Datasets:

jamescalam

/

ai-arxiv-chunked

like 32

1910.01108

0

DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter Victor SANH, Lysandre DEBUT, Julien CHAUMOND, Thomas WOLF Hugging Face {victor,lysandre,julien,thomas}@huggingface.co Abstract As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP), operating these large models in on-theedge and/or under constrained computational training or inference budgets remains challenging. In this work, we propose a method to pre-train a smaller generalpurpose language representation model, called DistilBERT, which can then be finetuned with good performances on a wide range of tasks like its larger counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage knowledge distillation during the pre-training phase and show that it is possible to reduce the size of a BERT model by 40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive biases learned by larger models during pre-training, we introduce a triple loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its

1910.01108

DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP), operating these large models in on-the-edge and/or under constrained computational training or inference budgets remains challenging. In this work, we propose a method to pre-train a smaller general-purpose language representation model, called DistilBERT, which can then be fine-tuned with good performances on a wide range of tasks like its larger counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage knowledge distillation during the pre-training phase and show that it is possible to reduce the size of a BERT model by 40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive biases learned by larger models during pre-training, we introduce a triple loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device study.

http://arxiv.org/pdf/1910.01108

February 2020 - Revision: fix bug in evaluation metrics, updated metrics, argumentation unchanged. 5 pages, 1 figure, 4 tables. Accepted at the 5th Workshop on Energy Efficient Machine Learning and Cognitive Computing - NeurIPS 2019

cs.CL

20191002

20200301

1910.01108

1

loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device study. 1 Introduction Figure 1: Parameter counts of several recently released pretrained language models.The last two years have seen the rise of Transfer Learning approaches in Natural Language Processing (NLP) with large-scale pre-trained language models becoming a basic tool in many NLP tasks [Devlin et al., 2018, Radford et al., 2019, Liu et al., 2019]. While these models lead to significant improvement, they often have several hundred million parameters and current research1on pre-trained models indicates that training even larger models still leads to better performances on downstream tasks. The trend toward bigger models raises several concerns. First is the environmental cost of exponentially scaling these models’ computational requirements as mentioned in Schwartz et al. [2019], Strubell et al. [2019]. Second, while operating these models on-device in real-time has the potential to enable novel and interesting language processing applications, the growing computational and memory requirements of these models may hamper wide adoption.

1910.01108

DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP), operating these large models in on-the-edge and/or under constrained computational training or inference budgets remains challenging. In this work, we propose a method to pre-train a smaller general-purpose language representation model, called DistilBERT, which can then be fine-tuned with good performances on a wide range of tasks like its larger counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage knowledge distillation during the pre-training phase and show that it is possible to reduce the size of a BERT model by 40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive biases learned by larger models during pre-training, we introduce a triple loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device study.

http://arxiv.org/pdf/1910.01108

February 2020 - Revision: fix bug in evaluation metrics, updated metrics, argumentation unchanged. 5 pages, 1 figure, 4 tables. Accepted at the 5th Workshop on Energy Efficient Machine Learning and Cognitive Computing - NeurIPS 2019

cs.CL

20191002

20200301

1910.01108

2

in real-time has the potential to enable novel and interesting language processing applications, the growing computational and memory requirements of these models may hamper wide adoption. 1See for instance the recently released MegatronLM ( https://nv-adlr.github.io/MegatronLM ) EMC^2: 5th Edition Co-located with NeurIPS’19arXiv:1910.01108v4 [cs.CL] 1 Mar 2020 In this paper, we show that it is possible to reach similar performances on many downstream-tasks using much smaller language models pre-trained with knowledge distillation, resulting in models that are lighter and faster at inference time, while also requiring a smaller computational training budget. Our general-purpose pre-trained models can be fine-tuned with good performances on several downstream tasks, keeping the flexibility of larger models. We also show that our compressed models are small enough to run on the edge, e.g. on mobile devices. Using a triple loss, we show that a 40% smaller Transformer (Vaswani et al. [2017]) pre-trained through distillation via the supervision of a bigger Transformer language model can achieve similar performance on a variety of downstream tasks, while being 60% faster at inference time. Further

1910.01108

DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP), operating these large models in on-the-edge and/or under constrained computational training or inference budgets remains challenging. In this work, we propose a method to pre-train a smaller general-purpose language representation model, called DistilBERT, which can then be fine-tuned with good performances on a wide range of tasks like its larger counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage knowledge distillation during the pre-training phase and show that it is possible to reduce the size of a BERT model by 40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive biases learned by larger models during pre-training, we introduce a triple loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device study.

http://arxiv.org/pdf/1910.01108

February 2020 - Revision: fix bug in evaluation metrics, updated metrics, argumentation unchanged. 5 pages, 1 figure, 4 tables. Accepted at the 5th Workshop on Energy Efficient Machine Learning and Cognitive Computing - NeurIPS 2019

cs.CL

20191002

20200301

1910.01108

3

through distillation via the supervision of a bigger Transformer language model can achieve similar performance on a variety of downstream tasks, while being 60% faster at inference time. Further ablation studies indicate that all the components of the triple loss are important for best performances. We have made the trained weights available along with the training code in the Transformers2 library from HuggingFace [Wolf et al., 2019]. 2 Knowledge distillation Knowledge distillation [Bucila et al., 2006, Hinton et al., 2015] is a compression technique in which a compact model - the student - is trained to reproduce the behaviour of a larger model - the teacher or an ensemble of models. In supervised learning, a classification model is generally trained to predict an instance class by maximizing the estimated probability of gold labels. A standard training objective thus involves minimizing the cross-entropy between the model’s predicted distribution and the one-hot empirical distribution of training labels. A model performing well on the training set will predict an output distribution with high probability on the correct class and with near-zero probabilities on other classes. But some of these "near-zero" probabilities are larger than others and reflect, in part, the generalization capabilities of the model and how well it will perform on the test set3.

1910.01108

DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP), operating these large models in on-the-edge and/or under constrained computational training or inference budgets remains challenging. In this work, we propose a method to pre-train a smaller general-purpose language representation model, called DistilBERT, which can then be fine-tuned with good performances on a wide range of tasks like its larger counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage knowledge distillation during the pre-training phase and show that it is possible to reduce the size of a BERT model by 40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive biases learned by larger models during pre-training, we introduce a triple loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device study.

http://arxiv.org/pdf/1910.01108

February 2020 - Revision: fix bug in evaluation metrics, updated metrics, argumentation unchanged. 5 pages, 1 figure, 4 tables. Accepted at the 5th Workshop on Energy Efficient Machine Learning and Cognitive Computing - NeurIPS 2019

cs.CL

20191002

20200301

1910.01108

4

generalization capabilities of the model and how well it will perform on the test set3. Training loss The student is trained with a distillation loss over the soft target probabilities of the teacher: Lce=P itilog(si)where ti(resp. si) is a probability estimated by the teacher (resp. the student). This objective results in a rich training signal by leveraging the full teacher distribution. Following Hinton et al. [2015] we used a softmax-temperature :pi=exp(zi=T)P jexp(zj=T) where Tcontrols the smoothness of the output distribution and ziis the model score for the class i. The same temperature Tis applied to the student and the teacher at training time, while at inference, Tis set to 1 to recover a standard softmax . The final training objective is a linear combination of the distillation loss Lcewith the supervised training loss, in our case the masked language modeling lossLmlm [Devlin et al., 2018]. We found it beneficial to add a cosine embedding loss (Lcos) which will tend to align the directions of the student and teacher hidden states vectors. 3 DistilBERT: a distilled version of BERT

1910.01108

DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP), operating these large models in on-the-edge and/or under constrained computational training or inference budgets remains challenging. In this work, we propose a method to pre-train a smaller general-purpose language representation model, called DistilBERT, which can then be fine-tuned with good performances on a wide range of tasks like its larger counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage knowledge distillation during the pre-training phase and show that it is possible to reduce the size of a BERT model by 40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive biases learned by larger models during pre-training, we introduce a triple loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device study.

http://arxiv.org/pdf/1910.01108

February 2020 - Revision: fix bug in evaluation metrics, updated metrics, argumentation unchanged. 5 pages, 1 figure, 4 tables. Accepted at the 5th Workshop on Energy Efficient Machine Learning and Cognitive Computing - NeurIPS 2019

cs.CL

20191002

20200301

1910.01108

5

and teacher hidden states vectors. 3 DistilBERT: a distilled version of BERT Student architecture In the present work, the student - DistilBERT - has the same general architecture as BERT. The token-type embeddings and the pooler are removed while the number of layers is reduced by a factor of 2. Most of the operations used in the Transformer architecture ( linear layer andlayer normalisation ) are highly optimized in modern linear algebra frameworks and our investigations showed that variations on the last dimension of the tensor (hidden size dimension) have a smaller impact on computation efficiency (for a fixed parameters budget) than variations on other factors like the number of layers. Thus we focus on reducing the number of layers. Student initialization In addition to the previously described optimization and architectural choices, an important element in our training procedure is to find the right initialization for the sub-network to converge. Taking advantage of the common dimensionality between teacher and student networks, we initialize the student from the teacher by taking one layer out of two. 2https://github.com/huggingface/transformers 3E.g. BERT-base’s predictions for a masked token in " I think this is the beginning of a

1910.01108

DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP), operating these large models in on-the-edge and/or under constrained computational training or inference budgets remains challenging. In this work, we propose a method to pre-train a smaller general-purpose language representation model, called DistilBERT, which can then be fine-tuned with good performances on a wide range of tasks like its larger counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage knowledge distillation during the pre-training phase and show that it is possible to reduce the size of a BERT model by 40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive biases learned by larger models during pre-training, we introduce a triple loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device study.

http://arxiv.org/pdf/1910.01108

February 2020 - Revision: fix bug in evaluation metrics, updated metrics, argumentation unchanged. 5 pages, 1 figure, 4 tables. Accepted at the 5th Workshop on Energy Efficient Machine Learning and Cognitive Computing - NeurIPS 2019

cs.CL

20191002

20200301

1910.01108

6

3E.g. BERT-base’s predictions for a masked token in " I think this is the beginning of a beautiful [MASK] " comprise two high probability tokens ( dayandlife) and a long tail of valid predictions (future ,story ,world . . . ). 2 Table 1: DistilBERT retains 97% of BERT performance. Comparison on the dev sets of the GLUE benchmark. ELMo results as reported by the authors. BERT and DistilBERT results are the medians of 5 runs with different seeds. Model Score CoLA MNLI MRPC QNLI QQP RTE SST-2 STS-B WNLI ELMo 68.7 44.1 68.6 76.6 71.1 86.2 53.4 91.5 70.4 56.3 BERT-base 79.5 56.3 86.7 88.6 91.8 89.6 69.3 92.7 89.0 53.5 DistilBERT 77.0 51.3 82.2 87.5 89.2 88.5 59.9 91.3 86.9 56.3 Table 2: DistilBERT yields to comparable performance on downstream tasks. Comparison on downstream tasks: IMDb (test accuracy) and SQuAD 1.1 (EM/F1 on dev set).

1910.01108

DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP), operating these large models in on-the-edge and/or under constrained computational training or inference budgets remains challenging. In this work, we propose a method to pre-train a smaller general-purpose language representation model, called DistilBERT, which can then be fine-tuned with good performances on a wide range of tasks like its larger counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage knowledge distillation during the pre-training phase and show that it is possible to reduce the size of a BERT model by 40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive biases learned by larger models during pre-training, we introduce a triple loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device study.

http://arxiv.org/pdf/1910.01108

February 2020 - Revision: fix bug in evaluation metrics, updated metrics, argumentation unchanged. 5 pages, 1 figure, 4 tables. Accepted at the 5th Workshop on Energy Efficient Machine Learning and Cognitive Computing - NeurIPS 2019

cs.CL

20191002

20200301

1910.01108

7

performance on downstream tasks. Comparison on downstream tasks: IMDb (test accuracy) and SQuAD 1.1 (EM/F1 on dev set). D: with a second step of distillation during fine-tuning. Model IMDb SQuAD (acc.) (EM/F1) BERT-base 93.46 81.2/88.5 DistilBERT 92.82 77.7/85.8 DistilBERT (D) - 79.1/86.9Table 3: DistilBERT is significantly smaller while being constantly faster. Inference time of a full pass of GLUE task STS-B (sentiment analysis) on CPU with a batch size of 1. Model # param. Inf. time (Millions) (seconds) ELMo 180 895 BERT-base 110 668 DistilBERT 66 410 Distillation We applied best practices for training BERT model recently proposed in Liu et al. [2019]. As such, DistilBERT is distilled on very large batches leveraging gradient accumulation (up to 4K examples per batch) using dynamic masking and without the next sentence prediction objective. Data and compute power We train DistilBERT on the same corpus as the original BERT model:

1910.01108

DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP), operating these large models in on-the-edge and/or under constrained computational training or inference budgets remains challenging. In this work, we propose a method to pre-train a smaller general-purpose language representation model, called DistilBERT, which can then be fine-tuned with good performances on a wide range of tasks like its larger counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage knowledge distillation during the pre-training phase and show that it is possible to reduce the size of a BERT model by 40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive biases learned by larger models during pre-training, we introduce a triple loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device study.

http://arxiv.org/pdf/1910.01108

February 2020 - Revision: fix bug in evaluation metrics, updated metrics, argumentation unchanged. 5 pages, 1 figure, 4 tables. Accepted at the 5th Workshop on Energy Efficient Machine Learning and Cognitive Computing - NeurIPS 2019

cs.CL

20191002

20200301

1910.01108

8

examples per batch) using dynamic masking and without the next sentence prediction objective. Data and compute power We train DistilBERT on the same corpus as the original BERT model: a concatenation of English Wikipedia and Toronto Book Corpus [Zhu et al., 2015]. DistilBERT was trained on 8 16GB V100 GPUs for approximately 90 hours. For the sake of comparison, the RoBERTa model [Liu et al., 2019] required 1 day of training on 1024 32GB V100. 4 Experiments General Language Understanding We assess the language understanding and generalization capabilities of DistilBERT on the General Language Understanding Evaluation (GLUE) benchmark [Wang et al., 2018], a collection of 9 datasets for evaluating natural language understanding systems. We report scores on the development sets for each task by fine-tuning DistilBERT without the use of ensembling or multi-tasking scheme for fine-tuning (which are mostly orthogonal to the present work). We compare the results to the baseline provided by the authors of GLUE: an ELMo (Peters et al. [2018]) encoder followed by two BiLSTMs.4 The results on each of the 9 tasks are showed on Table 1 along with the macro-score (average of

1910.01108

DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP), operating these large models in on-the-edge and/or under constrained computational training or inference budgets remains challenging. In this work, we propose a method to pre-train a smaller general-purpose language representation model, called DistilBERT, which can then be fine-tuned with good performances on a wide range of tasks like its larger counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage knowledge distillation during the pre-training phase and show that it is possible to reduce the size of a BERT model by 40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive biases learned by larger models during pre-training, we introduce a triple loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device study.

http://arxiv.org/pdf/1910.01108

February 2020 - Revision: fix bug in evaluation metrics, updated metrics, argumentation unchanged. 5 pages, 1 figure, 4 tables. Accepted at the 5th Workshop on Energy Efficient Machine Learning and Cognitive Computing - NeurIPS 2019

cs.CL

20191002

20200301

1910.01108

9

et al. [2018]) encoder followed by two BiLSTMs.4 The results on each of the 9 tasks are showed on Table 1 along with the macro-score (average of individual scores). Among the 9 tasks, DistilBERT is always on par or improving over the ELMo baseline (up to 19 points of accuracy on STS-B). DistilBERT also compares surprisingly well to BERT, retaining 97% of the performance with 40% fewer parameters. 4.1 Downstream task benchmark Downstream tasks We further study the performances of DistilBERT on several downstream tasks under efficient inference constraints: a classification task (IMDb sentiment classification - Maas et al. [2011]) and a question answering task (SQuAD v1.1 - Rajpurkar et al. [2016]). As shown in Table 2, DistilBERT is only 0.6% point behind BERT in test accuracy on the IMDb benchmark while being 40% smaller. On SQuAD, DistilBERT is within 3.9 points of the full BERT. We also studied whether we could add another step of distillation during the adaptation phase by

1910.01108

DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP), operating these large models in on-the-edge and/or under constrained computational training or inference budgets remains challenging. In this work, we propose a method to pre-train a smaller general-purpose language representation model, called DistilBERT, which can then be fine-tuned with good performances on a wide range of tasks like its larger counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage knowledge distillation during the pre-training phase and show that it is possible to reduce the size of a BERT model by 40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive biases learned by larger models during pre-training, we introduce a triple loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device study.

http://arxiv.org/pdf/1910.01108

February 2020 - Revision: fix bug in evaluation metrics, updated metrics, argumentation unchanged. 5 pages, 1 figure, 4 tables. Accepted at the 5th Workshop on Energy Efficient Machine Learning and Cognitive Computing - NeurIPS 2019

cs.CL

20191002

20200301

1910.01108

10

We also studied whether we could add another step of distillation during the adaptation phase by fine-tuning DistilBERT on SQuAD using a BERT model previously fine-tuned on SQuAD as a 4We use jiant [Wang et al., 2019] to compute the baseline. 3 Table 4: Ablation study. Variations are relative to the model trained with triple loss and teacher weights initialization. Ablation Variation on GLUE macro-score ;-Lcos-Lmlm -2.96 Lce-;-Lmlm -1.46 Lce-Lcos-; -0.31 Triple loss + random weights initialization -3.69 teacher for an additional term in the loss (knowledge distillation). In this setting, there are thus two successive steps of distillation, one during the pre-training phase and one during the adaptation phase. In this case, we were able to reach interesting performances given the size of the model: 79.8 F1 and 70.4 EM, i.e. within 3 points of the full model. Size and inference speed To further investigate the speed-up/size trade-off of DistilBERT, we compare (in Table 3) the number

1910.01108

DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP), operating these large models in on-the-edge and/or under constrained computational training or inference budgets remains challenging. In this work, we propose a method to pre-train a smaller general-purpose language representation model, called DistilBERT, which can then be fine-tuned with good performances on a wide range of tasks like its larger counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage knowledge distillation during the pre-training phase and show that it is possible to reduce the size of a BERT model by 40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive biases learned by larger models during pre-training, we introduce a triple loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device study.

http://arxiv.org/pdf/1910.01108

February 2020 - Revision: fix bug in evaluation metrics, updated metrics, argumentation unchanged. 5 pages, 1 figure, 4 tables. Accepted at the 5th Workshop on Energy Efficient Machine Learning and Cognitive Computing - NeurIPS 2019

cs.CL

20191002

20200301

1910.01108

11

Size and inference speed To further investigate the speed-up/size trade-off of DistilBERT, we compare (in Table 3) the number of parameters of each model along with the inference time needed to do a full pass on the STSB development set on CPU (Intel Xeon E5-2690 v3 Haswell @2.9GHz) using a batch size of 1. DistilBERT has 40% fewer parameters than BERT and is 60% faster than BERT. On device computation We studied whether DistilBERT could be used for on-the-edge applications by building a mobile application for question answering. We compare the average inference time on a recent smartphone (iPhone 7 Plus) against our previously trained question answering model based on BERT-base. Excluding the tokenization step, DistilBERT is 71% faster than BERT, and the whole model weighs 207 MB (which could be further reduced with quantization). Our code is available5. 4.2 Ablation study In this section, we investigate the influence of various components of the triple loss and the student initialization on the performances of the distilled model. We report the macro-score on GLUE. Table 4 presents the deltas with the full triple loss: removing the Masked Language Modeling loss has little

1910.01108

DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP), operating these large models in on-the-edge and/or under constrained computational training or inference budgets remains challenging. In this work, we propose a method to pre-train a smaller general-purpose language representation model, called DistilBERT, which can then be fine-tuned with good performances on a wide range of tasks like its larger counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage knowledge distillation during the pre-training phase and show that it is possible to reduce the size of a BERT model by 40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive biases learned by larger models during pre-training, we introduce a triple loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device study.

http://arxiv.org/pdf/1910.01108

February 2020 - Revision: fix bug in evaluation metrics, updated metrics, argumentation unchanged. 5 pages, 1 figure, 4 tables. Accepted at the 5th Workshop on Energy Efficient Machine Learning and Cognitive Computing - NeurIPS 2019

cs.CL

20191002

20200301

1910.01108

12

presents the deltas with the full triple loss: removing the Masked Language Modeling loss has little impact while the two distillation losses account for a large portion of the performance. 5 Related work Task-specific distillation Most of the prior works focus on building task-specific distillation setups. Tang et al. [2019] transfer fine-tune classification model BERT to an LSTM-based classifier. Chatterjee [2019] distill BERT model fine-tuned on SQuAD in a smaller Transformer model previously initialized from BERT. In the present work, we found it beneficial to use a general-purpose pre-training distillation rather than a task-specific distillation. Turc et al. [2019] use the original pretraining objective to train smaller student, then fine-tuned via distillation. As shown in the ablation study, we found it beneficial to leverage the teacher’s knowledge to pre-train with additional distillation signal. Multi-distillation Yang et al. [2019] combine the knowledge of an ensemble of teachers using multi-task learning to regularize the distillation. The authors apply Multi-Task Knowledge Distillation

1910.01108

DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP), operating these large models in on-the-edge and/or under constrained computational training or inference budgets remains challenging. In this work, we propose a method to pre-train a smaller general-purpose language representation model, called DistilBERT, which can then be fine-tuned with good performances on a wide range of tasks like its larger counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage knowledge distillation during the pre-training phase and show that it is possible to reduce the size of a BERT model by 40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive biases learned by larger models during pre-training, we introduce a triple loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device study.

http://arxiv.org/pdf/1910.01108

February 2020 - Revision: fix bug in evaluation metrics, updated metrics, argumentation unchanged. 5 pages, 1 figure, 4 tables. Accepted at the 5th Workshop on Energy Efficient Machine Learning and Cognitive Computing - NeurIPS 2019

cs.CL

20191002

20200301

1910.01108

13

multi-task learning to regularize the distillation. The authors apply Multi-Task Knowledge Distillation to learn a compact question answering model from a set of large question answering models. An application of multi-distillation is multi-linguality: Tsai et al. [2019] adopts a similar approach to us by pre-training a multilingual model from scratch solely through distillation. However, as shown in the ablation study, leveraging the teacher’s knowledge with initialization and additional losses leads to substantial gains. Other compression techniques have been studied to compress large models. Recent developments in weights pruning reveal that it is possible to remove some heads in the self-attention at test time without significantly degrading the performance Michel et al. [2019]. Some layers can be reduced to one head. A separate line of study leverages quantization to derive smaller models (Gupta et al. [2015]). Pruning and quantization are orthogonal to the present work. 5https://github.com/huggingface/swift-coreml-transformers 4 6 Conclusion and future work We introduced DistilBERT, a general-purpose pre-trained version of BERT, 40% smaller, 60% faster, that retains 97% of the language understanding capabilities. We showed that a general-purpose

1910.01108

DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP), operating these large models in on-the-edge and/or under constrained computational training or inference budgets remains challenging. In this work, we propose a method to pre-train a smaller general-purpose language representation model, called DistilBERT, which can then be fine-tuned with good performances on a wide range of tasks like its larger counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage knowledge distillation during the pre-training phase and show that it is possible to reduce the size of a BERT model by 40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive biases learned by larger models during pre-training, we introduce a triple loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device study.

http://arxiv.org/pdf/1910.01108

February 2020 - Revision: fix bug in evaluation metrics, updated metrics, argumentation unchanged. 5 pages, 1 figure, 4 tables. Accepted at the 5th Workshop on Energy Efficient Machine Learning and Cognitive Computing - NeurIPS 2019

cs.CL

20191002

20200301

1910.01108

14

that retains 97% of the language understanding capabilities. We showed that a general-purpose language model can be successfully trained with distillation and analyzed the various components with an ablation study. We further demonstrated that DistilBERT is a compelling option for edge applications. References Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. In NAACL-HLT , 2018. Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. Language models are unsupervised multitask learners. 2019. Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar S. Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke S. Zettlemoyer, and Veselin Stoyanov. Roberta: A robustly optimized bert pretraining approach. ArXiv , abs/1907.11692, 2019. Roy Schwartz, Jesse Dodge, Noah A. Smith, and Oren Etzioni. Green ai. ArXiv , abs/1907.10597, 2019. Emma Strubell, Ananya Ganesh, and Andrew McCallum. Energy and policy considerations for deep learning in

1910.01108

DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP), operating these large models in on-the-edge and/or under constrained computational training or inference budgets remains challenging. In this work, we propose a method to pre-train a smaller general-purpose language representation model, called DistilBERT, which can then be fine-tuned with good performances on a wide range of tasks like its larger counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage knowledge distillation during the pre-training phase and show that it is possible to reduce the size of a BERT model by 40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive biases learned by larger models during pre-training, we introduce a triple loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device study.

http://arxiv.org/pdf/1910.01108

February 2020 - Revision: fix bug in evaluation metrics, updated metrics, argumentation unchanged. 5 pages, 1 figure, 4 tables. Accepted at the 5th Workshop on Energy Efficient Machine Learning and Cognitive Computing - NeurIPS 2019

cs.CL

20191002

20200301

1910.01108

15

Emma Strubell, Ananya Ganesh, and Andrew McCallum. Energy and policy considerations for deep learning in nlp. In ACL, 2019. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. In NIPS , 2017. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Rémi Louf, Morgan Funtowicz, and Jamie Brew. Transformers: State-of-the-art natural language processing, 2019. Cristian Bucila, Rich Caruana, and Alexandru Niculescu-Mizil. Model compression. In KDD , 2006. Geoffrey E. Hinton, Oriol Vinyals, and Jeffrey Dean. Distilling the knowledge in a neural network. ArXiv , abs/1503.02531, 2015. Yukun Zhu, Ryan Kiros, Richard S. Zemel, Ruslan Salakhutdinov, Raquel Urtasun, Antonio Torralba, and Sanja

1910.01108

DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP), operating these large models in on-the-edge and/or under constrained computational training or inference budgets remains challenging. In this work, we propose a method to pre-train a smaller general-purpose language representation model, called DistilBERT, which can then be fine-tuned with good performances on a wide range of tasks like its larger counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage knowledge distillation during the pre-training phase and show that it is possible to reduce the size of a BERT model by 40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive biases learned by larger models during pre-training, we introduce a triple loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device study.

http://arxiv.org/pdf/1910.01108

February 2020 - Revision: fix bug in evaluation metrics, updated metrics, argumentation unchanged. 5 pages, 1 figure, 4 tables. Accepted at the 5th Workshop on Energy Efficient Machine Learning and Cognitive Computing - NeurIPS 2019

cs.CL

20191002

20200301

1910.01108

16

Yukun Zhu, Ryan Kiros, Richard S. Zemel, Ruslan Salakhutdinov, Raquel Urtasun, Antonio Torralba, and Sanja Fidler. Aligning books and movies: Towards story-like visual explanations by watching movies and reading books. 2015 IEEE International Conference on Computer Vision (ICCV) , pages 19–27, 2015. Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel R. Bowman. Glue: A multi-task benchmark and analysis platform for natural language understanding. In ICLR , 2018. Matthew E. Peters, Mark Neumann, Mohit Iyyer, Matt Gardner, Christopher Clark, Kenton Lee, and Luke Zettlemoyer. Deep contextualized word representations. In NAACL , 2018. Alex Wang, Ian F. Tenney, Yada Pruksachatkun, Katherin Yu, Jan Hula, Patrick Xia, Raghu Pappagari, Shuning Jin, R. Thomas McCoy, Roma Patel, Yinghui Huang, Jason Phang, Edouard Grave, Najoung Kim, Phu Mon Htut, Thibault F’evry, Berlin Chen, Nikita Nangia, Haokun Liu, Anhad Mohananey, Shikha Bordia, Nicolas

1910.01108

DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP), operating these large models in on-the-edge and/or under constrained computational training or inference budgets remains challenging. In this work, we propose a method to pre-train a smaller general-purpose language representation model, called DistilBERT, which can then be fine-tuned with good performances on a wide range of tasks like its larger counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage knowledge distillation during the pre-training phase and show that it is possible to reduce the size of a BERT model by 40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive biases learned by larger models during pre-training, we introduce a triple loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device study.

http://arxiv.org/pdf/1910.01108

February 2020 - Revision: fix bug in evaluation metrics, updated metrics, argumentation unchanged. 5 pages, 1 figure, 4 tables. Accepted at the 5th Workshop on Energy Efficient Machine Learning and Cognitive Computing - NeurIPS 2019

cs.CL

20191002

20200301

1910.01108

17

Htut, Thibault F’evry, Berlin Chen, Nikita Nangia, Haokun Liu, Anhad Mohananey, Shikha Bordia, Nicolas Patry, Ellie Pavlick, and Samuel R. Bowman. jiant 1.1: A software toolkit for research on general-purpose text understanding models. http://jiant.info/ , 2019. Andrew L. Maas, Raymond E. Daly, Peter T. Pham, Dan Huang, Andrew Y . Ng, and Christopher Potts. Learning word vectors for sentiment analysis. In ACL, 2011. Pranav Rajpurkar, Jian Zhang, Konstantin Lopyrev, and Percy Liang. Squad: 100, 000+ questions for machine comprehension of text. In EMNLP , 2016. Raphael Tang, Yao Lu, Linqing Liu, Lili Mou, Olga Vechtomova, and Jimmy Lin. Distilling task-specific knowledge from bert into simple neural networks. ArXiv , abs/1903.12136, 2019. Debajyoti Chatterjee. Making neural machine reading comprehension faster. ArXiv , abs/1904.00796, 2019. Iulia Turc, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Well-read students learn better: The impact

1910.01108

DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP), operating these large models in on-the-edge and/or under constrained computational training or inference budgets remains challenging. In this work, we propose a method to pre-train a smaller general-purpose language representation model, called DistilBERT, which can then be fine-tuned with good performances on a wide range of tasks like its larger counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage knowledge distillation during the pre-training phase and show that it is possible to reduce the size of a BERT model by 40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive biases learned by larger models during pre-training, we introduce a triple loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device study.

http://arxiv.org/pdf/1910.01108

February 2020 - Revision: fix bug in evaluation metrics, updated metrics, argumentation unchanged. 5 pages, 1 figure, 4 tables. Accepted at the 5th Workshop on Energy Efficient Machine Learning and Cognitive Computing - NeurIPS 2019

cs.CL

20191002

20200301

1910.01108

18

Iulia Turc, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Well-read students learn better: The impact of student initialization on knowledge distillation. ArXiv , abs/1908.08962, 2019. Ze Yang, Linjun Shou, Ming Gong, Wutao Lin, and Daxin Jiang. Model compression with multi-task knowledge distillation for web-scale question answering system. ArXiv , abs/1904.09636, 2019. Henry Tsai, Jason Riesa, Melvin Johnson, Naveen Arivazhagan, Xin Li, and Amelia Archer. Small and practical bert models for sequence labeling. In EMNLP-IJCNLP , 2019. Paul Michel, Omer Levy, and Graham Neubig. Are sixteen heads really better than one? In NeurIPS , 2019. Suyog Gupta, Ankur Agrawal, Kailash Gopalakrishnan, and Pritish Narayanan. Deep learning with limited numerical precision. In ICML , 2015. 5

1910.01108

DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP), operating these large models in on-the-edge and/or under constrained computational training or inference budgets remains challenging. In this work, we propose a method to pre-train a smaller general-purpose language representation model, called DistilBERT, which can then be fine-tuned with good performances on a wide range of tasks like its larger counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage knowledge distillation during the pre-training phase and show that it is possible to reduce the size of a BERT model by 40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive biases learned by larger models during pre-training, we introduce a triple loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device study.

http://arxiv.org/pdf/1910.01108

February 2020 - Revision: fix bug in evaluation metrics, updated metrics, argumentation unchanged. 5 pages, 1 figure, 4 tables. Accepted at the 5th Workshop on Energy Efficient Machine Learning and Cognitive Computing - NeurIPS 2019

cs.CL

20191002

20200301

1710.06481

0

Constructing Datasets for Multi-hop Reading Comprehension Across Documents Johannes Welbl1Pontus Stenetorp1Sebastian Riedel1;2 1University College London,2Bloomsbury AI fj.welbl,p.stenetorp,s.riedel [email protected] Abstract Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently no resources exist to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence – effectively performing multihop, alias multi-step, inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced,1and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

1

thematically linked documents. Two datasets from different domains are induced,1and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information; and providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 54.5% on an annotated test set, compared to human performance at 85.0%, leaving ample room for improvement. 1 Introduction Devising computer systems capable of answering questions about knowledge described using text has 1Available at http://qangaroo.cs.ucl.ac.uk The Hanging Gardens, in [Mumbai], also known as Pherozeshah Mehta Gardens, are terraced gardens … They provide sunset views over the [Arabian Sea] … Mumbai (also known as Bombay, the official name until 1995) is the capital city of the Indian state of Maharashtra. It is the most populous city in India …Q: (Hanging gardens of Mumbai, country, ?) Options: {Iran, India, Pakistan, Somalia, …}

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

2

The Arabian Sea is a region of the northern Indian Ocean bounded on the north by Pakistan and Iran, on the west by northeastern Somalia and the Arabian Peninsula, and on the east by India …Figure 1: A sample from the W IKIHOPdataset where it is necessary to combine information spread across multiple documents to infer the correct answer. been a longstanding challenge in Natural Language Processing (NLP). Contemporary end-to-end Reading Comprehension (RC) methods can learn to extract the correct answer span within a given text and approach human-level performance (Kadlec et al., 2016; Seo et al., 2017a). However, for existing datasets, relevant information is often concentrated locally within a single sentence, emphasizing the role of locating, matching, and aligning information between query and support text. For example, Weissenborn et al. (2017) observed that a simple binary word-in-query indicator feature boosted the relative accuracy of a baseline model by 27.9%. We argue that, in order to further the ability of machine comprehension methods to extract knowledge from text, we must move beyond a scenario where relevant information is coherently and explicitly stated within a single document. Methods with this

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

3

from text, we must move beyond a scenario where relevant information is coherently and explicitly stated within a single document. Methods with this capability would aid Information Extraction (IE) applications, such as discovering drug-drug interac-arXiv:1710.06481v2 [cs.CL] 11 Jun 2018 tions (Gurulingappa et al., 2012) by connecting protein interactions reported across different publications. They would also benefit search (Carpineto and Romano, 2012) and Question Answering (QA) applications (Lin and Pantel, 2001) where the required information cannot be found in a single location. Figure 1 shows an example from W IKIPEDIA , where the goal is to identify the country property of the Hanging Gardens of Mumbai . This cannot be inferred solely from the article about them without additional background knowledge, as the answer is not stated explicitly. However, several of the linked articles mention the correct answer India (and other countries), but cover different topics (e.g. Mumbai , Arabian Sea , etc.). Finding the answer requires multi-hop reasoning: figuring out that the Hanging Gardens are located in Mumbai , and then, from a second document, that Mumbai is a city in India .

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

4

Gardens are located in Mumbai , and then, from a second document, that Mumbai is a city in India . We define a novel RC task in which a model should learn to answer queries by combining evidence stated across documents. We introduce a methodology to induce datasets for this task and derive two datasets. The first, W IKIHOP, uses sets of WIKIPEDIA articles where answers to queries about specific properties of an entity cannot be located in the entity’s article. In the second dataset, M EDHOP, the goal is to establish drug-drug interactions based on scientific findings about drugs and proteins and their interactions, found across multiple M EDLINE abstracts. For both datasets we draw upon existing Knowledge Bases (KBs), W IKIDATA and D RUGBANK, as ground truth, utilizing distant supervision (Mintz et al., 2009) to induce the data – similar to Hewlett et al. (2016) and Joshi et al. (2017). We establish that for 74.1% and 68.0% of the samples, the answer can be inferred from the given documents by a human annotator. Still, constructing multi-document datasets is challenging; we encounter and prescribe remedies for several pitfalls

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

5

samples, the answer can be inferred from the given documents by a human annotator. Still, constructing multi-document datasets is challenging; we encounter and prescribe remedies for several pitfalls associated with their assembly – for example, spurious co-locations of answers and specific documents. For both datasets we then establish several strong baselines and evaluate the performance of two previously proposed competitive RC models (Seo et al., 2017a; Weissenborn et al., 2017). We find that one can integrate information across documents, but neither excels at selecting relevant information from a larger documents set, as their accuracy increases sig-nificantly when given only documents guaranteed to be relevant. The best model reaches 54.5% on an annotated test set, compared to human performance at 85.0%, indicating ample room for improvement. In summary, our key contributions are as follows: Firstly, proposing a cross-document multi-step RC task, as well as a general dataset induction strategy. Secondly, assembling two datasets from different domains and identifying dataset construction pitfalls and remedies. Thirdly, establishing multiple baselines, including two recently proposed RC models, as well as analysing model behaviour in detail through ablation studies. 2 Task and Dataset Construction Method We will now formally define the multi-hop RC task,

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

6

through ablation studies. 2 Task and Dataset Construction Method We will now formally define the multi-hop RC task, and a generic methodology to construct multi-hop RC datasets. Later, in Sections 3 and 4 we will demonstrate how this method is applied in practice by creating datasets for two different domains. Task Formalization A model is given a query q, a set of supporting documents Sq, and a set of candidate answers Cq– all of which are mentioned in Sq. The goal is to identify the correct answer a2Cq by drawing on the support documents Sq. Queries could potentially have several true answers when not constrained to rely on a specific set of support documents – e.g., queries about the parent of a certain individual. However, in our setup each sample has only one true answer among CqandSq. Note that even though we will utilize background information during dataset assembly, such information will not be available to a model: the document set will be provided in random order and without any metadata. While certainly beneficial, this would distract from our goal of fostering end-to-end RC methods that infer facts by combining separate facts stated in text. Dataset Assembly We assume that there exists a document corpus D, together with a KB containing

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

7

Dataset Assembly We assume that there exists a document corpus D, together with a KB containing fact triples (s; r; o )– with subject entity s, relation r, and object entity o. For example, one such fact could be (Hanging Gardens ofMumbai, country, India) . We start with individual KB facts and transform them into query-answer pairs by leaving the object slot empty, i.e. q= (s; r;?)anda=o. Next, we define a directed bipartite graph, where vertices on one side correspond to documents in D, and vertices on the other side are entities from the KB – see Figure 2 for an example. A document node dis connected to an entity eifeis mentioned in d, though there may be further constraints when defining the graph connectivity. For a given (q; a)pair, the candidates Cqand support documents SqDare identified by traversing the bipartite graph using breadth-first search; the documents visited will become the support documents Sq. As the traversal starting point, we use the node belonging to the subject entity sof the query q. As traversal end points, we use the set of all entity nodes

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

8

belonging to the subject entity sof the query q. As traversal end points, we use the set of all entity nodes that are type-consistent answers to q.2Note that whenever there is another fact (s; r; o0)in the KB, i.e. a fact producing the same qbut with a different a, we will not include o0into the set of end points for this sample. This ensures that precisely one of the end points corresponds to a correct answer to q.3 When traversing the graph starting at s, several end points will be visited, though generally not all; those visited define the candidate set Cq. If however the correct answer ais not among them we discard the(q; a)pair. The documents visited to reach the end points will define the support document set Sq. That is, Sqcomprises chains of documents leading not only from the query subject to the correct answer, but also to type-consistent false candidates. With this methodology, relevant textual evidence for(q; a)will be spread across documents along the chain connecting sanda– ensuring that multihop reasoning goes beyond resolving co-reference within a single document. Note that including other type-consistent candidates alongside aas end

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

9

the chain connecting sanda– ensuring that multihop reasoning goes beyond resolving co-reference within a single document. Note that including other type-consistent candidates alongside aas end points in the graph traversal – and thus into the support documents – renders the task considerably more challenging (Jia and Liang, 2017). Models could otherwise identify ain the documents by simply relying on type-consistency heuristics. It is worth pointing out that by introducing alternative candidates we counterbalance a type-consistency bias, in contrast to Hermann et al. (2015) and Hill et al. (2016) who instead rely on entity masking. 2To determine entities which are type-consistent for a query q, we consider all entities which are observed as object in a fact with ras relation type – including the correct answer. 3Here we rely on a closed-world assumption; that is, we assume that the facts in the KB state all true facts. Documents EntitiesKB (s, r, o)(s, r, o0)(s0,r ,o00)soo0o00 Figure 2: A bipartite graph connecting entities and documents mentioning them. Bold edges are those traversed

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

10

Figure 2: A bipartite graph connecting entities and documents mentioning them. Bold edges are those traversed for the first fact in the small KB on the right; yellow highlighting indicates documents in Sqand candidates in Cq. Check and cross indicate correct and false candidates. 3 W IKIHOP WIKIPEDIA contains an abundance of humancurated, multi-domain information and has several structured resources such as infoboxes and WIKIDATA (Vrande ˇci´c, 2012) associated with it. WIKIPEDIA has thus been used for a wealth of research to build datasets posing queries about a single sentence (Morales et al., 2016; Levy et al., 2017) or article (Yang et al., 2015; Hewlett et al., 2016; Rajpurkar et al., 2016). However, no attempt has been made to construct a cross-document multi-step RC dataset based on W IKIPEDIA . A recently proposed RC dataset is W IKIREADING (Hewlett et al., 2016), where W IKIDATA tuples (item, property, answer) are aligned with the W IKIPEDIA articles regarding their item . The tuples define a slot filling task with the goal of predicting the answer , given an article andproperty .

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

11

tuples define a slot filling task with the goal of predicting the answer , given an article andproperty . One problem with using W IKIREADING as an extractive RC dataset is that 54.4% of the samples do not state the answer explicitly in the given article (Hewlett et al., 2016). However, we observed that some of the articles accessible by following hyperlinks from the given article often state the answer, alongside other plausible candidates. 3.1 Assembly We now apply the methodology from Section 2 to create a multi-hop dataset with W IKIPEDIA as the document corpus and W IKIDATA as structured knowledge triples. In this setup, (item, property, answer) WIKIDATA tuples correspond to (s; r; o ) triples, and the item andproperty of each sample together form our query q– e.g., (Hanging Gardens of Mumbai, country, ?) . Similar to Yang et al. (2015) we only use the first paragraph of an article, as relevant information is more often stated in the beginning. Starting with all samples in W IKIREADING , we first remove samples where the answer is stated explicitly in the W IKIPEDIA article about the item .4

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

12

we first remove samples where the answer is stated explicitly in the W IKIPEDIA article about the item .4 The bipartite graph is structured as follows: (1) for edges from articles to entities: all articles mentioning an entity eare connected to e; (2) for edges from entities to articles: each entity eis only connected to the W IKIPEDIA article about the entity. Traversing the graph is then equivalent to iteratively following hyperlinks to new articles about the anchor text entities. For a given query-answer pair, the item entity is chosen as the starting point for the graph traversal. A traversal will always pass through the article about the item , since this is the only document connected from there. The end point set includes the correct answer alongside other type-consistent candidate expressions, which are determined by considering allfacts belonging to W IKIREADING training samples, selecting those triples with the same property as inqand keeping their answer expressions. As an example, for the W IKIDATA property country , this would be the set fFrance ;Russia ; :::g. We executed graph traversal up to a maximum chain length of 3 documents. To not pose unreasonable computational constraints, samples with more than 64 different support documents or 100 candidates

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

13

We executed graph traversal up to a maximum chain length of 3 documents. To not pose unreasonable computational constraints, samples with more than 64 different support documents or 100 candidates are removed, discarding 1% of the samples. 3.2 Mitigating Dataset Biases Dataset creation is always fraught with the risk of inducing unintended errors and biases (Chen et al., 2016; Schwartz et al., 2017). As Hewlett et al. (2016) only carried out limited analysis of their WIKIREADING dataset, we present an analysis of the downstream effects we observe on W IKIHOP. Candidate Frequency Imbalance A first observation is that there is a significant bias in the answer distribution of W IKIREADING . For example, in the majority of the samples the property country has theUnited States of America as the answer. A simple 4We thus use a disjoint subset of W IKIREADING compared to Levy et al. (2017) to construct W IKIHOP.majority class baseline would thus prove successful, but would tell us little about multi-hop reasoning. To combat this issue, we subsampled the dataset to ensure that samples of any one particular answer candidate make up no more than 0:1%of the dataset,

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

14

combat this issue, we subsampled the dataset to ensure that samples of any one particular answer candidate make up no more than 0:1%of the dataset, and omitted articles about the United States . Document-Answer Correlations A problem unique to our multi-document setting is the possibility of spurious correlations between candidates and documents induced by the graph traversal method. In fact, if we were notto address this issue, a model designed to exploit these regularities could achieve 74.6% accuracy (detailed in Section 6). Concretely, we observed that certain documents frequently co-occur with the correct answer, independently of the query. For example, if the article about London is present in Sq, the answer is likely to be the United Kingdom , independent of the query type or entity in question. Appendix C contains a list with several additional examples. We designed a statistic to measure this effect and then used it to sub-sample the dataset. The statistic counts how often a candidate cis observed as the correct answer when a certain document is present in Sqacross training set samples. More formally, for a given document dand answer candidatec, letcooccurrence (d; c)denote the total count of how often dco-occurs with cin a sample where cis also the correct answer. We use this statistic

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

15

of how often dco-occurs with cin a sample where cis also the correct answer. We use this statistic to filter the dataset, by discarding samples with at least one document-candidate pair (d; c)for which cooccurrence (d; c)>20. 4 M EDHOP Following the same general methodology, we next construct a second dataset for the domain of molecular biology – a field that has been undergoing exponential growth in the number of publications (Cohen and Hunter, 2004). The promise of applying NLP methods to cope with this increase has led to research efforts in IE (Hirschman et al., 2005; Kim et al., 2011) and QA for biomedical text (Hersh et al., 2007; Nentidis et al., 2017). There are a plethora of manually curated structured resources (Ashburner et al., 2000; The UniProt Consortium, 2017) which can either serve as ground truth or to induce training data using distant supervision (Craven and Kumlien, 1999; Bobic et al., 2012). Existing RC datasets are either severely limited in size (Hersh et al., 2007) or cover a very diverse set of query types (Nentidis

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

16

either severely limited in size (Hersh et al., 2007) or cover a very diverse set of query types (Nentidis et al., 2017), complicating the application of neural models that have seen successes for other domains (Wiese et al., 2017). A task that has received significant attention is detecting Drug-Drug Interactions (DDIs). Existing DDI efforts have focused on explicit mentions of interactions in single sentences (Gurulingappa et al., 2012; Percha et al., 2012; Segura-Bedmar et al., 2013). However, as shown by Peng et al. (2017), cross-sentence relation extraction increases the number of available relations. It is thus likely that cross-document interactions would further improve recall, which is of particular importance considering interactions that are never stated explicitly – but rather need to be inferred from separate pieces of evidence. The promise of multi-hop methods is finding and combining individual observations that can suggest previously unobserved DDIs, aiding the process of making scientific discoveries, yet not directly from experiments, but by inferring them from established public knowledge (Swanson, 1986). DDIs are caused by Protein-Protein Interaction (PPI) chains, forming biomedical pathways. If we consider PPI chains across documents,

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

17

DDIs are caused by Protein-Protein Interaction (PPI) chains, forming biomedical pathways. If we consider PPI chains across documents, we find examples like in Figure 3. Here the first document states that the drug Leuprolide causes GnRH receptor -induced synaptic potentiations, which can be blocked by the protein Progonadoliberin-1 . The last document states that another drug, Triptorelin , is a superagonist of the same protein. It is therefore likely to affect the potency of Leuprolide , describing a way in which the two drugs interact. Besides the true interaction there is also a false candidate Urofollitropin for which, although mentioned together with GnRH receptor within one document, there is no textual evidence indicating interactions with Leuprolide . 4.1 Assembly We construct M EDHOPusing D RUGBANK (Law et al., 2014) as structured knowledge resource and research paper abstracts from M EDLINE as documents. There is only one relation type for D RUGBANK facts, interacts with , that connects pairs of drugs – an example of a M EDHOPquery would thus Q: (Leuprolide, interacts_with, ?) Options: {Triptorelin, Urofollitropin}

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

18

Q: (Leuprolide, interacts_with, ?) Options: {Triptorelin, Urofollitropin} Leuprolide ... elicited a long-lasting potentiation of excitatory postsynaptic currents… [GnRH receptor]-induced synaptic potentiation was blocked … by [Progonadoliberin-1], a specific [GnRH receptor] antagonist… Analyses of gene expression demonstrated a dynamic response to the Progonadoliberin-1 superagonist Triptorelin. … our research to study the distribution, co-localization of Urofollitropin and its receptor[,] and co-localization of Urofollitropin and GnRH receptor…Figure 3: A sample from the M EDHOPdataset. be(Leuprolide, interacts with, ?) . We start by processing the 2016 M EDLINE release using the preprocessing pipeline employed for the BioNLP 2011 Shared Task (Stenetorp et al., 2011). We restrict the set of entities in the bipartite graph to drugs in D RUGBANK and human proteins in S WISS PROT (Bairoch et al., 2004). That is, the graph has drugs and proteins on one side, and M EDLINE abstracts on the other.

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

19

drugs and proteins on one side, and M EDLINE abstracts on the other. The edge structure is as follows: (1) There is an edge from a document to all proteins mentioned in it. (2) There is an edge between a document and a drug, if this document also mentions a protein known to be a target for the drug according to D RUGBANK. This edge is bidirectional, i.e. it can be traversed both ways, since there is no canonical document describing each drug – thus one can “hop” to any document mentioning the drug and its target. (3) There is an edge from a protein pto a document mentioning p, but only if the document also mentions another proteinp0which is known to interact with paccording to REACTOME (Fabregat et al., 2016). Given our distant supervision assumption, these additionally constraining requirements err on the side of precision. As a mention, similar to Percha et al. (2012), we consider any exact match of a name variant of a drug or human protein in D RUGBANK or S WISS PROT. For a given DDI (drug 1, interacts with, drug 2), we then select drug 1as the starting point for the graph traversal. As possible end points, we consider any other drug, apart from drug 1and those

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

20

drug 2), we then select drug 1as the starting point for the graph traversal. As possible end points, we consider any other drug, apart from drug 1and those interacting with drug 1other than drug 2. Similar to WIKIHOP, we exclude samples with more than 64 support documents and impose a maximum document length of 300 tokens plus title. Document Sub-sampling The bipartite graph for MEDHOPis orders of magnitude more densely connected than for W IKIHOP. This can lead to potentially large support document sets Sq, to a degree where it becomes computationally infeasible for a majority of existing RC models. After the traversal has finished, we subsample documents by first adding a set of documents that connects the drug in the query with its answer. We then iteratively add documents to connect alternative candidates until we reach the limit of 64 documents – while ensuring that all candidates have the same number of paths through the bipartite graph. Mitigating Candidate Frequency Imbalance Some drugs interact with more drugs than others –Aspirin for example interacts with 743 other drugs, but Isotretinoin with only 34. This leads to similar candidate frequency imbalance issues as with W IKIHOP– but due to its smaller size

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

21

drugs, but Isotretinoin with only 34. This leads to similar candidate frequency imbalance issues as with W IKIHOP– but due to its smaller size MEDHOPis difficult to sub-sample. Nevertheless we can successfully combat this issue by masking entity names, detailed in Section 6.2. 5 Dataset Analysis Table 1 shows the dataset sizes. Note that W IKIHOPinherits the train, development, and test set splits from W IKIREADING – i.e., the full dataset creation, filtering, and sub-sampling pipeline is executed on each set individually. Also note that subsampling according to document-answer correlation significantly reduces the size of W IKIHOPfrom 528K training samples to 44K. While in terms of samples, both W IKIHOPand M EDHOPare smaller than other large-scale RC datasets, such as SQuAD and W IKIREADING , the supervised learning signal available per sample is arguably greater. One could, for example, re-frame the task as binary path classification: given two entities and a document path

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

22

available per sample is arguably greater. One could, for example, re-frame the task as binary path classification: given two entities and a document path connecting them, determine whether a given relation holds. For such a case, W IKIHOPand M EDHOPwould have more than 1M and 150K paths to be classified, respectively. Instead, in our formulation, this corresponds to each single sample containing the supervised learning signal from an average of 19.5 and 59.8 unique document paths. Table 2 shows statistics on the number of candidates and documents per sample on the respective training sets. For M EDHOP, the majority of samples have 9 candidates, due to the way documentsTrain Dev Test Total WIKIHOP 43,738 5,129 2,451 51,318 MEDHOP 1,620 342 546 2,508 Table 1: Dataset sizes for our respective datasets. min max avg median # cand. – WH 2 79 19.8 14 # docs. – WH 3 63 13.7 11 # tok/doc – WH 4 2,046 100.4 91 # cand. – MH 2 9 8.9 9 # docs. – MH 5 64 36.4 29 # tok/doc – MH 5 458 253.9 264

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

23

# cand. – MH 2 9 8.9 9 # docs. – MH 5 64 36.4 29 # tok/doc – MH 5 458 253.9 264 Table 2: Candidates and documents per sample and document length statistics. WH: W IKIHOP; MH: M EDHOP. are selected up until a maximum of 64 documents is reached. Few samples have less than 9 candidates, and samples would have far more false candidates if more than 64 support documents were included. The number of query types in W IKIHOPis 277, whereas in M EDHOPthere is only one: interacts with . 5.1 Qualitative Analysis To establish the quality of the data and analyze potential distant supervision errors, we sampled and annotated 100 samples from each development set. WIKIHOP Table 3 lists characteristics along with the proportion of samples that exhibit them. For 45%, the true answer either uniquely follows from multiple texts directly or is suggested as likely. For 26%, more than one candidate is plausibly supported by the documents, including the correct answer. This is often due to hypernymy, where the appropriate level of granularity for the answer is difficult to predict – e.g. (west suffolk, administrative entity, ?) with candidates

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

24

the appropriate level of granularity for the answer is difficult to predict – e.g. (west suffolk, administrative entity, ?) with candidates suffolk and england . This is a direct consequence of including type-consistent false answer candidates from W IKIDATA , which can lead to questions with several true answers. For 9% of the cases a single document suffices; these samples contain a document that states enough information about item and answer together. For example, the query (Louis Auguste, father, ?) has the correct answer Louis XIV of France , and French Unique multi-step answer. 36% Likely multi-step unique answer. 9% Multiple plausible answers. 15% Ambiguity due to hypernymy. 11% Only single document required. 9% Answer does not follow. 12% WIKIDATA /WIKIPEDIA discrepancy. 8% Table 3: Qualitiative analysis of W IKIHOPsamples. king Louis XIV is mentioned within the same document as Louis Auguste . Finally, although our task is significantly more complex than most previous tasks where distant supervision has been applied, the distant supervision assumption is only violated for 20% of the samples – a proportion similar to previous work (Riedel et al., 2010). These

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

25

cases can either be due to conflicting information between W IKIDATA and W IKIPEDIA (8%), e.g. when the date of birth for a person differs between W IKIDATA and what is stated in the W IKIPEDIA article, or because the answer is consistent but cannot be inferred from the support documents (12%). When answering 100 questions, the annotator knew the answer prior to reading the documents for 9%, and produced the correct answer after reading the document sets for 74% of the cases. On 100 questions of a validated portion of the Dev set (see Section 5.3), 85% accuracy was reached. MEDHOP Since both document complexity and number of documents per sample were significantly larger compared to W IKIHOP, (see Figure 4 in Appendix B) it was not feasible to ask an annotator to read allsupport documents for 100 samples. We opted to verify the dataset quality by providing only the subset of documents relevant to support the correct answer, i.e., those traversed along the path reaching the answer. The annotator was asked if the answer to the query “follows” ,“is likely” , or“does not follow” , given the relevant documents. 68% of

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

26

answer to the query “follows” ,“is likely” , or“does not follow” , given the relevant documents. 68% of the cases were considered as “follows” or as “is likely” . The majority of cases violating the distant supervision assumption were due to lacking a necessary PPI in one of the connecting documents.5.2 Crowdsourced Human Annotation We asked human annotators on Amazon Mechanical Turk to evaluate samples of the W IKIHOPdevelopment set. Similar to our qualitative analysis of M EDHOP, annotators were shown the query-answer pair as a fact and the chain of relevant documents leading to the answer. They were then instructed to answer (1) whether they knew the fact before; (2) whether the fact follows from the texts (with options “fact follows” ,“fact is likely” , and “fact does not follow” ); and (3); whether a single or several of the documents are required. Each sample was shown to three annotators and a majority vote was used to aggregate the annotations. Annotators were familiar with the fact 4.6% of the time; prior knowledge of the fact is thus not likely to be a confounding effect on the other judgments. Inter-annotator agreement

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

27

with the fact 4.6% of the time; prior knowledge of the fact is thus not likely to be a confounding effect on the other judgments. Inter-annotator agreement as measured by Fleiss’ kappa is 0.253 in (2), and 0.281 in (3) – indicating a fair overall agreement, according to Landis and Koch (1977). Overall, 9.5% of samples have no clear majority in (2). Among samples with a majority judgment, 59.8% are cases where the fact “follows” , for 14.2% the fact is judged as “likely” , and as “not follow” for 25.9%. This again provides good justification for the distant supervision strategy. Among the samples with a majority vote for (2) of either “follows” or“likely” , 55.9% were marked with a majority vote as requiring multiple documents to infer the fact, and 44.1% as requiring only a single document. The latter number is larger than initially expected, given the construction of samples through graph traversal. However, when inspecting cases judged as “single” more closely, we observed

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

28

initially expected, given the construction of samples through graph traversal. However, when inspecting cases judged as “single” more closely, we observed that many indeed provide a clear hint about the correct answer within one document, but without stating it explicitly. For example, for the fact (witold cichy, country ofcitizenship, poland) with documents d1: Witold Cichy (born March 15, 1986 in Wodzisaw lski) is a Polish footballer[...] andd2: Wodzisaw lski[...] is a town in Silesian Voivodeship, southern Poland[...] , the information provided in d1 suffices for a human given the background knowledge that Polish is an attribute related to Poland , removing the need for d2to infer the answer. 5.3 Validated Test Sets While training models on distantly supervised data is useful, one should ideally evaluate methods on a manually validated test set. We thus identified subsets of the respective test sets for which the correct answer can be inferred from the text. This is in contrast to prior work such as Hermann et al. (2015), Hill et al. (2016), and Hewlett et al. (2016), who evaluate only on distantly supervised samples. For

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

29

Hill et al. (2016), and Hewlett et al. (2016), who evaluate only on distantly supervised samples. For WIKIHOP, we applied the same annotation strategy as described in Section 5.2. The validated test set consists of those samples labeled by a majority of annotators (at least 2 of 3) as “follows” , and requiring“multiple” documents. While desirable, crowdsourcing is not feasible for M EDHOPsince it requires specialist knowledge. In addition, the number of document paths is 3x larger, which along with the complexity of the documents greatly increases the annotation time. We thus manually annotated 20% of the M EDHOPtest set and identified the samples for which the text implies the correct answer and where multiple documents are required. 6 Experiments This section describes experiments on W IKIHOP and M EDHOPwith the goal of establishing the performance of several baseline models, including recent neural RC models. We empirically demonstrate the importance of mitigating dataset biases, probe whether multi-step behavior is beneficial for solving the task, and investigate if RC models can learn to perform lexical abstraction. Training will be conducted on the respective training sets, and evaluation

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

30

to perform lexical abstraction. Training will be conducted on the respective training sets, and evaluation on both the full test set and validated portion (Section 5.3) allowing for a comparison between the two. 6.1 Models Random Selects a random candidate; note that the number of candidates differs between samples. Max-mention Predicts the most frequently mentioned candidate in the support documents Sqof a sample – randomly breaking ties. Majority-candidate-per-query-type Predicts the candidate c2Cqthat was most frequently observed as the true answer in the training set, given the querytype of q. For W IKIHOP, the query type is the propertypof the query; for M EDHOPthere is only the single query type – interacts with . TF-IDF Retrieval-based models are known to be strong QA baselines if candidate answers are provided (Clark et al., 2016; Welbl et al., 2017). They search for individual documents based on keywords in the question, but typically do not combine information across documents. The purpose of this baseline is to see if it is possible to identify the correct answer from a single document alone through lexical correlations. The model forms its prediction as follows: For each candidate c, the concatenation of the query qwithcis fed as an ORquery into the

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

31

follows: For each candidate c, the concatenation of the query qwithcis fed as an ORquery into the whoosh text retrieval engine.5It then predicts the candidate with the highest TF-IDF similarity score: arg max c2Cq[max s2Sq(TF-IDF (q+c; s))] (1) Document-cue During dataset construction we observed that certain document-answer pairs appear more frequently than others, to the effect that the correct candidate is often indicated solely by the presence of certain documents in Sq. This baseline captures how easy it is for a model to exploit these informative document-answer co-occurrences. It predicts the candidate with highest score across Cq: arg max c2Cq[max d2Sq(cooccurrence (d; c))] (2) Extractive RC models: FastQA and BiDAF In our experiments we evaluate two recently proposed LSTM -based extractive QA models: the Bidirectional Attention Flow model ( BiDAF , Seo et al. (2017a)), and FastQA (Weissenborn et al., 2017), which have shown a robust performance across several datasets. These models predict an answer span

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

32

(2017a)), and FastQA (Weissenborn et al., 2017), which have shown a robust performance across several datasets. These models predict an answer span within a single document. We adapt them to a multidocument setting by sequentially concatenating all d2Sqin random order into a superdocument, adding document separator tokens. During training, the first answer mention in the concatenated document serves as the gold span.6At test time, we measured accuracy based on the exact match between 5https://pypi.python.org/pypi/Whoosh/ 6We also tested assigning the gold span randomly to any one of the mention of the answer, with insignificant changes. the prediction and answer, both lowercased, after removing articles, trailing white spaces and punctuation, in the same way as Rajpurkar et al. (2016). To rule out any signal stemming from the order of documents in the superdocument, this order is randomized both at training and test time. In a preliminary experiment we also trained models using different random document order permutations, but found that performance did not change significantly. ForBiDAF , the default hyperparameters from the implementation of Seo et al. (2017a) are used, with

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

33

ForBiDAF , the default hyperparameters from the implementation of Seo et al. (2017a) are used, with pretrained GloVe (Pennington et al., 2014) embeddings. However, we restrict the maximum document length to 8,192 tokens and hidden size to 20, and train for 5,000 iterations with batchsize 16 in order to fit the model into memory.7ForFastQA we use the implementation provided by the authors, also with pre-trained GloVe embeddings, no characterembeddings, no maximum support length, hidden size 50, and batch size 64 for 50 epochs. While BiDAF andFastQA were initially developed and tested on single-hop RC datasets, their usage of bidirectional LSTMs and attention over the full sequence theoretically gives them the capacity to integrate information from different locations in the (super-)document. In addition, BiDAF employs iterative conditioning across multiple layers, potentially making it even better suited to integrate information found across the sequence. 6.2 Lexical Abstraction: Candidate Masking The presence of lexical regularities among answers is a problem in RC dataset assembly – a phenomenon already observed by Hermann et al. (2015). When comprehending a text, the correct answer should become clear from its context – rather

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

34

phenomenon already observed by Hermann et al. (2015). When comprehending a text, the correct answer should become clear from its context – rather than from an intrinsic property of the answer expression. To evaluate the ability of models to rely on context alone, we created masked versions of the datasets: we replace any candidate expression randomly using 100 unique placeholder tokens, e.g. “Mumbai is the most populous city in MASK7 . ” Masking is consistent within one sample, but generally different for the same expression across samples. This not only removes answer frequency cues, 7The superdocument has a larger number of tokens compared to e.g. SQuAD , thus the additional memory requirements.Model Unfiltered Filtered Document-cue 74.6 36.7 Maj. candidate 41.2 38.8 TF-IDF 43.8 25.6 Train set size 527,773 43,738 Table 4: Accuracy comparison for simple baseline models on W IKIHOPbefore andafter filtering. it also removes statistical correlations between frequent answer strings and support documents. Models consequently cannot base their prediction on intrinsic properties of the answer expression, but have to rely on the context surrounding the mentions. 6.3 Results and Discussion

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

35

to rely on the context surrounding the mentions. 6.3 Results and Discussion Table 5 shows the experimental outcomes for W IKIHOPand M EDHOP, together with results for the masked setting; we will first discuss the former. A first observation is that candidate mention frequency does not produce better predictions than a random guess. Predicting the answer most frequently observed at training time achieves strong results: as much as 38.8% / 44.2% and 58.4% / 67.3% on the two datasets, for the full and validated test sets respectively. That is, a simple frequency statistic together with answer type constraints alone is a relatively strong predictor, and the strongest overall for the“unmasked” version of M EDHOP. The TF-IDF retrieval baseline clearly performs better than random for W IKIHOP, but is not very strong overall. That is, the question tokens are helpful to detect relevant documents, but exploiting only this information compares poorly to the other baselines. On the other hand, as no co-mention of an interacting drug pair occurs within any single document in M EDHOP, the TF-IDF baseline performs worse than random. We conclude that lexical matching with a single support document is not enough to build a strong predictive model for both datasets.

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

36

worse than random. We conclude that lexical matching with a single support document is not enough to build a strong predictive model for both datasets. TheDocument-cue baseline can predict more than a third of the samples correctly, for both datasets, even after sub-sampling frequent document-answer pairs for W IKIHOP. The relative strength of this and other baselines proves to be an important issue when designing multi-hop datasets, which we addressed through the measures described in SecWIKIHOP MEDHOP standard masked standard masked Model test test* test test* test test* test test* Random 11.5 12.2 12.2 13.0 13.9 20.4 14.1 22.4 Max-mention 10.6 15.9 13.9 20.1 9.5 16.3 9.2 16.3 Majority-candidate-per-query-type 38.8 44.2 12.0 13.7 58.4 67.3 10.4 6.1 TF-IDF 25.6 36.7 14.4 24.2 9.0 14.3 8.8 14.3 Document-cue 36.7 41.7 7.4 20.3 44.9 53.1 15.2 16.3

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

37

Document-cue 36.7 41.7 7.4 20.3 44.9 53.1 15.2 16.3 FastQA 25.7 27.2 35.8 38.0 23.1 24.5 31.3 30.6 BiDAF 42.9 49.7 54.5 59.8 47.8 61.2 33.7 42.9 Table 5: Test accuracies for the W IKIHOPand M EDHOPdatasets, both in standard (unmasked) and masked setup. Columns marked with asterisk are for the validated portion of the dataset. WIKIHOP MEDHOP standard gold chain standard gold chain Model test test* test test* test test* test test* BiDAF 42.9 49.7 57.9 63.4 47.8 61.2 86.4 89.8 BiDAF mask 54.5 59.8 81.2 85.7 33.7 42.9 99.3 100.0 FastQA 25.7 27.2 44.5 53.5 23.1 24.5 54.6 59.2 FastQA mask 35.8 38.0 65.3 70.0 31.3 30.6 51.8 55.1 Table 6: Test accuracy comparison when only using documents leading to the correct answer (gold chain). Columns

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

38

Table 6: Test accuracy comparison when only using documents leading to the correct answer (gold chain). Columns with asterisk hold results for the validated samples. tion 3.2. In Table 4 we compare the two relevant baselines on W IKIHOPbefore and after applying filtering measures. The absolute strength of these baselines before filtering shows how vital addressing this issue is: 74.6% accuracy could be reached through exploiting the cooccurrence (d; c)statistic alone. This underlines the paramount importance of investigating and addressing dataset biases that otherwise would confound seemingly strong RC model performance. The relative drop demonstrates that the measures undertaken successfully mitigate the issue. A downside to aggressive filtering is a significantly reduced dataset size, rendering it infeasible for smaller datasets like M EDHOP. Among the two neural models, BiDAF is overall strongest across both datasets – this is in contrast to the reported results for SQuAD where their performance is nearly indistinguishable. This is possibly due to the iterative latent interactions in the BiDAF architecture: we hypothesize that these are of increased importance for our task, where informationis distributed across documents. It is worth emphasizing that unlike the other baselines, both FastQA andBiDAF predict the answer by extracting a span

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

39

andBiDAF predict the answer by extracting a span from the support documents without relying on the candidate options Cq. In the masked setup all baseline models reliant on lexical cues fail in the face of the randomized answer expressions, since the same answer option has different placeholders in different samples. Especially on M EDHOP, where dataset sub-sampling is not a viable option, masking proves to be a valuable alternative, effectively circumventing spurious statistical correlations that RC models can learn to exploit. Both neural RC models are able to largely retain or even improve their strong performance when answers are masked: they are able to leverage the textual context of the candidate expressions. To understand differences in model behavior between W IKIHOPand M EDHOP, it is worth noting that drug mentions in M EDHOPare normalized to a unique single-word identifier, and performance drops under WIKIHOP MEDHOP test test* test test* BiDAF 54.5 59.8 33.7 42.9 BiDAF rem 44.6 57.7 30.4 36.7 FastQA 35.8 38.0 31.3 30.6 FastQA rem 38.0 41.2 28.6 24.5 Table 7: Test accuracy (masked) when only documents

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

40

FastQA rem 38.0 41.2 28.6 24.5 Table 7: Test accuracy (masked) when only documents containing answer candidates are given ( rem). masking. In contrast, for the open-domain setting of WIKIHOP, a reduction of the answer vocabulary to 100 random single-token mask expressions clearly helps the model in selecting a candidate span, compared to the multi-token candidate expressions in the unmasked setting. Overall, although both neural RC models clearly outperform the other baselines, they still have large room for improvement compared to human performance at 74% / 85% for W IKIHOP. Comparing results on the full and validated test sets, we observe that the results consistently improve on the validated sets. This suggests that the training set contains the signal necessary to make inference on valid samples at test time, and that noisy samples are harder to predict. 6.4 Using only relevant documents We conducted further experiments to examine the RC models when presented with only the relevant documents in Sq, i.e., the chain of documents leading to the correct answer. This allows us to investigate the hypothetical performance of the models if they were able to select and read only relevant documents: Table 6 summarizes these results. Models improve greatly in this gold chain setup, with up

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

41

they were able to select and read only relevant documents: Table 6 summarizes these results. Models improve greatly in this gold chain setup, with up to 81.2% / 85.7% on W IKIHOPin the masked setting for BiDAF . This demonstrates that RC models are capable of identifying the answer when few or no plausible false candidates are mentioned, which is particularly evident for M EDHOP, where documents tend to discuss only single drug candidates. In the masked gold chain setup, models can then pick up on what the masking template looks like and achieve almost perfect scores. Conversely, these results also show that the models’ answer selection process is not robust to the introduction of unrelated documents with type-consistent candidates.This indicates that learning to intelligently select relevant documents before RC may be among the most promising directions for future model development. 6.5 Removing relevant documents To investigate if the neural RC models can draw upon information requiring multi-step inference we designed an experiment where we discard all documents that do not contain candidate mentions, including the first documents traversed. Table 7 shows the results: we can observe that performance drops across the board for BiDAF . There is a significant drop of 3.3%/6.2% on M EDHOP, and 10.0%/2.1%

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611

1710.06481

42

drop of 3.3%/6.2% on M EDHOP, and 10.0%/2.1% on W IKIHOP, demonstrating that BiDAF , is able to leverage cross-document information. FastQA shows a slight increase of 2.2%/3.2% for W IKIHOP and a decrease of 2.7%/4.1% on M EDHOP. While inconclusive, it is clear that FastQA with fewer latent interactions than BiDAF has problems integrating cross-document information. 7 Related Work Related Datasets End-to-end text-based QA has witnessed a surge in interest with the advent of largescale datasets, which have been assembled based on F REEBASE (Berant et al., 2013; Bordes et al., 2015), W IKIPEDIA (Yang et al., 2015; Rajpurkar et al., 2016; Hewlett et al., 2016), web search queries (Nguyen et al., 2016), news articles (Hermann et al., 2015; Onishi et al., 2016), books (Hill et al., 2016; Paperno et al., 2016), science exams (Welbl et al., 2017), and trivia (Boyd-Graber

1710.06481

Constructing Datasets for Multi-hop Reading Comprehension Across Documents

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

http://arxiv.org/pdf/1710.06481

This paper directly corresponds to the TACL version (https://transacl.org/ojs/index.php/tacl/article/view/1325) apart from minor changes in wording, additional footnotes, and appendices

Transactions of the Association for Computational Linguistics (TACL), Vol 6 (2018), pages 287-302

cs.CL

20171017

20180611