Dataset origin: https://lrec2020.lrec-conf.org/sharedlrs2020/367_res_1.zip

Description

This resource contains a context-independent gold standard for English-Dutch and French-Dutch cognate detection. To this end, automatic word alignment was applied on the Dutch Parallel Corpus, and all term equivalents with a Normalized Levenshtein distance smaller than 0.5 were extracted. This resulted in a list with 28,503 English-Dutch candidate cognate pairs, and 22,715 French-Dutch candidate cognate pairs, which were subsequently manually labeled according to the guidelines established in Labat et al. 2019. The following labels were annotated: (1) Cognate: words which have a similar form and meaning in all contexts, (2) Partial cognate: words which have a similar form, but only share the same meaning in some contexts, (3) False friend: words which have a similar form but a different meaning, (4) Proper name: proper nouns (e.g. persons, companies, cities, coun-tries, etc.) and their derivations, (5) Error: word alignment errors and compound nouns of which one part is a cognate but the other part is missing in one of the languages, and (6) No standard: words that do not occur in the dictionary of that particular language.

Citation

@inproceedings{lefever-etal-2020-identifying,
    title = "Identifying Cognates in {E}nglish-{D}utch and {F}rench-{D}utch by means of Orthographic Information and Cross-lingual Word Embeddings",
    author = "Lefever, Els  and
      Labat, Sofie  and
      Singh, Pranaydeep",
    editor = "Calzolari, Nicoletta  and
      B{\'e}chet, Fr{\'e}d{\'e}ric  and
      Blache, Philippe  and
      Choukri, Khalid  and
      Cieri, Christopher  and
      Declerck, Thierry  and
      Goggi, Sara  and
      Isahara, Hitoshi  and
      Maegaard, Bente  and
      Mariani, Joseph  and
      Mazo, H{\'e}l{\`e}ne  and
      Moreno, Asuncion  and
      Odijk, Jan  and
      Piperidis, Stelios",
    booktitle = "Proceedings of the Twelfth Language Resources and Evaluation Conference",
    month = may,
    year = "2020",
    address = "Marseille, France",
    publisher = "European Language Resources Association",
    url = "https://aclanthology.org/2020.lrec-1.504",
    pages = "4096--4101",
    abstract = "This paper investigates the validity of combining more traditional orthographic information with cross-lingual word embeddings to identify cognate pairs in English-Dutch and French-Dutch. In a first step, lists of potential cognate pairs in English-Dutch and French-Dutch are manually labelled. The resulting gold standard is used to train and evaluate a multi-layer perceptron that can distinguish cognates from non-cognates. Fifteen orthographic features capture string similarities between source and target words, while the cosine similarity between their word embeddings represents the semantic relation between these words. By adding domain-specific information to pretrained fastText embeddings, we are able to obtain good embeddings for words that did not yet have a pretrained embedding (e.g. Dutch compound nouns). These embeddings are then aligned in a cross-lingual vector space by exploiting their structural similarity (cf. adversarial learning). Our results indicate that although the classifier already achieves good results on the basis of orthographic information, the performance further improves by including semantic information in the form of cross-lingual word embeddings.",
    language = "English",
    ISBN = "979-10-95546-34-4",
}