Update README.md

README.md

@@ -27,6 +27,26 @@ Our dataset treats ''unfair contractual term'' as the equivalent of an abusive c
 
 On the EU and at the national such as the Polish levels, agencies cannot check possible agreements by hand. Hence, we took the first step to evaluate the possibility of accelerating this process. We created a dataset and machine learning models to automate potentially abusive clauses detection partially. Consumer protection organizations and agencies can use these resources to make their work more effective and efficient. Moreover, consumers can automatically analyze contracts and understand what they agree upon.
 
+## Tasks (input, output and metrics)
+
+Abusive Clauses Detection
+
+**Input** ('*text'* column): text of agreement 
+
+**Output** ('*label'* column): binary label (`BEZPIECZNE_POSTANOWIENIE_UMOWNE`: correct agreement statement, `KLAUZULA_ABUZYWNA`: abusive clause)
+
+**Domain**: legal agreement
+
+**Measurements**: Accuracy, F1 Macro
+
+**Example***:* 
+
+Input: *`Wszelka korespondencja wysyłana przez Pożyczkodawcę na adres zamieszkania podany w umowie oraz na e-mail zostaje uznana za skutecznie doręczoną. Zmiana adresu e-mail oraz adresu zamieszkania musi być dostarczona do Pożyczkodawcy osobiście`* 
+
+Input (translated by DeepL): *`All correspondence sent by the Lender to the residential address provided in the agreement and to the e-mail address shall be deemed effectively delivered. Change of e-mail address and residential address must be delivered to the Lender in person`*
+
+Output: `KLAUZULA_ABUZYWNA` (abusive clause)
+
 ## Data splits
 
 | Subset      | Cardinality (sentences) |