Cognition-aware Cognate Detection

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 3281–3292 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 3281 Cognition-aware Cognate Detection Diptesh Kanojia†,♣,⋆, Prashant Sharma⋄, Sayali Ghodekar‡, Pushpak Bhattacharyya†, Gholamreza Haffari⋆, and Malhar Kulkarni† †IIT Bombay, India, ⋄Hitachi CRL, Japan, ‡RingCentral, India ♣IITB-Monash Research Academy, India ⋆Monash University, Australia †{diptesh,pb,malhar}@iitb.ac.in, ⋄prashaantsharmaa@gmail.com ‡sayalighodekar26@gmail.com , ⋆gholamreza.haffari@monash.edu Abstract Automatic detection of cognates helps down- stream NLP tasks of Machine Translation, Cross-lingual Information Retrieval, Computa- tional Phylogenetics and Cross-lingual Named Entity Recognition. Previous approaches for the task of cognate detection use orthographic, phonetic and semantic similarity based fea- tures sets. In this paper, we propose a novel method for enriching the feature sets, with cog- nitive features extracted from human readers’ gaze behaviour. We collect gaze behaviour data for a small sample of cognates and show that extracted cognitive features help the task of cognate detection. However, gaze data col- lection and annotation is a costly task. We use the collected gaze behaviour data to predict cognitive features for a larger sample and show that predicted cognitive features, also, signifi- cantly improve the task performance. We re- port improvements of 10% with the collected gaze features, and 12% using the predicted gaze features, over the previously proposed approaches. Furthermore, we release the col- lected gaze behaviour data along with our code and cross-lingual models. 1 Introduction Cognates are word pairs, across languages, which have a common etymological origin. For example, the French and English word pair, Libert´e - Liberty, reveals itself to be a cognate through orthographic similarity. The task of automatic cognate detection across languages1 requires one to detect word pairs which are etymologically related, and carry the same meaning. Such word pairs share a formal and/or semantic affinity and facilitate the second language acquisition process, particularly between related languages. Although they can accelerate vocabulary acquisition, language learners also need 1Cognates can also exist in the same language. Such word pairs/sets are commonly referred to as doublets. to be aware of false friends and partial cognates, at times, leading to unrelated semantic coupling. For example, “gift” in German means “poison”, which is a known example of a False Friend pair. For an example of a partial cognate, the word “police” can translate to “police”, “policy” or “font”, in French, depending on the context in which it was used. Manual detection of such cognate sets requires a human expert with a good linguistic background in multiple languages. Moreover, manual annotation of cognate sets is a costly task in terms of time and human effort. Automatic Cognate Detection (ACD) is a well-known task, which has been explored for a range of languages using different methods; and has shown to help Cross-lingual Information Re- trieval (Meng et al., 2001), Machine Translation (MT) (Al-Onaizan et al., 1999), and Computational Phylogenetics (Rama et al., 2018). In the tradi- tional approaches to automatic cognate detection, words with similar meanings or forms are used as probable cognates (J¨ager et al., …

Content truncated to 500 words