Interpretability for Morphological Inflection: from Character-level Predictions to Subword-level Rules

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 3189–3201 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 3189 Interpretability for Morphological Inflection: from Character-level Predictions to Subword-level Rules Tatyana Ruzsics Olga Sozinova Ximena Gutierrez-Vasques Tanja Samardˇzi´c Text Group, URPP Language and Space, University of Zurich, Switzerland {tatiana.ruzsics, olga.sozinova, ximena.gutierrezvasques, tanja.samardzic}@uzh.ch Abstract Neural models for morphological inflection have recently attained very high results. How- ever, their interpretation remains challeng- ing. Towards this goal, we propose a simple linguistically-motivated variant to the encoder- decoder model with attention. In our model, character-level cross-attention mechanism is complemented with a self-attention module over substrings of the input. We design a novel approach for pattern extraction from at- tention weights to interpret what the model learn. We apply our methodology to analyze the model’s decisions on three typologically- different languages and find that a) our pattern extraction method applied to cross-attention weights uncovers variation in form of inflec- tion morphemes, b) pattern extraction from self-attention shows triggers for such variation, c) both types of patterns are closely aligned with grammar inflection classes and class as- signment criteria, for all three languages. Ad- ditionally, we find that the proposed encoder attention component leads to consistent perfor- mance improvements over a strong baseline. 1 Introduction With the rise of deep learning, neural networks have been nowadays used in the process of deci- sion making in various domains, as different as trading, medicine and government. Ethical consid- erations of such decisions have led to an increas- ing need for interpreting neural models which is a vivid research topic in machine learning commu- nity (Lipton, 2018; Gilpin et al., 2018). Although interpretability research in NLP is partly driven by ethics (Jacovi and Goldberg, 2020), there is a growing body of work exploring what linguis- tic properties emerge in neural models (Belinkov and Glass, 2019; Manning et al., 2020). The latter line of work aims to aid and scale up linguistic research, which is also the topic of this paper. Lin- guistics research focuses on uncovering patterns and regularities in language. Retrieving and ana- lyzing structures of languages learned by neural agents can systematize our knowledge and, ide- ally, help us to come up with new regularities. Re- cent advances (Schrimpf et al., 2020) in testing hypotheses about human language processing us- ing the growing suite of modern interpretable NLP are, indeed, inspiring, but still relatively limited to few languages. To scale up linguistic research, we require truly language-independent models de- veloped for languages other than English (Bender, 2011). In return, understanding the model’s deci- sions can lead to new ideas, how to improve the performance of the model on hard cases, i.e. on a particular linguistic phenomenon or language. In our work, we concentrate on interpretability methods suitable for examining what knowledge of inflection morphology is captured by neural net- works. Specifically, we consider a neural model that learns a mapping from a lemma and an abstract morpho-syntactic definition (MSD) …

Content truncated to 500 words