On the Linguistic Representational Power of Neural Machine Translation Models

On the Linguistic Representational Power of Neural Machine Translation Models Yonatan Belinkov∗ Massachusetts Institute of Technology Computer Science and Artificial Intelligence Laboratory Harvard University John F. Paulson School of Engineering and Applied Sciences belinkov@mit.edu Nadir Durrani∗ Qatar Computing Research Institute HBKU Research Complex ndurrani@qf.org.qa Fahim Dalvi Qatar Computing Research Institute HBKU Research Complex faimaduddin@qf.org.qa Hassan Sajjad Qatar Computing Research Institute HBKU Research Complex hsajjad@qf.org.qa James Glass Massachusetts Institute of Technology Computer Science and Artificial Intelligence Laboratory glass@mit.edu ∗ Authors contributed equally. Submission received: 30 November 2018; revised version received: 21 July 2019; accepted for publication: 17 September 2019. https://doi.org/10.1162/COLI a 00367 © 2020 Association for Computational Linguistics Published under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0) license Computational Linguistics Volume 46, Number 1 Despite the recent success of deep neural networks in natural language processing and other spheres of artificial intelligence, their interpretability remains a challenge. We analyze the repre- sentations learned by neural machine translation (NMT) models at various levels of granularity and evaluate their quality through relevant extrinsic properties. In particular, we seek answers to the following questions: (i) How accurately is word structure captured within the learned rep- resentations, which is an important aspect in translating morphologically rich languages? (ii) Do the representations capture long-range dependencies, and effectively handle syntactically divergent languages? (iii) Do the representations capture lexical semantics? We conduct a thorough investigation along several parameters: (i) Which layers in the architecture capture each of these linguistic phenomena; (ii) How does the choice of translation unit (word, character, or subword unit) impact the linguistic properties captured by the underlying representations? (iii) Do the encoder and decoder learn differently and independently? (iv) Do the representations learned by multilingual NMT models capture the same amount of linguistic information as their bilingual counterparts? Our data-driven, quantitative evaluation illuminates important aspects in NMT models and their ability to capture various linguistic phenomena. We show that deep NMT models trained in an end-to-end fashion, without being provided any direct supervision during the training process, learn a non-trivial amount of linguistic information. Notable findings include the following observations: (i) Word morphology and part-of-speech information are captured at the lower layers of the model; (ii) In contrast, lexical semantics or non-local syntactic and semantic dependencies are better represented at the higher layers of the model; (iii) Representations learned using characters are more informed about word-morphology compared to those learned using subword units; and (iv) Representations learned by multilingual models are richer compared to bilingual models. 1. Introduction Deep neural networks have quickly become the predominant approach to most tasks in artificial intelligence, including machine translation (MT). Compared with their tradi- tional counterparts, these models are trained in an end-to-end fashion, providing a sim- ple yet elegant mechanism. This simplicity, however, comes at the price of opaqueness. Unlike traditional systems that contain specialized modules carrying specific sub-tasks, neural MT (NMT) systems train one large network, optimized toward the overall task. For example, non-neural statistical MT systems have sub-components to handle fluency (Heafield …

Content truncated to 500 words