Climbing towards NLU: On Meaning, Form, and Understanding in the Age of Data

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 5185–5198 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 5185 Climbing towards NLU: On Meaning, Form, and Understanding in the Age of Data Emily M. Bender University of Washington Department of Linguistics ebender@uw.edu Alexander Koller Saarland University Dept. of Language Science and Technology koller@coli.uni-saarland.de Abstract The success of the large neural language mod- els on many NLP tasks is exciting. However, we find that these successes sometimes lead to hype in which these models are being de- scribed as “understanding” language or captur- ing “meaning”. In this position paper, we ar- gue that a system trained only on form has a priori no way to learn meaning. In keeping with the ACL 2020 theme of “Taking Stock of Where We’ve Been and Where We’re Going”, we argue that a clear understanding of the dis- tinction between form and meaning will help guide the field towards better science around natural language understanding. 1 Introduction The current state of affairs in NLP is that the large neural language models (LMs), such as BERT (De- vlin et al., 2019) or GPT-2 (Radford et al., 2019), are making great progress on a wide range of tasks, including those that are ostensibly meaning- sensitive. This has led to claims, in both academic and popular publications, that such models “under- stand” or “comprehend” natural language or learn its “meaning”. From our perspective, these are overclaims caused by a misunderstanding of the relationship between linguistic form and meaning. We argue that the language modeling task, be- cause it only uses form as training data, cannot in principle lead to learning of meaning. We take the term language model to refer to any system trained only on the task of string prediction, whether it operates over characters, words or sentences, and sequentially or not. We take (linguistic) meaning to be the relation between a linguistic form and communicative intent. Our aim is to advocate for an alignment of claims and methodology: Human-analogous natural lan- guage understanding (NLU) is a grand challenge of artificial intelligence, which involves mastery of the structure and use of language and the ability to ground it in the world. While large neural LMs may well end up being important components of an eventual full-scale solution to human-analogous NLU, they are not nearly-there solutions to this grand challenge. We argue in this paper that gen- uine progress in our field — climbing the right hill, not just the hill on whose slope we currently sit — depends on maintaining clarity around big picture notions such as meaning and understanding in task design and reporting of experimental results. After briefly reviewing the ways in which large LMs are spoken about and summarizing the re- cent flowering of “BERTology” papers (§2), we offer a working definition for “meaning” (§3) and a series of thought experiments illustrating the im- possibility of learning meaning when it is not in the training signal (§4,5). We …

Content truncated to 500 words