Language (Re)modelling: Towards Embodied Language Understanding

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 6268–6281 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 6268 Language (Re)modelling: Towards Embodied Language Understanding Ronen Tamari† Chen Shani† Tom Hope⋆∗ Miriam R. L. Petruck‡ Omri Abend† Dafna Shahaf† †The Hebrew University of Jerusalem ⋆Allen Institute for Artificial Intelligence ∗Paul G. Allen School of Computer Science & Engineering, University of Washington ‡International Computer Science Institute, Berkeley, CA {ronent,chenxshani,oabend,dshahaf}@cs.huji.ac.il tomh@allenai.org miriamp@icsi.berkeley.edu Abstract While natural language understanding (NLU) is advancing rapidly, today’s technology dif- fers from human-like language understanding in fundamental ways, notably in its inferior efficiency, interpretability, and generalization. This work proposes an approach to represen- tation and learning based on the tenets of em- bodied cognitive linguistics (ECL). According to ECL, natural language is inherently exe- cutable (like programming languages), driven by mental simulation and metaphoric map- pings over hierarchical compositions of struc- tures and schemata learned through embodied interaction. This position paper argues that the use of grounding by metaphoric inference and simulation will greatly benefit NLU systems, and proposes a system architecture along with a roadmap towards realizing this vision. 1 Introduction “Not those speaking the same language, but those sharing the same feeling under- stand each other.” – Jalal ad-Din Rumi While current NLU systems “speak” human lan- guage by learning strong statistical models, they do not possess anything like the rich mental repre- sentations that people utilize for language under- standing. Indeed, despite the tremendous progress in NLU, recent work shows that today’s state-of- the-art (SOTA) systems differ from human-like lan- guage understanding in crucial ways, in particular in their generalization, grounding, reasoning, and explainability capabilities (Glockner et al., 2018; McCoy et al., 2019a,b; Nie et al., 2019; Yogatama et al., 2019; Lake et al., 2019). Question-answering (QA) is currently one of the predominant methods of training deep-learning models for general, open-domain language under- standing (Gardner et al., 2019b). While QA is a ver- satile, broadly-applicable framework, recent stud- ies have shown it to be fraught with pitfalls (Gard- ner et al., 2019a; Mudrakarta et al., 2018). A recent workshop on QA for reading comprehension sug- gested that “There is growing realization that the traditional supervised learning paradigm is broken [...] – we’re fitting artifacts” (Gardner, 2019). In many respects, the problems of NLU mirror those of artificial intelligence (AI) research in gen- eral. Lake et al.’s (2017a) seminal work identified a significant common factor at the root of problems in general AI. The current deep-learning paradigm is a statistical pattern-recognition approach predom- inantly applied to relatively narrow task-specific prediction. In contrast, human cognition supports a wide range of inferences (planning, action, ex- plaining, etc.), hinting at a view of intelligence fo- cused on model-building, specifically, mental mod- els: rich, structured, manipulable, and explainable representations useful for performing in dynamic, uncertain environments. This distinction motivates the quest for a new cognitively-inspired model- building learning paradigm for general AI, which has inspired fruitful subsequent research and dis- cussion (e.g., Lake et …

Content truncated to 500 words