Assessing Phrasal Representation and Composition in Transformers

Abstract

Deep transformer models have pushed performance on NLP tasks to new limits, suggesting sophisticated treatment of complex linguistic inputs, such as phrases. However, we have limited understanding of how these models handle representation of phrases, and whether this reflects sophisticated composition of phrase meaning like that done by humans. In this paper, we present systematic analysis of phrasal representations in state-of-the-art pre-trained transformers. We use tests leveraging human judgments of phrase similarity and meaning shift, and compare results before and after control of word overlap, to tease apart lexical effects versus composition effects. We find that phrase representation in these models relies heavily on word content, with little evidence of nuanced composition. We also identify variations in phrase representation quality across models, layers, and representation types, and make corresponding recommendations for usage of representations from these models.

Dynabench: Rethinking Benchmarking in NLP

“…Whether our models have learned to solve tasks in robust and generalizable ways has been a topic of much recent interest. Challenging test sets have shown that many state-of-the-art NLP models struggle with compositionality Kim and Linzen, 2020;Yu and Ettinger, 2020;White et al, 2020), and find it difficult to pass the myriad stress tests for social May et al, 2019;Nangia et al, 2020) and/or linguistic competencies (Geiger et al, 2018;Naik et al, 2018;Glockner et al, 2018;White et al, 2018;Warstadt et al, 2019;Gauthier et al, 2020;Hossain et al, 2020;Jeretic et al, 2020;Lewis et al, 2020;Saha et al, 2020;Schuster et al, 2020;Sugawara et al, 2020;. Yet, challenge sets may suffer from performance instability (Liu et al, 2019a;Rozen et al, 2019; and often lack sufficient statistical power (Card et al, 2020), suggesting that, although they may be valuable assessment tools, they are not sufficient for ensuring that our models have achieved the learning targets we set for them.…”

On the Interplay Between Fine-tuning and Composition in Transformers

“…Phrase and sentence composition has drawn frequent attention in analysis of neural models, often focusing on analysis of internal representations and downstream task behavior (Ettinger et al, 2018;Conneau et al, 2019;Nandakumar et al, 2019;Yu and Ettinger, 2020;Bhathena et al, 2020;Mu and Andreas, 2020;Andreas, 2019). Some work investigates compositionality via constructing linguistic (Keysers et al, 2019) and non-linguistic (Liška et al, 2018;Hupkes et al, 2018;Baan et al, 2019) and Ettinger (2020).…”

“…The versatility of these pre-trained models suggests that they may acquire fairly robust linguistic knowledge and capacity for natural language "understanding". However, an emerging body of analysis demonstrates a level of superficiality in these models' handling of language (Niven and Kao, 2019;Kim and Linzen, 2020;Ettinger, 2020;Yu and Ettinger, 2020).…”

“…In particular, although composition-a model's capacity to combine meaning units into more com-plex units reflecting phrase meanings-is an indispensable component of language understanding, when testing for composition in pre-trained transformer representations, Yu and Ettinger (2020) report that these representations reflect word content of phrases, but don't show signs of more sophisticated humanlike composition beyond word content. In the present paper we perform a direct followup of that study, asking whether models will show better evidence of composition after fine-tuning on tasks that are good candidates for requiring composition: 1) the Quora Question Pairs dataset in Paraphrase Adversaries from Word Scrambling (PAWS-QQP) , an adversarial paraphrase dataset forcing models to classify paraphrases with high lexical overlap, and 2) the Stanford Sentiment Treebank (Socher et al, 2013), a sentiment dataset with fine-grained phrase labels to promote composition.…”

“…In the present paper we perform a direct followup of that study, asking whether models will show better evidence of composition after fine-tuning on tasks that are good candidates for requiring composition: 1) the Quora Question Pairs dataset in Paraphrase Adversaries from Word Scrambling (PAWS-QQP) , an adversarial paraphrase dataset forcing models to classify paraphrases with high lexical overlap, and 2) the Stanford Sentiment Treebank (Socher et al, 2013), a sentiment dataset with fine-grained phrase labels to promote composition. We base our analysis on the tests proposed by Yu and Ettinger (2020), which rely on alignment with human judgments of phrase pair similarities, and which leverage control of lexical overlap to target compositionality. We fine-tune and evaluate the same models and representation types tested in that paper, for optimal comparison.…”