Grammatical illusions in BERT: Attraction effects of subject-verb agreement and reflexive-antecedent dependencies
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
The phenomenon of attraction effects, whereby a verb erroneously retrieves a syntactically inaccessible but feature-matching noun, is a type of grammatical illusions (Phillips, Wagers, and Lau 2011) that can occur in long-distance subject-verb agreement in human sentence processing (Wagers et al. 2009). In contrast, reflexive-antecedent dependencies have been claimed to lack attraction effects when the reflexive and the antecedent mismatch (Dillon et al. 2013). Yet, some other studies have shown that attraction effects have been observed in reflexive-antecedent dependencies, when the number of feature mismatch between the reflexive and the antecedent increases (Parker and Philips 2017). These findings suggest that there are different cue weightings based on the predictability of the dependency, and these cues are combined according to different cue-combination scheme, such as a linear or a non-linear cue-combination rule (Parker 2019). These linguistic phenomena can be used to analyze how linguistic features are accessed and combined within the internal states of Deep Neural Network (DNN) language models. In the linguistic representations of BERT (Devlin et al. 2018), one of the pre-trained DNN language models, various types of linguistic information are encoded in each layer (Jawahar et al. 2019) and combined while passing through the layers. By measuring the performance of Masked Language Model (MLM), this study finds that both subject-verb agreement and reflexive-antecedent dependencies show attraction effects and follow the linear-combinatoric rule in BERT. The different results from human sentence processing suggest that the self-attention mechanism of BERT may not be able to capture the differences in the predictability of the dependency as effectively as memory retrieval mechanisms in humans. These findings have important implications for developing more understandable and interpretable explainable-AI (xAI) systems that better capture the complexities of human language processing.
Abstract
The phenomenon of attraction effects, whereby a verb erroneously retrieves a syntactically inaccessible but feature-matching noun, is a type of grammatical illusions (Phillips, Wagers, and Lau 2011) that can occur in long-distance subject-verb agreement in human sentence processing (Wagers et al. 2009). In contrast, reflexive-antecedent dependencies have been claimed to lack attraction effects when the reflexive and the antecedent mismatch (Dillon et al. 2013). Yet, some other studies have shown that attraction effects have been observed in reflexive-antecedent dependencies, when the number of feature mismatch between the reflexive and the antecedent increases (Parker and Philips 2017). These findings suggest that there are different cue weightings based on the predictability of the dependency, and these cues are combined according to different cue-combination scheme, such as a linear or a non-linear cue-combination rule (Parker 2019). These linguistic phenomena can be used to analyze how linguistic features are accessed and combined within the internal states of Deep Neural Network (DNN) language models. In the linguistic representations of BERT (Devlin et al. 2018), one of the pre-trained DNN language models, various types of linguistic information are encoded in each layer (Jawahar et al. 2019) and combined while passing through the layers. By measuring the performance of Masked Language Model (MLM), this study finds that both subject-verb agreement and reflexive-antecedent dependencies show attraction effects and follow the linear-combinatoric rule in BERT. The different results from human sentence processing suggest that the self-attention mechanism of BERT may not be able to capture the differences in the predictability of the dependency as effectively as memory retrieval mechanisms in humans. These findings have important implications for developing more understandable and interpretable explainable-AI (xAI) systems that better capture the complexities of human language processing.
참고문헌 (61)
- Baayen, R. Harald, "Mixed-effects modeling with crossed random effects for subjects and items", Journal of Memory and Language, 59(4), pp.390~412 (2008)
- Bacon, Geoff, "Does BERT agree? Evaluating knowledge of structure dependence through agreement relations", arXiv:1908.09892, ~ (2019)
- Badecker, William, "The processing role of structural constraints on interpretation of pronouns and anaphors", Journal of Experimental Psychology: Learning, Memory, and Cognition, 28(4), pp.748~769 (2002)
- Bates, Douglas, "Fitting linear mixed-effects models using lme4", Journal of Statistical Software, 67(1), pp.1~48 (2015)
- Belinkov, Yonatan, "Analysis methods in neural language processing: A survey", Transactions of the Association for Computational Linguistics, 7, pp.49~72 (2019)
- Bock, Kathryn, "Broken agreement", Cognitive Psychology, 23(1), pp.45~93 (1991)
- Child, Rewon, "Generating long sequences with sparse transformers", arXiv:1904.10509, ~ (2019)
- Cho, Won Ik, "Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021", pp.2922~2929 (2021), Association for Computational Linguistics
- Chomsky, Noam, "Lectures on government and binding" (1981), Foris
- Correia, Gonçalo M., "Adaptively sparse transformers", Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pp.2174~2184 (2019)
- Cunnings, Ian, "The role of working memory in the processing of reflexives", Language and Cognitive Processes, 28(1-2), pp.188~219 (2013)
- Cunnings, Ian, "Coargumenthood and the processing of reflexives", Journal of Memory and Language, 75, pp.117~139 (2014)
- Dai, Zihang, "Transformer-xl: Attentive language models beyond a fixed-length context", Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pp.2978~2988 (2019)
- Devlin, Jacob, "Bert: Pre-training of deep bidirectional transformers for language understanding", Proceedings of the North American Chapter of the Association for Computational Linguistics-Human Language Technologies, pp.4171~4186 (2018), Association for Computational Linguistics
- Dillon, Brian William, "Structured access in sentence comprehension", PhD, pp.2011
- Dillon, Brian, "Contrasting intrusion profiles for agreement and anaphora: Experimental and modeling evidence", Journal of Memory and Language, 69(2), pp.85~103 (2013)
- Goldberg, Yoav, "Assessing BERT's syntactic abilities", arXiv:1901.05287, ~ (2019)
- Gronlund, Scott D., "Comparison of the retrieval of item versus spatial position information", Journal of Experimental Psychology: Learning, Memory, and Cognition, 23(5), pp.1261~1274 (1997)
- Gulordava, Kristina, "Colorless green recurrent networks dream hierarchically", Proceedings of the North American Chapter of the Association for Computational Linguistics-Human Language Technologies, pp.1195~1205 (2018), Association for Computational Linguistics
- Hale, John, "A probabilistic Earley parser as a psycholinguistic model", Proceedings of the second meeting of the North American Chapter of the Association for Computational Linguistics, pp.159~166 (2001), Association for Computational Linguistics
- Hewitt, John, "A structural probe for finding syntax in word representations", Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 1, pp.4129~4138 (2019), Association for Computational Linguistics
- Holzinger, Andreas, "Causability and explainability of artificial intelligence in medicine", Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, 9(4), pp.e1312~ (2019)
- Jawahar, Ganesh, "What does BERT learn about the structure of language?", Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pp.3651~3657 (2019), Association for Computational Linguistics
- Kim, Nayoun, "The online processing of noun phrase ellipsis and mechanisms of antecedent retrieval", Language, Cognition and Neuroscience, 34(2), pp.190~213 (2019)
- King, Joseph, "Do reflexives always find a grammatical antecedent for themselves", Poster presented at the 25th Annual Meeting of the CUNY Conference on Human Sentence Processing (2012), The CUNY Graduate Center
- Klafka, Josef, "Spying on your neighbors: Fine-grained probing of contextual embeddings for information about surrounding words", Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pp.4801~4811 (2020)
- Kush, Dave W, "Respecting relations: Memory access and antecedent retrieval in incremental sentence processing", PhD, pp.2013
- Lago, Sol, "Agreement attraction in Spanish comprehension", Journal of Memory and Language, 82, pp.133~149 (2015)
- Levy, Roger, "Expectation-based syntactic comprehension", Cognition, 106(3), pp.1126~1177 (2008)
- Lewis, Richard L., "An activation-based model of sentence processing as skilled memory retrieval", Cognitive Science, 29(3), pp.375~419 (2005)
- Lin, Yongjie, "Open Sesame: getting inside BERT's linguistic knowledge", Proceedings of the 2019 ACL Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP, pp.241~253 (2019)
- Linzen, Tal, "Assessing the ability of LSTMs to learn syntax-sensitive dependencies", Transactions of the Association for Computational Linguistics, 4, pp.521~535 (2016)
- Liu, Nelson F., "Linguistic knowledge and transferability of contextual representations", Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 1, pp.1073~1094 (2019), Association for Computational Linguistics
- Marvin, Rebecca, "Targeted syntactic evaluation of language models", Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pp.1192~1202 (2018)
- McElree, Brian, "Sentence comprehension is mediated by content-addressable memory structures", Journal of Psycholinguistic Research, 29(2), pp.111~123 (2000)
- McElree, Brian, "Working memory and focal attention", Journal of Experimental Psychology : Learning, Memory, and Cognition, 27(3), pp.817~ (2001)
- McElree, Brian, "Accessing recent events", Psychology of Learning and Motivation, 46, pp.155~200 (2006)
- McElree, Brian, "Serial retrieval processes in the recovery of order information", Journal of Experimental Psychology : General, 122(3), pp.291~315 (1993)
- McElree, Brian, "Memory structures that subserve sentence comprehension", Journal of Memory and Language, 48(1), pp.67~91 (2003)
- Nicol, Janet, "The role of structure in coreference assignment during sentence comprehension", Journal of Psycholinguistic Research, 18(1), pp.5~19 (1989)
- Parker, Dan, "Reflexive attraction in comprehension is selective", Journal of Memory and Language, 94, pp.272~290 (2017)
- Parker, Dan, "Cue combinatorics in memory retrieval for anaphora", Cognitive Science, 43(3), pp.e12715~ (2019)
- Patil, Umesh, "Retrieval interference in syntactic processing: The case of reflexive binding in English", Frontiers in Psychology, 7, pp.329~ (2016)
- Pearl, Judea, "The book of why: The new science of cause and effect" (2018), Basic Books
- Pearlmutter, Neal J., "Agreement processes in sentence comprehension", Journal of Memory and Language, 41(3), pp.427~456 (1999)
- Perconti, Pietro, "Deep learning and cognitive science", Cognition, 203, pp.104365~ (2020)
- Phillips, Colin, "Experiments at the interfaces", pp.147~180 (2011), Emerald Group Publishing Limited
- Roy, Aurko, "Efficient content-based sparse attention with routing transformers", Transactions of the Association for Computational Linguistics, 9, pp.53~68 (2021)
- Sturt, Patrick, "The time-course of the application of binding constraints in reference resolution", Journal of Memory and Language, 48(3), pp.542~562 (2003)
- Tanner, Darren, "The time-course of feature interference in agreement comprehension: Multiple mechanisms and asymmetrical attraction", Journal of Memory and Language, 76, pp.195~215 (2014)
- Tenney, Ian, "What do you learn from context? probing for sentence structure in contextualized word representations", Presented at 2019 International Conference on Learning Representations (ICLR 2019) Debugging Machine Learning Models Workshop (2019)
- Tucker, Matthew A., "Representing number in the real-time processing of agreement: Self-paced reading evidence from Arabic", Frontiers in Psychology, 6, pp.347~ (2015)
- Van Dyke, Julie A, "Interference effects from grammatically unavailable constituents during sentence processing", Journal of Experimental Psychology : Learning, Memory, and Cognition, 33(2), pp.407~430 (2007)
- Van Dyke, Julie A., "Distinguishing effects of structure and decay on attachment and repair: A cue-based parsing account of recovery from misanalyzed ambiguities", Journal of Memory and Language, 49(3), pp.285~316 (2003)
- Van Dyke, Julie A., "Retrieval interference in sentence comprehension", Journal of Memory and Language, 55(2), pp.157~166 (2006)
- Van Dyke, Julie A., "Cue-dependent interference in comprehension", Journal of Memory and Language, 65(3), pp.247~263 (2011)
- Vaswani, Ashish, "Attention is all you need", Proceedings of the Neural Information Processing Systems(NIPS), pp.5999~6009 (2017)
- Vig, Jesse, "BertViz: A tool for visualizing multihead self-attention in the BERT model", Presented at 2019 International Conference on Learning Representations (ICLR 2019) Debugging Machine Learning Models Workshop (2019)
- Wagers, Matthew W., "Agreement attraction in comprehension: Representations and processes", Journal of Memory and Language, 61(2), pp.206~237 (2009)
- Wolf, Thomas, "Transformers: State-of-the-art natural language processing", Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pp.38~45 (2020)
- Xiang, Ming, "Illusory licensing effects across dependency types: ERP evidence", Brain and Language, 108(1), pp.40~55 (2009)