Negative Polarity Items in English: A Deep Learning Model and Statistical Analysis
1
조회
0
다운로드
2
인용
0
참고문헌
키워드
초록
This paper examined English negative polarity items (NPIs) with a deep learning model and statistical analysis. Warstadt et al. (2019a) developed a deep learning model with the Bidirectional Encoder Representations from Transformers (BERT), and the study analyzed English NPIs. Their dataset included a total of 400,000 sentences, and they encoded a few linguistic variables per each sentence. This paper took the same dataset in Warstadt et al. (2019a), but one more factor (Monotonicity) was included. As for the deep learning model, this paper took the BERTLARGE model in Lee (2021), where the syntactic acceptability was calculated with numeric scores (0~100), rather than measured by the binary classification. After the acceptability scores were calculated per each sentence, the scores were converted with z-scores, and statistical analysis was conducted with six linguistic factors and interactions of three factors. Through the analysis, the followings were observed: (i) an NPI had to exist within the scope of the licensor, (ii) downward entailment also played a significant role in the NPI licensing, and (iii) the interaction between NPI licensor and scope played more crucial roles in the NPI licensing.
Abstract
This paper examined English negative polarity items (NPIs) with a deep learning model and statistical analysis. Warstadt et al. (2019a) developed a deep learning model with the Bidirectional Encoder Representations from Transformers (BERT), and the study analyzed English NPIs. Their dataset included a total of 400,000 sentences, and they encoded a few linguistic variables per each sentence. This paper took the same dataset in Warstadt et al. (2019a), but one more factor (Monotonicity) was included. As for the deep learning model, this paper took the BERTLARGE model in Lee (2021), where the syntactic acceptability was calculated with numeric scores (0~100), rather than measured by the binary classification. After the acceptability scores were calculated per each sentence, the scores were converted with z-scores, and statistical analysis was conducted with six linguistic factors and interactions of three factors. Through the analysis, the followings were observed: (i) an NPI had to exist within the scope of the licensor, (ii) downward entailment also played a significant role in the NPI licensing, and (iii) the interaction between NPI licensor and scope played more crucial roles in the NPI licensing.
참고문헌 (36)
- Annamoradnejad, Issa, "ColBERT : Using BERT Sentence Embedding for Humor Detection", Computation and Language (2020)[DOI]
- Bard, Ellen, "Magnitude Estimation of Linguistic Acceptability", Language, 72, pp.32~68 (1996)
- Charniak, Eugene, "BLLIP 1987-89 WSJ Corpus Release 1 LDC2000T43" (2000), Linguistic Data Consortium
- Chierchia, Gennaro, "Logic in Grammar : Polarity, Free Choice, and Intervention" (2013), Oxford University Press
- Cowart, Wayne, "Experimental Syntax: Applying Objective Methods to Sentence Judgments" (1997), Sage Publications
- Devlin, Jacob, "BERT:Pre-training of Deep Bidirectional Transformers for Language Understanding", Computation and Language, pp.04805~ (2019)[DOI]
- Fauconnier, Gilles, "Polarity and the Scale Principle", Proceedings of the Chicago Linguistics Society, 11, pp.188~199 (1975)
- Giannakidou, Anastasia, "Polarity Sensitivity as (Non)Veridical Dependency" (1998), John Benjamins Publishing
- Goldberg, Yoav, "Assessing BERT’s Syntactic Abilities", Computation and Language, pp.05287~ (2019)[DOI]
- Goodfellow, Ian, "Deep Learning" (2016), MIT Press
- Gries, Stefan, "Statistics for Linguistics with R: A Practical Introduction" (2021), Mouton
- Hu, Jennifer, "A Closer Look at the Performance of Neural Language Models on Reflexive Anaphor Licensing", Proceedings of the Society for Computation in Linguistics, pp.323~333 (2020)
- Hu, Jennifer, "A Systematic Assessment of Syntactic Generalization in Neural Language Models", Computation and Language, pp.03692~ (2020)[DOI]
- Jumelet, Japp, "Language Models Use Monotonicity to Assess NPI Licensing", Computation and Language, pp.13818~ (2021)[DOI]
- Kadmon, Nirit, "Any", Linguistics and Philosophy, 16(4), pp.353~422 (1993)
- Keller, Frank, "Gradient in Grammar : Experimental and Computational Aspects of Degrees of Grammaticality", Doctor, pp.2000
- Ladusaw, Willaim, "Negative Polarity Items as Inherent Scope Relations", Doctor, pp.1979
- Lee, Yong-hun, "Corpus Linguistics and Statistics Using R" (2016), Hankuk Publishing Co
- 이용훈, "English Island Constraints Revisited: Experimental vs. Deep Learning Approach", 영어학연구, 27(3), pp.21~45 (2021), 한국영어학학회[DOI]
- Levy, Roger, "Expectation-based Syntactic Comprehension", Cognition, 106(3), pp.1126~1177 (2008)
- Linebarger, Marcia, "The Grammar of Negative Polarity", Doctor, pp.1980
- Marvin, Rebecca, "Targeted Syntactic Evaluation of Language Models", Computation and Language, pp.09031~ (2018)[DOI]
- Kwonsik Park, "Deep learning can contrast the minimal pairs of syntactic data", 언어연구, 38(2), pp.395~424 (2021), 언어정보연구소[DOI]
- R Core Team, "R: A Language and Environment for Statistical Computing" (2021), R Foundation for Statistical Computing
- Saeed, John, "Semantics" (2016), Wiley Blackwell
- Schütze, Carson, "The Empirical Base of Linguistics: Grammaticality Judgments and Linguistic Methodology" (1996), University of Chicago Press
- Sprouse, Jon, "Magnitude Estimation and the Non-Linearity of Acceptability Judgments", Proceedings of the 27th West Coast Conference on Formal Linguistics, pp.397~403 (2008), Cascadilla Proceedings Project
- Vaswani, Ashish, "Attention Is All You Need", Computation and Language (2017)[DOI]
- Wang, Alex, "GLUE: A Multi-task Benchmark and Analysis Platform for Natural Language Understanding", Computation and Language (2019)[DOI]
- Wang, Alex, "SuperGLUE : A Stickier Benchmark for General-purpose Language Understanding Systems", Computation and Language (2020)[DOI]
- Warstadt, A., "Investigating BERT’s Knowledge of Language : Five Analysis Methods with NPIs", Computation and Language (2019)[DOI]
- Warstadt, Alex, "Neural Network Acceptability Judgments", Computation and Language (2019)[DOI]
- Wilcox, Ethan, "What Syntactic Structures Block Dependencies in RNN Language Models?", Computation and Language (1905)[DOI]
- Wilcox, Ethan, "Hierarchical Representation in Neural Language Models: Suppression and Recovery of Expectations", Computation and Language (2019)[DOI]
- Wilcox, Ethan, "What Do RNN Language Models Learn about Filler-Gap Dependencies?", Computation and Language (2018)[DOI]
- Zhu, Yukun, "Aligning Books and Movies:Towards Story-like Visual Explanations by Watching Movies and Reading Books", Proceedings of the IEEE International Conference on Computer Vision, pp.19~27 (2015)