As a natural language understanding task, commonsense question answering (CQA) is significantly more challenging than conventional question answering tasks. It requires the model to possess stronger commonsense reasoning capabilities. Currently, unsupervised methods for CQA have achieved relatively good performance on several datasets, but these approaches struggle to adequately mine and utilize commonsense knowledge, limiting the model’s reasoning ability in complex scenarios. To address this issue, this paper proposed a novel unsupervised CQA method, whose core advantage lay in effectively integrating external commonsense knowledge through unsupervised learning, thereby enhancing the model’s generalization capability and reasoning depth. Firstly, the method classifies questions into scientific commonsense questions and everyday event questions. Then, it generates corresponding knowledge prefixes based on the question type. Next, these knowledge prefixes are input into a pre-trained language model to produce multi-granularities commonsense knowledge through large model prompts. Finally, the multi-grained knowledge is leveraged to assist the answer generation module in reasoning. The adoption of an unsupervised approach not only reduces the reliance on annotated data but also better adapts to diverse commonsense scenarios, demonstrating its flexibility and generalizability in practical applications. Experimental results show that the proposed method significantly outperforms baseline models on relevant datasets, validating its correctness and rationality in unsupervised CQA tasks.
如表 3 所示, 疑问句“What is the effect of this?”和“What is the cause of this?”对应的陈述形式分别为“As a result, ”和“Because”。拼接而成的语句分别为“The grape juice fermented.As a result,the juice turned to wine”和“The man got a discount on his groceries.Because he used a coupon”。这两个句子分别为科学常识和日常事件。
WANGW Q, FANGT Q, DINGW X, et al. CAR: conceptualization-augmented reasoner for zero-shot commonsense question answering[C]//Conference on Empirical Methods in Natural Language Processing,2023:13520-13545.
[2]
HEJ, GUTIÉRREZ-BASULTOV, PANJ Z. BUCA: A binary classification approach to unsupervi- sed commonsense question answering[C]//61st Annual Meeting of the Association for Computational Linguistics,2023:376-387.
[3]
SHIH, WANGW, FANGT, et al. QADYNAMICS: Training dynamics-driven synthetic QA diagnostic for zero-shot commonsense question answering[C]//Findings of the Association for Computational Linguistics: EMNLP 2023, 2023: 15329-15341.
[4]
ZHAOZ, HUL, ZHAOH, et al. Knowledgeable parameter efficient tuning network for commonsense question answering[C]//Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics,2023:9051-9063.
[5]
ZHANGL, LIR. KE-GCL: Knowledge enhanced graph contrastive learning for commonsense question answering[C]//Findings of the Association for Comp- utational Linguistics: EMNLP 2022, 2022:76-87.
[6]
WANGY, ZHANGH, LIANGJ, et al. Dynamic heterogeneous-graph reasoning with language models and knowledge representation learning for commonse- nse question answering[C]//61st Annual Meeting of the Association for Computational Linguistics,2023: 14048-14063.
[7]
SUNY, SHIQ, QIL, et al. JointLK: Joint reasoning with language models and knowledge graphs for commonsense question answering[C]//2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,2022:5049-5060.
[8]
LIUJ, LIUA, LUX, et al. Generated knowledge prompting for commonsense reasoning[DB/OL]. (2021-10-15)[2024-07-20].
[9]
SUNY, ZHANGY, QIL, et al. TSGP: Two-stage generative prompting for unsupervised commonsense question answering[C]//Findings of the Association for Computational Linguistics: EMNLP 2022:968-980.
[10]
CHENQ, XUG, YANM, et al. Distinguish before answer: Generating contrastive explanation as knowledge for commonsense question answering[C]//Findings of the Association for Computational Lingui- stics: ACL 2023,2023:13207-13224.
[11]
WANGC, CAOP, LIJ, et al. Leros: Learning explicit reasoning on synthesized data for commonse- nse question answering[C]//2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), 2024:10303-10315.
[12]
LIUJ, HALLINANS, LUX, et al. Rainier: Reinforced knowledge introspector for commonsense question answering[C]//Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing,2022:8938-8958.
[13]
NIUY L, HUANGF, LIANGJ M, et al. A semantic-based method for unsupervised commonsen- se question answering[C]//59th Annual Meeting of the Association for Computational Linguistics,2021: 3037-3049.
[14]
SHWARTZV, WESTP, LE BRASR, et al. Unsupervised commonsense question answering with self-talk[C]//2020 Conference on Empirical Methods in Natural Language Processing (EMNLP),2020:4615-4629.
[15]
WANGJ, ZHAOH. ArT: All-round thinker for unsupervised commonsense question answering[C]//29th International Conference on Computational Linguistics,2022:1490-1501.
[16]
CHOMSKYN, SCHÜTZENBERGERM P. The algebraic theory of context-free languages[M]. Amsterdam: Elsevier, 1959.
[17]
BIRDS. NLTK: The natural language toolkit[C]// Proceedings of the COLING/ACL 2006 Interactive Presentation Sessions, 2006:69-72.
[18]
PETERSM E, NEUMANNM, IYYERM, et al. Deep contextualized word representations[C]//2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2018: 2227-2237.
[19]
HOLTZMANA, BUYSJ, DUL, et al. The curious case of neural text degeneration[C]//International Conference on Learning Representations 2020,2020:1-16.
[20]
ROEMMELEM, BEJANC A, GORDONA S. Choice of plausible alternatives: An evaluation of commonsense causal reasoning[C]//AAAI Spring Symposium: Logical Formalizations of Commonsense Reasoning, 2011: 90-95.
[21]
SAP M, RASHKINH, CHEND, et al. SocialIQa: Commonsense reasoning about social interactions[C]//2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), 2019: 4463-4473.
[22]
MOSTAFAZADEHN, CHAMBERSN, HEX, et al. A corpus and cloze evaluation for deeper understanding of commonsense stories[C]//2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language,2016: 839-849.
[23]
RADFORDA, WUJ, CHILDR, et al. Language models are unsupervised multitask learners[EB/OL]. (2019-02- 14) [2024-07-17].