GPT를 활용한 한국어 반어 병렬코퍼스 자동 구축
Automated Construction of a Korean Irony Parallel Corpus Using GPT
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
본 연구는 GPT 모델을 활용한 한국어 반어 병렬 코퍼스의 자동 구축을 목표로 한다. 반어 병렬 코퍼스는 직설 문장과 같은 의미를 담고 있는 반어 문장 쌍으로 구성되는데 이 직설 문장은 일반 코퍼스에서 추출된 부정적인 서술어를 담고 있는 문장이다. 직설 문장에서 사용된 부정적인 서술어를 감성 반의어를 이용하여 대응되는 긍정 서술어로 변환하여 반어 문장을 생성한다. 생성된 반어 문장이 명확한 반어적 의미를 전달하지 못할 경우, 관련된 맥락을 제공하여 반어 의미를 명확히 전달하고자 한다. GPT 모델을 이용하여 직설 문장이 도출할 수 있는 부정적 개념을 추출하고 이를 기반으로 코퍼스 검색과 GPT를 이용한 문장 생성을 통해 맥락 지원 문장을 생성한다. 구축된 반어 병렬 코퍼스의 수동 평가 결과 생성된 문장들의 문법적 정확성은 수용할 만한 수준인 반면, 반어적 표현력은 다소 개선이 필요하였다. 본 연구는 한국어 반어 병렬 코퍼스 자동 구축의 최초 시도이며, 다양한 표현으로 만들어진 여러 유형의 문장을 포함할 수 있었다.
Abstract
This research presents a Korean irony parallel corpus which was automatically constructed utilizing a GPT model. The corpus is composed of a set of paired sentences including indicative and ironic ones. All the indicative sentences are extracted from a general purposed corpus and convey negative sentiment. They are used to generate a corresponding ironic sentence, using a dictionary of sentiment reversal which contains paired sentiment antonyms. When a generated ironic sentence does not convey clear ironic meaning, relevant context is provided in a form of a supporting sentence. The sentences are retrieved through searching a corpus and generated utilizing a GPT model based on semantic incongruity. Human evaluation suggests grammaticality of the generated sentences are acceptable while sarcasticness need to be improved. Automatic evaluation shows the corpus includes various types of sentences created various expressions.
참고문헌 (15)
- Han Kil, "How to Realize Rhetorical Irony in Korean", STUDIES IN HUMANITIES, 13, pp.1~35 (2005)
- Lee Kong Joo, "Korean Irony Corpus Construction", LANGUAGE AND INFORMATION, 27(1), pp.19~36 (2023)
- Tuhin Chakrabarty, "Rˆ3 : Reverse, Retrieve, and Rank for Sarcasm Generation with Commonsense Knowledge", Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pp.7976~7986 (2020), Association for Computational Linguistics
- Andrea Esuli, "Sentiwordnet : A publicly available lexical resource for opinion mining", LREC, 6, pp.417~422 (2006)
- George AMiller, "Wordnet : a lexical database for English", Communications of the ACM, 38(11), pp.39~41 (1995)
- SamuelBowman, "A large annotated corpus for learning natural language inference", Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pp.632~642 (2015)
- Debanjan Ghosh, "Interpreting verbal irony : Linguistic strategies and the connection to the type of semantic incongruity", Proceedings of the Society for Computation in Linguistics, 3(9), pp.76~87 (2020)
- Lotem Peled, "Sarcasm SIGN : Interpreting sarcasm with sentiment based monolingual machine translation", Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, 1, pp.1690~1700 (2017), Association for Computational Linguistics
- Abhijit Mishra, "A modular architecture for unsupervised sarcasm generation", 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.6146~6155 (2019)
- Salvatore Attardo, "Irony markers and functions : Towards a goal-oriented theory of irony and its processing", Rask, 12(1), pp.3~20 (2000)
- Christian Burgers, "Verbal irony differences in usage across written genres", Journal of Language and Social Psychology, 31(3), pp.290~310 (2012)
- Li, Jiwei, "A diversitypromoting objective function for neural conversation models", Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp.110~119 (2016)
- Papineni, Kishore, "BLEU : a method for automatic evaluation of machine translation", Proceedings of the 40th annual meeting of the Association for Computational Linguistics, pp.311~318 (2002)
- Zhu, Yizhe, "Texygen : A benchmarking platform for text generation models", Proceedings of the 41st International ACM SIGIR Conference on Research & Development in Information Retrieval, pp.1097~1100 (2018)