ESSLLI2016 DTS Lecture Day 5-1: Introduction to day 5
0 likes38 views
The document discusses the need for computational semantics to compare and evaluate formal semantic frameworks as they have become increasingly complex. It emphasizes the importance of developing flexible computational architectures and datasets to benchmark the performance of these semantic systems. Additionally, it highlights two key issues in implementing formal semantics systems for natural language inference: creating a flexible platform and integrating external ontological knowledge.
ESSLLI2016 DTS Lecture Day 5-1: Introduction to day 5
- 1. Introduction References Introduction to Dependent Type Semantics: Day 5 – Introduction Daisuke Bekki1,2 Koji Mineshima1,2 1Ochanomizu University / 2CREST, Japan Science and Technology Agency version: August 19, 2016 ESSLLI 2016 course, Bolzano, Italy, August 15-19, 2016. 1 / 10
- 3. Introduction References Why computational semantics? ▶ As formal semantics advanced, it became more and more difficult to compare and evaluate the semantic frameworks and analyses proposed. 3 / 10
- 4. Introduction References Why computational semantics? ▶ As formal semantics advanced, it became more and more difficult to compare and evaluate the semantic frameworks and analyses proposed. ▶ It is often not easy to figure out the potential capacity of each system. • What prediction does this system provide for a range of constructions, including those that are not explicitly mentioned in a paper? 3 / 10
- 5. Introduction References Why computational semantics? ▶ As formal semantics advanced, it became more and more difficult to compare and evaluate the semantic frameworks and analyses proposed. ▶ It is often not easy to figure out the potential capacity of each system. • What prediction does this system provide for a range of constructions, including those that are not explicitly mentioned in a paper? ▶ Usually, one paper focuses on a set of specific linguistic phenomena (”Let’s neglect tense”). • Does this theory provide a correct prediction for a case in which phenomena X interacts with phenomena Y? 3 / 10
- 6. Introduction References From formal semantics to computational semantics ▶ So, computational modeling is needed to specify detailed semantic architectures, which are often left implicit or neglected in a published paper. 4 / 10
- 7. Introduction References From formal semantics to computational semantics ▶ So, computational modeling is needed to specify detailed semantic architectures, which are often left implicit or neglected in a published paper. ▶ It would make it possible to compute the predictions of each system quickly and precisely. 4 / 10
- 8. Introduction References From formal semantics to computational semantics ▶ So, computational modeling is needed to specify detailed semantic architectures, which are often left implicit or neglected in a published paper. ▶ It would make it possible to compute the predictions of each system quickly and precisely. ▶ A necessary step towards establishing a meaningful and systematic way to compare and evaluate each semantic framework. 4 / 10
- 9. Introduction References From formal semantics to computational semantics ... flexible computational architectures which make it possible to experiment with semantic representations, semantic construction strategies, and inference, must be designed and implemented. Blackburn and Bos (2005) 5 / 10
- 10. Introduction References From formal semantics to computational semantics ▶ To test the capacity of formal semantics systems, we would also need a dataset (“benchmark”) that consists of a comprehensive list of linguistic problems known to the community. ▶ For further discussion on datasets and the evaluation of semantic theories, see the CFP of the Workshop: Unshared Task at LENLS 13 http://www.compling.jp/fracas task/index.html 6 / 10
- 11. Introduction References Implemented “formal semantics” systems Textbooks/education tools ▶ Blackburn and Bos (2005): Prolog implementation ▶ Champollion et al. (2007): Lambda Calculator ▶ Bird et al. (2009): NLTK, Python ▶ van Eijck and Unger (2010): Haskell implementation With wide-coverage parsers ▶ Bos et al. (2004): Boxer (CCG + DRT), English ▶ Moot (2010): TLG + DRT, French ▶ Butler and Yoshimoto (2012): SCT + Treebank Semantics, English and Japanese ▶ Abzianidze (2015): CCG + Natural Logic Tableau prover, English ▶ Mineshima et al. (2015): CCG + HOL, English and Japanese 7 / 10
- 12. Introduction References From proof system to computational NLI system Two issues: 1. Flexible platform to implement a “formal semantics” system and a prover for natural language inference (NLI) 2. External/ontological knowledge in signature: how to acquire and use it? 8 / 10
- 13. Introduction References From proof system to computational NLI system Two issues: 1. Flexible platform to implement a “formal semantics” system and a prover for natural language inference (NLI) 2. External/ontological knowledge in signature: how to acquire and use it? Two steps towards the implementation of DTS: 1. Build a platform (parsers/prover) for formal semantics 2. Extend it with underspecification semantics (@-terms) 8 / 10
- 14. Introduction References Reference I Abzianidze, L. (2015) “A Tableau Prover for Natural Logic and Language”, In the Proceedings of Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing. Lisbon, Portugal, pp.2492–2502, Association for Computational Linguistics. Bird, S., E. Klein, and E. Loper. (2009) Natural Language Processing with Python. O’Reilly Media, Inc. Blackburn, P. and J. Bos. (2005) Representation and Inference for Natural Language: A First Course in Computational Semantics. Stanford, CA, CSLI Publications. Bos, J., S. Clark, M. Steedman, J. R. Curran, and J. Hockenmaier. (2004) “Wide-coverage semantic representations from a CCG parser”, In the Proceedings of Proceedings of the 20th international conference on Computational Linguistics. pp.1240–1246. Butler, A. and K. Yoshimoto. (2012) “Banking meaning representations from treebanks”, Linguistic Issues in Language Technology 7(1). 9 / 10
- 15. Introduction References Reference II Champollion, L., J. Tauberer, and M. Romero. (2007) “The Penn Lambda Calculator: Pedagogical Software for Natural Language Semantics”, In the Proceedings of T. H. King (ed.): Proceedings of the Grammar Engineering Across Frameworks (GEAF0 7) Workshop. Stanford, pp.106–127, CSLI Publications. Mineshima, K., P. Mart´ınez-G´omez, Y. Miyao, and D. Bekki. (2015) “Higher-order logical inference with compositional semantics”, In the Proceedings of Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing. Lisbon, Portugal, pp.2055–2061, Association for Computational Linguistics. Moot, R. (2010) “Wide-coverage French syntax and semantics using Grail”, In the Proceedings of TALN 2010. van Eijck, J. and C. Unger. (2010) Computational Semantics with Functional Programming. Cambridge University Press. 10 / 10