Pronoun Resolution in Turkish with Deep Learning

Abstract

In language, to avoid repetitive use of a word/phrase, it is common to use pronouns that refer to the corresponding antecedent word/phrase. Matching a pronoun with the antecedent to which it refers is called pronoun resolution. This study evaluates the performance of deep learning methods in pronoun resolution in Turkish texts. Within the scope of the study, a dataset was compiled using 10 Turkish children's stories and 12 features were identified to be used in the experiments. Multilayer perceptron, convolutional and recurrent neural networks were tested in several different configurations with varying numbers of neurons and layers and the performance was measured with F1 score. The results show that the highest performance in Turkish pronoun resolution is achieved by a multilayer perceptron neural network with a medium number of layers using too many neurons.

Keywords

• Natural language processing
• Pronoun resolution
• Convolutional neural networks
• Multilayer Perceptron
• Recurrent neural networks

Derin Öğrenme ile Türkçede Adıl Çözümleme

Abstract

Dilde, bir sözcüğün/sözcük öbeğinin sürekli tekrar eden kullanımını önlemek için, ilgili öncül sözcüğe/sözcük öbeğine atıfta bulunan adılların kullanımına sık rastlanır. Bir adılın atıfta bulunduğu öncül ile eşleştirilmesi adıl çözümleme olarak adlandırılır. Bu çalışmada Türkçe metinlerde adılların çözümlenmesinde derin öğrenme yöntemlerinin başarımı değerlendirilmiştir. Çalışma kapsamında 10 Türkçe çocuk hikayesi kullanılarak bir veri kümesi derlenmiş, deneylerde kullanılmak üzere 12 öznitelik belirlenmiştir. Çok katmanlı algılayıcı, evrişimsel (konvolüsyonel) ve tekrarlayan sinir ağları nöron ve katman sayılarının değiştiği bir dizi farklı konfigürasyonla uygulanarak F1 ölçüsü ile başarım ölçülmüştür. Sonuçlar, Türkçe adıl çözümlemesinde en yüksek başarımın, çok fazla nöron kullanan orta sayıda katmana sahip çok katmanlı algılayıcı sinir ağı tarafından elde edildiğini göstermiştir.

Keywords

• Doğal dil işleme
• Adıl çözümleme
• Evrişimsel sinir ağları
• Çok katmanlı algılayıcı
• Tekrarlayan sinir ağları

References

1. Grosz B. J., Weinstein S., Joshi A.K., Centering: A Framework for Modelling the Local Coherence of Discourse Centering: A Framework for Modelling the Local Coherence of Discourse, Computational Linguistics, 1995, 21: 203–225.
2. Halliday M.A., Hasan R., Cohesion in English 1st ed., Routledge,https://www.routledge.com/Cohesion-in-English/Halliday-Hasan/p/book/9780582550414 UK, 1976, 1-392.
3. Van Valin R.D., LaPolla R.J., Syntax: Structure, Meaning, and Function 1st ed., Cambridge University Press, UK, 1997, 1-744.
4. Van Valin R.D., A Summary of Role and reference Grammar, Papers of the Summer Institute of Linguistics, University of North Dakota Session, 2005, 37(5):1–30.
5. Smeaton A.F., Progress in the application of natural language processing to information retrieval tasks, Computing Journal, 1992, 35(268):268-278.
6. Yıldırım S., Kılıcaslan Y., A Machine Learning Approach to Personal Pronoun Resolution in Turkish, FLAIRS Conference, USA:Florida, 2007, 269–270.
7. Kılıcaslan Y., Guner E.S., Yıldırım S., Learning-based pronoun resolution for Turkish with a comparative evaluation, Computers Speech and Language, 2009, 23: 311–331.
8. Kornfilt J., Turkish 1st ed., Routledge, UK, 1997, 1-608.

9. Hamann J., Hamann J., On the Syntax and Morphology of Double Agreement in Lavukaleve, In agreement to Sebastian Bank, Doreen Georgi & Jochen Trommer (eds.) Linguistische Arbeits Berichte, 2010, 197–225.
10. Tufekci P., Kılıcaslan Y., A Computational Model for Resolving Pronominal Anaphora in Turkish Using Hobbs’ Naïve Algorithm, Journal of Computing and Information Science in Engineering, 2007, 1: 854–858.
11. Hobbs J.R., Resolving pronoun references, Lingua, 1978, 44: 311–338.
12. Lappin S., Leass H.J., An Algorithm for Pronominal Anaphora Resolution, Computational Linguistics, 1994, 20: 535–561.
13. Kennedy C., Boguraev B., Anaphora for Everyone : Pronominal Anaphora Resolution without a Parser, Proceedings of the 16th International Conference on Computational Linguistics, 1996, 113–118.
14. Aone C., Bennett S., Evaluating automated and manual acquisition of anaphora resolution strategies, Procedings of 33rd Annual Meeting of the Association for Computational Linguistics, Massachusetts, 1995, 122–129.
15. McCarthy J.F., Lehnert W.G., Using Decision Trees for Coreference Resolution, Proceedings of Fourteenth International Joint Conference of Artificial Intelligence, Montreal, Quebec, Canada, 1995, 1–5.
16. Soon W.M., Ng H.T., Lim D.C.Y., A Machine Learning Approach to Coreference Resolution of Noun Phrases, Computational Linguistics, 2001, 27: 521–544.
17. Mitkov R., Evans R., Orasan C., A New, Fully Automatic Version of Mitkov’s Knowledge-Poor Pronoun Resolution Method, Proceedings of CICLing2002, Mexico City, Mexico, 2002, 2276: 168–186.
18. Preiss J., Choosing a Parser for Anaphora Resolution, Proceedings of 4th Discourse Anaphora Anaphora Resolutition Colloq. (DAARC 2002), Lisbon, 2002, 175–180.
19. Turan U.D., Null vs. Overt Subjects in Turkish Discourse: A Centering Analysis, Phd. Thesis, University of Pennsylvania, Pennsylvania, 1995, 1-27.
20. Yuksel O., Bozsahin C., Contextually appropriate reference generation, 2002, Natural Language Engineering, 8: 69–89.
21. Chomsky N., on Binding, Linguist Inquires; 1980, 11: 1–46.
22. Erguvanlı-Taylan E., Pronominal versus zero representation of anaphora in Turkish, Studies in Turkish Linguistics, 1986, 209–231.
23. Tın E., Akman V., Situated Processing of Pronominal Anaphora, Verarbeitung Natürlicher Spr, 1998, 369–378.
24. Elizondo D., The Linear Separability Problem: Some Testing Methods, IEEE Transactions Neural Network, 2006, 17: 330–344.
25. Zipf, G. K. Human behavior and the principle of least effort, Cambridge: Addison Wesley Press, 1949.
26. Pinkus A., Approximation theory of the MLP model in neural networks, Acta Numerica, 1999, 8(143): 143-195.
27. Zurada, J.M., Introduction to Artificial Neural Systems 1st ed., West Group, Minnesota:St. Paul,1992, 1-759.
28. Oord A.V.D., Dieleman S., Schrauwen B., Deep content-based music recommendation, Advances in Neural Information Processing Systems 26 (NIPS 2013), Nevada, 2013, 643-2651.
29. Sermanet P., Eigen D., Zhang X., et al., OverFeat: Integrated Recognition, Localization and Detection using Convolutional Networks, 2nd International Conference on Learning Representations (ICLR 2014), Banff, Canada, 2014, 1-17.
30. Marhon, S.A., Cameron, C.J.F., Kremer, S.C., Recurrent Neural Networks, Handbook on Neural Information Processing. Intelligent Systems Reference Library, Springer, Berlin, Heidelberg, 2013, 49:29-65.
31. Kısla T., Karaoglan B., A hybrid Statistical Approach to Stemming in Turkish: An Agglutinative Language, Anadolu University-Journal of Science and Technology, 2016, 17: 401–412.
32. Abadi, M., Agarwal, A., Barham, P., et.al., Tensorflow: large-scale machine learning on heterogeneous distributed systems, Arxiv:1603.04467, 2016.
33. https://deeplearning4j.konduit.ai. Access date: 29.12.2021.
34. Ruck D.W., Rogers S.K., Kabrisky M., et al., Letters: The Multilayer Perceptron as an Approximation to a Bayes Optimal Discriminant Function, IEEE Transactions on Neural Networks, 1990, 1: 296–298.
35. Glorot X., Bengio Y., Understanding the difficulty of training deep feedforward neural networks, Proceedings of 13th International Conference on Artificial Intelligence and Statistics (AISTATS), Italy, 2010, 9: 249–256.
36. Krizhevsky A., Sutskever I., Geoffrey E.H., ImageNet Classification with Deep Convolutional Neural Networks, Advances in Neural Information Processing Systems, 2012, 25:1–9.