Corpora and Baselines for Humour Recognition in Portuguese

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 1278–1285 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 1278 Corpora and Baselines for Humour Recognition in Portuguese Hugo Gonçalo Oliveira1,2, André Clemêncio1,2, Ana Alves1,3 1CISUC, University of Coimbra 2DEI, University of Coimbra 3ISEC, Polytechnic Institute of Coimbra hroliv@dei.uc.pt, adfc@student.dei.uc.pt, ana@dei.uc.pt Abstract Having in mind the lack of work on the automatic recognition of verbal humour in Portuguese, a topic connected with fluency in a natural language, we describe the creation of three corpora, covering two styles of humour and four sources of non-humorous text, that may be used for related studies. We then report on some experiments where the created corpora were used for training and testing computational models that exploit content and linguistic features for humour recognition. Obtained results helped us taking some conclusions about this challenge and may be seen as baselines for those willing to tackle it in the future, using the same corpora. Keywords: corpora creation, humour recognition, processing of Portuguese, text classification, feature extraction 1. Introduction Computational Humour (Binsted et al., 2006) is a field of Artificial Intelligence that uses computers for detecting, analysing or producing humour. The automatic recognition of verbal humour is a branch of this field, with obvious connections with Natural Language Processing, because recognising humour expressed in a language, is a sign of fluency on that language (Tagnin, 2005). This means that an artificial agent that uses natural language for commu- nication purposes should be able to recognise humour for better handling different situations. For instance, a news aggregator should have the ability of filtering out humorous news; or a chatbot should recognise humorous interactions and change its response, possibly ignoring it, or answering with generated humour as well. Despite some work for other languages, especially English, humour recognition has not been an area of interest for Portuguese, until recently. While developing a computa- tional model for humour recognition in Portuguese, we de- cided to tackle two styles of humour and gathered some texts for validation purposes (Clemêncio et al., 2019). The current paper presents three new corpora created out of those texts, but including additional styles of negative ex- amples (why-questions and proverbs), and with balancing concerns, not only between the classes of humour and non- humour, but also between different sources. For instance, the corpus covering more sources has the positive examples balanced between two styles of humour (one-liners and hu- morous headlines) and the negative between four styles of non-humour (general knowledge questions, why-questions, proverbs, news headlines). We further describe three new experiments where we tackle humour recognition as a supervised learning task, using the created corpora for validation, training and evaluation. Again, we went further than in previous work and present results for the best methods in the new corpora, after a more exhaustive search for the best parameters with 10-fold cross validation, and tests in unseen evaluation data, also exploring some additional features. This helped us take …

Content truncated to 500 words