Do not let the history haunt you: Mitigating Compounding Errors in Conversational Question Answering

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 189–194 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 189 Design of BCCWJ-EEG: Balanced Corpus with Human Electroencephalography Yohei Oseki†,‡, Masayuki Asahara‡ † Waseda University, 3-4-1 Okubo, Shinjuku, Tokyo 169-8555 ‡ National Institute for Japanese Language and Linguistics, 10-2 Midoricho, Tachikawa, Tokyo 190-8561 oseki@aoni.waseda.jp, masayu-a@ninjal.ac.jp Abstract The past decade has witnessed the happy marriage between natural language processing (NLP) and the cognitive science of language. Moreover, given the historical relationship between biological and artificial neural networks, the advent of deep learning has re-sparked strong interests in the fusion of NLP and the neuroscience of language. Importantly, this inter-fertilization between NLP, on one hand, and the cognitive (neuro)science of language, on the other, has been driven by the language resources annotated with human language processing data. However, there remain several limitations with those language resources on annotations, genres, languages, etc. In this paper, we describe the design of a novel language resource called BCCWJ-EEG, the Balanced Corpus of Contemporary Written Japanese (BCCWJ) experimentally annotated with human electroencephalography (EEG). Specifically, after extensively reviewing the language resources currently available in the literature with special focus on eye-tracking and EEG, we summarize the details concerning (i) participants, (ii) stimuli, (iii) procedure, (iv) data preprocessing, (v) corpus evaluation, (vi) resource re- lease, and (vii) compilation schedule. In addition, potential applications of BCCWJ-EEG to neuroscience and NLP will also be discussed. Keywords: Balanced Corpus of Contemporary Written Japanese, Electroencephalography, Cognitive Modeling 1. Introduction The past decade has witnessed the happy marriage between natural language processing (NLP) and the cognitive sci- ence of language. For the NLP → cognitive science direc- tion, engineering models originally proposed in NLP have been employed as computational models of human lan- guage processing and evaluated against human behavioral data such as self-paced reading (Roark et al., 2009; Frank et al., 2013) and eye-tracking (Frank and Bod, 2011; Fos- sum and Levy, 2012). In contrast, for the cognitive science → NLP direction, human behavioral data, especially eye- tracking, have been used to train engineering models and improve performance in part-of-speech tagging (Barrett et al., 2016), sentiment analysis (Mishra et al., 2016), named entity recognition (Hollenstein and Zhang, 2019), and at- tention in artificial neural networks (Barrett et al., 2018). Moreover, given the historical relationship between bi- ological and artificial neural networks (Amari, 1967; Fukushima, 1980), the advent of deep learning has re- sparked strong interests in the fusion of NLP and the neu- roscience of language. For example, neuro-computational models of human language processing have been con- structed based on symbolic automata and neural networks and evaluated against human neural data such as elec- troencephalography (EEG) (Frank et al., 2015; Brennan and Hale, 2019), functional resonance magnetic imaging (fMRI) (Brennan et al., 2016; Henderson et al., 2016), magnetoencephalography (MEG) (Brennan and Pylkk¨anen, 2017), and also electrocorticography (ECoG) (Nelson et al., 2017). In addition, human MEG and ECoG data have been employed to fine-tune state-of-the-art …

Content truncated to 500 words