Preserving Distributional Information in Dialogue Act Classification

Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 2151–2156 Copenhagen, Denmark, September 7–11, 2017. c⃝2017 Association for Computational Linguistics Preserving Distributional Information in Dialogue Act Classification Quan Hung Tran and Ingrid Zukerman and Gholamreza Haffari Faculty of Information Technology Monash University, Australia hung.tran,ingrid.zukerman,gholamreza.haffari@monash.edu Abstract This paper introduces a novel train- ing/decoding strategy for sequence label- ing. Instead of greedily choosing a la- bel at each time step, and using it for the next prediction, we retain the proba- bility distribution over the current label, and pass this distribution to the next pre- diction. This approach allows us to avoid the effect of label bias and error propaga- tion in sequence learning/decoding. Our experiments on dialogue act classification demonstrate the effectiveness of this ap- proach. Even though our underlying neu- ral network model is relatively simple, it outperforms more complex neural mod- els, achieving state-of-the-art results on the MapTask and Switchboard corpora. 1 Introduction Dialogue Act (DA) classification is a sequence- labeling task, where a sequence of utterances is mapped into a sequence of DAs. The DAs are se- mantic classifications of the utterances, and differ- ent corpora usually have their own DA labels. Two of the most popular DA classification datasets are Switchboard (Godfrey et al., 1992; Ju- rafsky et al., 1997) and MapTask (Anderson et al., 1991). There have been many works on DA classi- fication applied to these two datasets; some focus on textual data (Kalchbrenner and Blunsom, 2013; Stolcke et al., 2000), while others explore speech data (Julia et al., 2010). The classification meth- ods used can be broadly divided into instance- based methods (Julia et al., 2010; Gamb¨ack et al., 2011) and sequence-labeling methods (Stolcke et al., 2000; Kalchbrenner and Blunsom, 2013; Ji et al., 2016; Shen and Lee, 2016; Tran et al., 2017). Instance-based methods treat each utter- ance as an independent data point, which allows the application of general machine learning mod- els, such as Support Vector Machines. Sequence- labeling methods include methods based on Hid- den Markov Models (HMMs) (Stolcke et al., 2000) and neural networks (Kalchbrenner and Blunsom, 2013; Ji et al., 2016; Shen and Lee, 2016; Tran et al., 2017). Stolcke et al. employed an HMM, using a Language Model to produce emission probabili- ties. The neural models are particularly success- ful, posting a higher accuracy on Switchboard than the HMM. Specifically, Kalchbrenner and Blun- som (2013) model a DA sequence with a recur- rent neural network (RNN) where sentence repre- sentations are constructed by means of a convolu- tional neural network (CNN); Ji et al. (2016) treat the labels as latent variables in a generative RNN; Shen and Lee (2016) employ attentional RNNs for the independent prediction of DAs; and Tran et al. (2017) model the DAs in a conversation by means of a hierarchical RNN. In this paper, we also rely on RNNs, but our architecture is much simpler than the above neural models, while post- ing competitive results. Most neural network models for …

Content truncated to 500 words