Multi-domain Neural Network Language Generation for Spoken Dialogue Systems

Proceedings of NAACL-HLT 2016, pages 120–129, San Diego, California, June 12-17, 2016. c⃝2016 Association for Computational Linguistics Multi-domain Neural Network Language Generation for Spoken Dialogue Systems Tsung-Hsien Wen, Milica Gaˇsi´c, Nikola Mrkˇsi´c, Lina M. Rojas-Barahona, Pei-Hao Su, David Vandyke, Steve Young Cambridge University Engineering Department, Trumpington Street, Cambridge, CB2 1PZ, UK {thw28,mg436,nm480,lmr46,phs26,djv27,sjy}@cam.ac.uk Abstract Moving from limited-domain natural lan- guage generation (NLG) to open domain is difficult because the number of semantic in- put combinations grows exponentially with the number of domains. Therefore, it is im- portant to leverage existing resources and ex- ploit similarities between domains to facilitate domain adaptation. In this paper, we propose a procedure to train multi-domain, Recurrent Neural Network-based (RNN) language gen- erators via multiple adaptation steps. In this procedure, a model is first trained on counter- feited data synthesised from an out-of-domain dataset, and then fine tuned on a small set of in-domain utterances with a discriminative ob- jective function. Corpus-based evaluation re- sults show that the proposed procedure can achieve competitive performance in terms of BLEU score and slot error rate while signifi- cantly reducing the data needed to train gen- erators in new, unseen domains. In subjective testing, human judges confirm that the proce- dure greatly improves generator performance when only a small amount of data is available in the domain. 1 Introduction Modern Spoken Dialogue Systems (SDS) are typi- cally developed according to a well-defined ontol- ogy, which provides a structured representation of the domain data that the dialogue system can talk about, such as searching for a restaurant or shop- ping for a laptop. Unlike conventional approaches employing a substantial amount of handcrafting for each individual processing component (Ward and Is- sar, 1994; Bohus and Rudnicky, 2009), statistical ap- proaches to SDS promise a domain-scalable frame- work which requires a minimal amount of human in- tervention (Young et al., 2013). Mrkˇsi´c et al. (2015) showed improved performance in belief tracking by training a general model and adapting it to specific domains. Similar benefit can be observed in Gaˇsi´c et al. (2015), in which a Bayesian committee ma- chine (Tresp, 2000) was used to model policy learn- ing in a multi-domain SDS regime. In past decades, adaptive NLG has been stud- ied from linguistic perspectives, such as systems that learn to tailor user preferences (Walker et al., 2007), convey a specific personality trait (Mairesse and Walker, 2008; Mairesse and Walker, 2011), or align with their conversational partner (Isard et al., 2006). Domain adaptation was first ad- dressed by Hogan et al. (2008) using a generator based on the Lexical Functional Grammar (LFG) f- structures (Kaplan and Bresnan, 1982). Although these approaches can model rich linguistic phe- nomenon, they are not readily adaptable to data since they still require many handcrafted rules to define the search space. Recently, RNN-based lan- guage generation has been introduced (Wen et al., 2015a; Wen et al., 2015b). This class of statistical generators can learn generation decisions directly from dialogue act (DA)-utterance pairs without any semantic annotations (Mairesse …

Content truncated to 500 words