A Deep Ensemble Model with Slot Alignment for Sequence-to-Sequence Natural Language Generation

Proceedings of NAACL-HLT 2018, pages 152–162 New Orleans, Louisiana, June 1 - 6, 2018. c⃝2018 Association for Computational Linguistics A Deep Ensemble Model with Slot Alignment for Sequence-to-Sequence Natural Language Generation Juraj Juraska, Panagiotis Karagiannis, Kevin K. Bowden and Marilyn A. Walker Natural Language and Dialogue Systems Lab University of California, Santa Cruz {jjuraska,pkaragia,kkbowden,mawalker}@ucsc.edu Abstract Natural language generation lies at the core of generative dialogue systems and conversa- tional agents. We describe an ensemble neural language generator, and present several novel methods for data representation and augmen- tation that yield improved results in our model. We test the model on three datasets in the restaurant, TV and laptop domains, and re- port both objective and subjective evaluations of our best model. Using a range of automatic metrics, as well as human evaluators, we show that our approach achieves better results than state-of-the-art models on the same datasets. 1 Introduction There has recently been a substantial amount of research in natural language processing (NLP) in the context of personal assistants, such as Cortana or Alexa. The capabilities of these conversational agents are still fairly limited and lacking in vari- ous aspects, one of the most challenging of which is the ability to produce utterances with human- like coherence and naturalness for many different kinds of content. This is the responsibility of the natural language generation (NLG) component. Our work focuses on language generators whose inputs are structured meaning representa- tions (MRs). An MR describes a single dialogue act with a list of key concepts which need to be conveyed to the human user during the dialogue. Each piece of information is represented by a slot- value pair, where the slot identifies the type of in- formation and the value is the corresponding con- tent. Dialogue act (DA) types vary depending on the dialogue manager, ranging from simple ones, such as a goodbye DA with no slots at all, to com- plex ones, such as an inform DA containing multi- ple slots with various types of values (see example in Table 1). MR inform (name [The Golden Curry], food [Japanese], priceRange [moderate], fami- lyFriendly [yes], near [The Bakers]) Utt. Located near The Bakers, kid-friendly restau- rant, The Golden Curry, offers Japanese cui- sine with a moderate price range. Table 1: An example of an MR and a corresponding reference utterance. A natural language generator must produce a syntactically and semantically correct utterance from a given MR. The utterance should express all the information contained in the MR, in a natu- ral and conversational way. In traditional language generator architectures, the assembling of an utter- ance from an MR is performed in two stages: sen- tence planning, which enforces semantic correct- ness and determines the structure of the utterance, and surface realization, which enforces syntactic correctness and produces the final utterance form. Earlier work on statistical NLG approaches were typically hybrids of a handcrafted compo- nent and a statistical training method (Langkilde and Knight, 1998; Stent et al., 2004; Rieser and Lemon, 2010). …

Content truncated to 500 words