Slot Tagging for Task Oriented Spoken Language Understanding in Human-to-Human Conversation Scenarios

Proceedings of the 23rd Conference on Computational Natural Language Learning, pages 757–767 Hong Kong, China, November 3-4, 2019. c⃝2019 Association for Computational Linguistics 757 Slot Tagging for Task Oriented Spoken Language Understanding in Human-to-human Conversation Scenarios Kunho Kim†, Rahul Jha†, Kyle Williams†, Alex Marin†, Imed Zitouni‡∗ †Microsoft Corporation, Redmond, WA, USA ‡Google, Mountain View, CA, USA {kuki, rajh, kywillia, alemari}@microsoft.com, izitouni@google.com Abstract Task oriented language understanding (LU) in human-to-machine (H2M) conversations has been extensively studied for personal digital assistants. In this work, we extend the task ori- ented LU problem to human-to-human (H2H) conversations, focusing on the slot tagging task. Recent advances on LU in H2M con- versations have shown accuracy improvements by adding encoded knowledge from different sources. Inspired by this, we explore sev- eral variants of a bidirectional LSTM architec- ture that relies on different knowledge sources, such as Web data, search engine click logs, ex- pert feedback from H2M models, as well as previous utterances in the conversation. We also propose ensemble techniques that aggre- gate these different knowledge sources into a single model. Experimental evaluation on a four-turn Twitter dataset in the restaurant and music domains shows improvements in the slot tagging F1-score of up to 6.09% compared to existing approaches. 1 Introduction Spoken Language Understanding (SLU) is the first component in digital assistants geared towards task completion, such as Amazon Alexa or Mi- crosoft Cortana. The input to an SLU compo- nent is a natural language utterance from the user and its output is a structured representation that can be used by the downstream dialog components to select the next action. The structured repre- sentation used by most standard dialog agents is a semantic frame consisting of domains, intents and slots (Tur and De Mori, 2011). For example, the structured representation of “Find me a cheap Italian restaurant” is the domain Restaurant, the intent find place, and slots [cheap]price range, ∗Work done while the author was at Microsoft Corpora- tion Figure 1: Example of language understanding for task completion on a H2H conversation. In this work, our goal is to identify useful slots (marked with red rectan- gles). [Italian]cuisine, [restaurant]place type. Different sub-tasks within SLU have been extensively stud- ied for human-to-machine (H2M) task completion scenarios (Sarikaya et al., 2016). We extend the task oriented SLU problem to human-to-human (H2H) conversations. A digi- tal assistant can listen to the conversation between two or more humans and provide relevant informa- tion or suggest actions based on the structured rep- resentation captured with SLU. Figure 1 shows an example of capturing intents and slots expressed implicitly during a conversation between two hu- mans. The digital assistant can show general in- formation about the restaurant Mua, and provide the opening hours based on the captured structured representation. These types of H2H task comple- tion scenarios may allow digital assistants to sug- gest useful information to users in advance with- out them needing to explicitly ask questions. In this paper, we investigate SLU oriented to- 758 wards task completion …

Content truncated to 500 words