Implementing Spoken Language Understanding

Abstract

There is a wide range of tools that support various tasks in spoken language, some of which are particularly relevant for processing spoken language understanding in conversational interfaces. Here, the main task is to detect the user's intent and to extract any further information that is required to understand the utterance. This chapter provides a tutorial on the Api.ai platform that has been widely used to support the development of mobile and wearable devices as well as applications for smart homes and automobiles. The chapter also reviews some similar tools provided by Wit.ai, Amazon Alexa, and Microsoft LUIS, and looks briefly at other tools that have been widely used in natural language processing and that are potentially relevant for conversational interfaces. 9.1 Introduction As we saw in Chap. 8, spoken language understanding is not a uniform technology. Some tasks such as tokenization and part-of-speech tagging are used for low-level processing that will contribute to subsequent stages of analysis, while others per-form more high-level tasks such as providing a semantic interpretation of an utterance. For conversational interfaces, a widely used approach involves detecting the intent behind the user's utterance and extracting the relevant entities. The intent might be some action such as setting an alarm, scheduling a meeting, sending a text message, or booking a table at a restaurant. The entities are those elements of meaning that are essential to the execution of the action, such as the time for the alarm or the meeting, the recipient of the text message and its content, or the number of people for the restaurant booking. A number of spoken language understanding platforms take the approach of intent recognition and entity extraction, including Api.ai, Wit.ai, Amazon Alexa, and Microsoft LUIS. In this chapter, we will focus mainly on the Api.ai platform. In the final part of the chapter, we will provide an overview of some other tools that are used widely for spoken language understanding.