CASA-NLU: Context-Aware Self-Attentive Natural Language Understanding for Task-Oriented Chatbots

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 1285–1290, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 1285 CASA-NLU: Context-Aware Self-Attentive Natural Language Understanding for Task-Oriented Chatbots Arshit Gupta† Peng Zhang‡ Garima Lalwani‡ Mona Diab† †Amazon AI, Seattle ‡Amazon AI, East Palo Alto {arshig, pezha, glalwani, diabmona}@amazon.com Abstract Natural Language Understanding (NLU) is a core component of dialog systems. It typi- cally involves two tasks - intent classification (IC) and slot labeling (SL), which are then fol- lowed by a dialogue management (DM) com- ponent. Such NLU systems cater to utterances in isolation, thus pushing the problem of con- text management to DM. However, contextual information is critical to the correct prediction of intents and slots in a conversation. Prior work on contextual NLU has been limited in terms of the types of contextual signals used and the understanding of their impact on the model. In this work, we propose a context- aware self-attentive NLU (CASA-NLU) model that uses multiple signals, such as previous intents, slots, dialog acts and utterances over a variable context window, in addition to the current user utterance. CASA-NLU outper- forms a recurrent contextual NLU baseline on two conversational datasets, yielding a gain of up to 7% on the IC task for one of the datasets. Moreover, a non-contextual variant of CASA- NLU achieves state-of-the-art performance for IC task on standard public datasets - SNIPS and ATIS. 1 Introduction With the advent of smart conversational agents such as Amazon Alexa, Google Assistant, etc., di- alogue systems are becoming ubiquitous. In the context of enterprises, the majority of these sys- tems target task oriented dialogues with the user trying to achieve a goal, e.g. booking flight tick- ets or ordering food. Natural Language Under- standing (NLU) captures the semantic meaning of a user’s utterance within each dialogue turn, by identifying intents and slots. An intent specifies the goal underlying the expressed utterance while slots are additional parameters for these intents. These tasks are typically articulated as intent clas- Hello! What can I do for you today? uh, help! Help Sure! You can signup for or modify an existing service. You can also query for your data usage or bills. increase my data plan please O O UpgradeService O O O O O What new plan would you like? unlimited UpgradeService [DataPlanType] Sure! Your data plan has now been changed to unlimited. Is there anything else you would like to do? sign me up for cable StartService O O O O [ServiceType] (1) (2) (3) (4) Figure 1: Snippet of a sample Cable data conversation with intents shown in boxes and slots aligned with user request sification (IC) coupled with sequence tagging task of slot labeling (SL). Over time, human-machine interactions have become more complex with greater reliance on contextual cues for utterance understanding (Fig- ure 1). With traditional NLU frameworks, the res- olution of contextual utterances is typically ad- …

Content truncated to 500 words