Probabilistic matching for dialog state tracking with limited training data

Abstract

A system and method for dialog state tracking employ an ontology in which a set of values are indexed by slot. A segment of a dialog is processed to detect mentions. Candidate slot values are extracted from the ontology, based on the detected mentions. The candidate slot values are ranked. A dialog state is updated, based on the ranking of the candidate slot values, which may be conditioned on the output of a temporal model, which predicts whether the value of the slot has been instantiated, modified, or is unchanged.

Description

BACKGROUND[0001]The exemplary embodiment relates to dialog systems and finds particular application in connection with a system and method for tracking a dialog state using probabilistic matching with an ontology.[0002]Automated dialog systems interact with users in a natural language to help them achieve a goal. As an example, a user may be interested in finding a restaurant and may have a set of constraints, such as geographic location, date, and time. The system offers the name of a restaurant that satisfies the constraints. The user may then request additional information about the restaurant. The dialogue continues until the user's questions are answered. There are many other applications where dialog systems could be advantageous. For example, in the context of customer call centers, efficient automation could bring a gain of productivity by increasing the probability of success of each call while reducing the overall cost.[0003]The use of autonomous dialog systems is rapidly ...

AI Technical Summary

Benefits of technology

The patent describes a method and system for tracking the state of a dialog, which involves detecting mentions in an input speech segment, extracting candidate slot values from an ontology, ranking the values, and updating the dialog state based on the ranking. The system can also condition the updating of the dialog state based on the prediction of a temporal model that predicts whether the value of a slot has been instantiated, modified, or is unchanged. The technical effect of this invention is to enable a more efficient and accurate tracking of the state of a dialog, which can be useful in various applications such as speech recognition or natural language processing.

Problems solved by technology

The use of autonomous dialog systems is rapidly growing with the spread of smart mobile devices but still faces challenges to becoming a primary user interface for natural interaction using conversations.

In particular, when dialogs are conducted in noisy environments or when utterances themselves are noisy, it can be difficult for the system to recognize or understand the user utterances.

However, in spoken dialog systems, where the user utterance is input as a voice recording, the errors incurred by Automatic Speech Recognition (ASR) and Natural Language Understanding (NLU) means that the true user utterance may not be directly observable.

This makes it difficult to compute the true dialog state.

A primary drawback of these two statistically-based approaches is the need of extensive data to embed the knowledge for inferring a state tracking model.

While gathering data is often a feasible task, annotating a gathered dialog corpus can be time-consuming and costly.

Images