Vladimir Popescu

Logic-based rhetorical structuring for natural language generation in human-computer dialogue

Usually, human-computer dialogue systems rely on ad-hoc solutions for the component performing speech turn generation, in natural language. However, integration of taskspecific and general world knowledge in order to provide a more reliable and natural interaction with humans also through more sophisticated language generation techniques becomes needed. In this paper we present performance improvements of a module simulating in first-order logic Segmented Discourse Representation Theory for language generation in dialogue. These improvements concern reductions in computational costs and enhancements in rhetorical coherence for the discourse structures obtained, and are obtained using speech-act related information for driving rhetorical relations computations.

Effective Speaker Tracking Strategies for Multi-party Human-Computer Dialogue

Human-computer dialogue is already a rather mature research field [10] that already boiled down to several commercial applications, either service or task-oriented [11]. Nevertheless, several issues remain to be tackled, when unrestricted, spontaneous dialogue is concerned: barge-in (when users interrupt the system or interrupt each other) must be properly handled, hence Voice Activity Detection is a crucial point [13]. Moreover, when multi-party interactions are allowed (i.e., the machine engages simultaneously in dialogue with several users), supplementary robustness constraints occur: the speakers have to be properly tracked, so that each utterance is mapped to a certain speaker that had produced it. This is needed in order to perform a reliable analysis of input utterances [2].

A constraint satisfaction approach to context-sensitive utterance generation in multi-party dialogue systems

In this paper we present a constraint satisfaction approach to the pragmatic control of the language generation process in dialogue systems. After a brief problem statement providing motivational background for a constraint satisfaction approach to context sensitive language generation, a theoretical framework designed and implemented in part by the authors (in the context of several research and development projects) will be argued for in concise manner. Particularly, the way in which constraints influence the generation of context-sensitive, relevant utterances in simultaneous interaction with several human speakers will be considered in detail, providing a computational framework for handling “pragmatic” aspects, such as the degree of illocutionary force, semantic ellipsis, or lexical selection of concessive connectors. Then, limiting cases of our approach will be pointed out, showing in what measure by acting on the constraints we can enlarge these limits.

Word error rate improvement and complexity reduction in Automatic Speech Recognition by analyzing acoustic model uncertainty and confusion

In this paper, a study about the uncertainty of the trained acoustic models and the confusion among these models is made in the context of speech recognition. The purpose is to find the most relevant voice features, hence the analysis is made on a per-feature basis. Model uncertainty is defined as a measure of feature distribution overlapping. A model is compared only to the models it is more similar to. Hence, confusion matrices are built from both feature distributions and recognition results. Next, the voice features are weighted according to their relevance in order to increase the discrimination among models, while relevance itself is deduced from the values of model uncertainty. Experimental results show that, by appropriate weighting, the recognition accuracy, in terms of Word Error Rate (WER), improves. Moreover, by removing the features with lower weights, the recognition accuracy is maintained, but the number of calculations is significantly reduced.

Pragmatically-Motivated Utterance Fine-Tuning in Human-Computer Dialogue

Pragmatically-Motivated Utterance Fine-Tuning in Human-Computer Dialogue The naturalness and user-friendliness of the utterances generated by the computer when engaging in dialogue with humans is a key point for the success of spoken language interaction-based computer applications. This article addresses the issue by proposing a mechanism for controlling the strength of the illocutionary force conveyed by the utterances produced by the machine. The degree of strength for a speech act roughly quantifies the pressure that this act puts on its recipient. Few research efforts are reported for controlling this pressure in utterance generation. This article provides the means to adjust this force, relying on the discourse structure of the dialogue and on the public commitments that the speakers make during the dialogue. After a concise statement of the formal framework, the proposed approach is presented in a detailed manner and qualitatively assessed via relevant examples.

Optimizing a discourse structuring component for utterance generation in human-computer dialogue

In this paper we describe a series of optimizations brought to a discourse structuring module for answer generation in dialogue systems. Starting from a baseline rhetorical structure updating algorithm, based on a first-order logic simulation of SDRT (“Segmented Discourse Representation Theory”), several improved versions are obtained, applying a series of pragmatically-driven, psychologically-motivated or purely computational optimizations. These optimizations stem from a reduction in the complexity of the baseline algorithm, as well as from a better conformance of the rhetorical structures obtained to linguistic predictions.