Sequential Decision Strategies for Machine Interpretation of Speech

Abstract

Abstract-Recognition errors made by automatic speech recognition (ASR) systems may not prevent the development of useful dialogue applications if the interpretation strategy has an introspection capability for evaluating the reliability of the results. This paper proposes an interpretation strategy which is particularly effective when applications are developed with a training corpus of moderate size. From the lattice of word hypotheses generated by an ASR system, a short list of conceptual structures is obtained with a set of finite state machines (FSM). Interpretation or a rejection decision is then performed by a tree-based strategy. The nodes of the tree correspond to elaboration-decision units containing a redundant set of classifiers. A decision tree based and two large margin classifiers are trained with a development set to become interpretation knowledge sources. Discriminative training of the classifiers selects linguistic and confidence-based features for contributing to a cooperative assessment of the reliability of an interpretation. Such an assessment leads to the definition of a limited number of reliability states. The probability that a proposed interpretation is correct is provided by its reliability state and transmitted to the dialogue manager. Experimental results are presented for a telephone service application.Index Terms-Confidence measures, decision strategy, speech recognition, spoken dialogue systems, spoken language understanding.

Automatic Allocation of Training Data for Speech Understanding Based on Multiple Model Combinations

“…It is also important to use a variety of language understanding methods [5], [6] such as ones using grammatical constraints and ones using less-structured statistical models. This is because no single understanding method achieves both robustness and accuracy.…”

“…Raymond et al developed a sequential decision strategy for LU results [5]. Hahn et al employed six different LU methods with a single ASR result and combined them using a weighted ROVER method [6].…”

Spoken language understanding: a survey

“…Combinations of human-crafted knowledge and the results of automatic learning of semantic knowledge are proposed in [25]. The concurrent use of SCT, Boosting, and SVM classifiers is proposed in [22] to increase classification robustness.…”

“…In [22] the conceptual decoding process is seen as a translation process in which stochastic Language Models are implemented by Finite State Machines (FSM) which output labels for semantic constituents. There is an FSM for each elementary conceptual constituent.…”

“…In [22] an interpretation strategy implemented by a decision tree is proposed. In addition to confidence measures given at the concept level, a limited set of reliability states is defined for characterizing a whole utterance interpretation.…”