Non-deterministic stochastic language models for speech recognition

Riccardi, Giuseppe; Bocchieri, Enrico; Pieraccini, Roberto

doi:10.1109/icassp.1995.479408

Cited by 33 publications

(31 citation statements)

References 3 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…We denote by P(w, ′ ) the probability of generating the word w and visiting the state ′ given that the state is occupied. There is no requirement that the language model be deterministic (in the sense of Riccardi et al, 1995). One advantage of accommodating a general finite state machine as a language model is that it enables us to use our search software for training as well as recognition.…”

Section: The Language Modelmentioning

confidence: 99%

Bi-directional graph search strategies for speech recognition

Zhi-shun¹,

Labute²,

Barszcz³

et al. 1996

Computer Speech & Language

View full text Add to dashboard Cite

Section: The Language Modelmentioning

confidence: 99%

Bi-directional graph search strategies for speech recognition

Zhi-shun¹,

Labute²,

Barszcz³

et al. 1996

Computer Speech & Language

View full text Add to dashboard Cite

“…Two of the methods-Variable Length Markov Models (VLMMs) (Guyon & Pereira, 1995) and Variable N-gram Stochastic Automata (VNSA) (Riccardi, Bocchieri & Pieraccini, 1995) are the most recent developments reported in the literature. The third method, called Refined Probabilistic Finite Automata (RPFA), which will be introduced in this paper, is based on the highly successful text compression algorithm called Dynamic Markov Compression (DMC) (Cormack & Horspool, 1987).…”

Section: Introductionmentioning

confidence: 99%

Language modeling using stochastic automata with variable length contexts

Hu¹,

Turin

Brown³

1997

Computer Speech & Language

View full text Add to dashboard Cite

“…Then an SLU module that uses the output of the ASR is trained and optimized for the understanding performance. However, as it has been pointed out many times, the hypothesis that gives a better recognition performance does not always yield a better understanding performance [108,127,38]. If the end goal of SLS is to understand what the user means and respond accordingly, both modules can be optimized jointly such that the system "understands better what it recognizes" and "recognizes better what it understands".…”

Section: Motivationmentioning

confidence: 99%

“…SELMs are important because as it has been argued multiple times [108,127] improving WER does not always yield an improvement on the understanding performance. This is especially important for spoken language systems which extracts semantic structures out of the utterances.…”

Section: Chapter 9 Deep Encodings For Semantic Language Modelsmentioning

confidence: 99%

Semantic language models with deep neural networks

Bayer

Riccardi

2016

Computer Speech & Language

View full text Add to dashboard Cite

Spoken language systems (SLS) communicate with users in natural language through speech. There are two main problems related to processing the spoken input in SLS. The first one is automatic speech recognition (ASR) which recognizes what the user says. The second one is spoken language understanding (SLU) which understands what the user means. We focus on the language model (LM) component of SLS. LMs constrain the search space that is used in the search for the best hypothesis. Therefore, they play a crucial role in the performance of SLS.It has long been discussed that an improvement in the recognition performance does not necessarily yield a better understanding performance. Therefore, optimization of LMs for the understanding performance is crucial. In addition, long-range dependencies in languages are hard to handle with statistical language models. These two problems are addressed in this thesis.We investigate two different LM structures. The first LM that we investigate enable SLS to understand better what they recognize by searching the ASR hypotheses for the best understanding performance. We refer to these models as joint LMs. They use lexical and semantic units jointly in the LM. The second LM structure uses the semantic context of an utterance, which can also be described as "what the system understands", to search for a better hypothesis that improves the recognition and the understanding performance. We refer to these models as semantic LMs (SELMs). SELMs use features that are based on a well established theory of lexical semantics, namely the theory of frame semantics. They incorporate the semantic features which are extracted from the ASR hypothesis into the LM and handle long-range dependencies by using the semantic relationships between words and semantic context. ASR noise is propagated to the semantic features, to suppress this noise we introduce the use of deep semantic encodings for semantic feature extraction. In this way, SELMs optimize both the recognition and the understanding performance.

show abstract

Non-deterministic stochastic language models for speech recognition

Cited by 33 publications

References 3 publications

Bi-directional graph search strategies for speech recognition

Bi-directional graph search strategies for speech recognition

Language modeling using stochastic automata with variable length contexts

Semantic language models with deep neural networks

Contact Info

Product

Resources

About