Lightweight Spoken Utterance Classification with CFG, tf-idf and Dynamic Programming

Rayner, Manny; Tsourakis, Nikos; Gerlach, Johanna

doi:10.1007/978-3-319-68456-7_12

Cited by 5 publications

(11 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As noted, the fuzzy matching method is much more accurate than the strict grammar-based method; the experiments in [8] show relative reductions in 1-best semantic error rate by 35% and 2-best semantic error rate by 45% on realistic unseen speech data. This is unsurprising, given that commercial large-vocabulary recognisers, with a little domain tuning, can achieve word error rates of under 15% on BabelDr data, while the grammar-based recogniser's WER is typically in the neighbourhood of 30-40%.…”

Section: The Babeldr Medical Phraselatormentioning

confidence: 88%

“…The point is that the grammar-based recogniser's WER is much lower on the high-confidence portion of the data, and with a suitable threshold can be reduced to a point substantially under that of the large-vocabulary recogniser. The experiments in [8] show the hybrid method achieving a relative reduction in 1-best semantic error rate by 8% and 2-best semantic error rate by 20%, compared to the plain fuzzy matching method.…”

Section: Hybrid Processing and Machine Learningmentioning

confidence: 97%

“…We have found, however, that directly performing fuzzy matching of the grammar against the recogniser output is a highly competitive alternative. By weighting the words with tf-idf scores [9] and doing the match with a bottom-up dynamic programming algorithm 1 , we have obtained accuracy not worse than any of the machine learning methods we have so far investigated, and much better than the strict grammar-based method [8]. The fuzzy matching process is fast enough that response times are in practice dominated by the recognition speed of the large-vocabulary recogniser.…”

Section: The Babeldr Medical Phraselatormentioning

confidence: 99%

See 2 more Smart Citations

Handling Ellipsis in a Spoken Medical Phraselator

Rayner

Gerlach

Bouillon

et al. 2018

Statistical Language and Speech Processing

Self Cite

View full text Add to dashboard Cite

We consider methods for handling incomplete (elliptical) utterances in spoken phraselators, and describe how they have been implemented inside BabelDr, a substantial spoken medical phraselator. The challenge is to extend the phrase matching process so that it is sensitive to preceding dialogue context. We contrast two methods, one using limited-vocabulary strict grammar-based speech and language processing and one using large-vocabulary speech recognition with fuzzy grammar-based processing, and present an initial evaluation on a spoken corpus of 821 context-sentence/elliptical-phrase pairs. The large-vocabulary/fuzzy method strongly outperforms the limited-vocabulary/strict method over the whole corpus, though it is slightly inferior for the subset that is within grammar coverage. We investigate possibilities for combining the two processing paths, using several machine learning frameworks, and demonstrate that hybrid methods strongly outperform the largevocabulary/fuzzy method.

show abstract

Section: The Babeldr Medical Phraselatormentioning

confidence: 88%

Section: Hybrid Processing and Machine Learningmentioning

confidence: 97%

Section: The Babeldr Medical Phraselatormentioning

confidence: 99%

See 1 more Smart Citation

Handling Ellipsis in a Spoken Medical Phraselator

Rayner

Gerlach

Bouillon

et al. 2018

Statistical Language and Speech Processing

Self Cite

View full text Add to dashboard Cite

show abstract

“…In our experiments, the keyword number increases from 10 to 50. Each keyword follows TF-IDFweight [9]. k value (ie the number of results returned to the subscriber) increases from 2 to 10.…”

Section: Methodsmentioning

confidence: 99%

“…Each keyword t i is associated with a weight w(t i ). w(t i ) is TF-IDF[9] weight of keyword t i in s. (2) s G denotes s's geography description which is composed of latitude and longitude. (3) δ is a preference parameter.…”

mentioning

confidence: 99%

ETGP: Top-K Geography-Text P/S Approach without Threshold

2018

Teh. vjesn.

View full text Add to dashboard Cite

Social media are more and more popular. Subsequently, geography-text data has caused wide attention. Different from the traditional publish/subscribe (P/S), geography-text data is published and subscribed in the form of dynamic data flow in the mobile network. The difference raises higher demands for facility. However, previous top-k geography-text P/S approaches want to set a set of thresholds. A user should take time to set a threshold for each subscription, which is not facile enough. The threshold yields many weaknesses to users. Therefore, we herein propose an efficient top-k geography-text P/S approach that excludes the threshold, called ETGP. Our approach does not need users to set any threshold. Subsequently, the ETGP returns the highest score results to the subscriber without setting a threshold. Therefore, our approach can lessen redundant computations, promote the query integrity rate, and make P/S system easier for the user to use. Comprehensive experiments prove the efficiency of the proposed approach with high facility.

show abstract