Combining Relevance Language Modeling and Clarity Measure for Extractive Speech Summarization

Shih Hung Liu, Kuan Yu Chen, Berlin Chen, Hsin Min Wang, Hsu Chun Yen, Wen Lian Hsu

Research output: Contribution to journalArticle

18 Citations (Scopus)

Abstract

Extractive speech summarization, which purports to select an indicative set of sentences from a spoken document so as to succinctly represent the most important aspects of the document, has garnered much research over the years. In this paper, we cast extractive speech summarization as an ad-hoc information retrieval (IR) problem and investigate various language modeling (LM) methods for important sentence selection. The main contributions of this paper are four-fold. First, we explore a novel sentence modeling paradigm built on top of the notion of relevance, where the relationship between a candidate summary sentence and a spoken document to be summarized is discovered through different granularities of context for relevance modeling. Second, not only lexical but also topical cues inherent in the spoken document are exploited for sentence modeling. Third, we propose a novel clarity measure for use in important sentence selection, which can help quantify the thematic specificity of each individual sentence that is deemed to be a crucial indicator orthogonal to the relevance measure provided by the LM-based methods. Fourth, in an attempt to lessen summarization performance degradation caused by imperfect speech recognition, we investigate making use of different levels of index features for LM-based sentence modeling, including words, subword-level units, and their combination. Experiments on broadcast news summarization seem to demonstrate the performance merits of our methods when compared to several existing well-developed and/or state-of-the-art methods.

Original languageEnglish
Article number7063924
Pages (from-to)957-969
Number of pages13
JournalIEEE Transactions on Audio, Speech and Language Processing
Volume23
Issue number6
DOIs
Publication statusPublished - 2015 Jun 1

Fingerprint

clarity
sentences
Information retrieval
Speech recognition
Degradation
Experiments
information retrieval
news
cues
speech recognition
casts
degradation

Keywords

  • Clarity measure
  • KL divergence
  • language modeling
  • relevance modeling
  • speech summarization

ASJC Scopus subject areas

  • Acoustics and Ultrasonics
  • Electrical and Electronic Engineering

Cite this

Combining Relevance Language Modeling and Clarity Measure for Extractive Speech Summarization. / Liu, Shih Hung; Chen, Kuan Yu; Chen, Berlin; Wang, Hsin Min; Yen, Hsu Chun; Hsu, Wen Lian.

In: IEEE Transactions on Audio, Speech and Language Processing, Vol. 23, No. 6, 7063924, 01.06.2015, p. 957-969.

Research output: Contribution to journalArticle

Liu, Shih Hung ; Chen, Kuan Yu ; Chen, Berlin ; Wang, Hsin Min ; Yen, Hsu Chun ; Hsu, Wen Lian. / Combining Relevance Language Modeling and Clarity Measure for Extractive Speech Summarization. In: IEEE Transactions on Audio, Speech and Language Processing. 2015 ; Vol. 23, No. 6. pp. 957-969.
@article{98b4b7f1027a45aca931beee8297cf9a,
title = "Combining Relevance Language Modeling and Clarity Measure for Extractive Speech Summarization",
abstract = "Extractive speech summarization, which purports to select an indicative set of sentences from a spoken document so as to succinctly represent the most important aspects of the document, has garnered much research over the years. In this paper, we cast extractive speech summarization as an ad-hoc information retrieval (IR) problem and investigate various language modeling (LM) methods for important sentence selection. The main contributions of this paper are four-fold. First, we explore a novel sentence modeling paradigm built on top of the notion of relevance, where the relationship between a candidate summary sentence and a spoken document to be summarized is discovered through different granularities of context for relevance modeling. Second, not only lexical but also topical cues inherent in the spoken document are exploited for sentence modeling. Third, we propose a novel clarity measure for use in important sentence selection, which can help quantify the thematic specificity of each individual sentence that is deemed to be a crucial indicator orthogonal to the relevance measure provided by the LM-based methods. Fourth, in an attempt to lessen summarization performance degradation caused by imperfect speech recognition, we investigate making use of different levels of index features for LM-based sentence modeling, including words, subword-level units, and their combination. Experiments on broadcast news summarization seem to demonstrate the performance merits of our methods when compared to several existing well-developed and/or state-of-the-art methods.",
keywords = "Clarity measure, KL divergence, language modeling, relevance modeling, speech summarization",
author = "Liu, {Shih Hung} and Chen, {Kuan Yu} and Berlin Chen and Wang, {Hsin Min} and Yen, {Hsu Chun} and Hsu, {Wen Lian}",
year = "2015",
month = "6",
day = "1",
doi = "10.1109/TASLP.2015.2414820",
language = "English",
volume = "23",
pages = "957--969",
journal = "IEEE Transactions on Audio, Speech and Language Processing",
issn = "1558-7916",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
number = "6",

}

TY - JOUR

T1 - Combining Relevance Language Modeling and Clarity Measure for Extractive Speech Summarization

AU - Liu, Shih Hung

AU - Chen, Kuan Yu

AU - Chen, Berlin

AU - Wang, Hsin Min

AU - Yen, Hsu Chun

AU - Hsu, Wen Lian

PY - 2015/6/1

Y1 - 2015/6/1

N2 - Extractive speech summarization, which purports to select an indicative set of sentences from a spoken document so as to succinctly represent the most important aspects of the document, has garnered much research over the years. In this paper, we cast extractive speech summarization as an ad-hoc information retrieval (IR) problem and investigate various language modeling (LM) methods for important sentence selection. The main contributions of this paper are four-fold. First, we explore a novel sentence modeling paradigm built on top of the notion of relevance, where the relationship between a candidate summary sentence and a spoken document to be summarized is discovered through different granularities of context for relevance modeling. Second, not only lexical but also topical cues inherent in the spoken document are exploited for sentence modeling. Third, we propose a novel clarity measure for use in important sentence selection, which can help quantify the thematic specificity of each individual sentence that is deemed to be a crucial indicator orthogonal to the relevance measure provided by the LM-based methods. Fourth, in an attempt to lessen summarization performance degradation caused by imperfect speech recognition, we investigate making use of different levels of index features for LM-based sentence modeling, including words, subword-level units, and their combination. Experiments on broadcast news summarization seem to demonstrate the performance merits of our methods when compared to several existing well-developed and/or state-of-the-art methods.

AB - Extractive speech summarization, which purports to select an indicative set of sentences from a spoken document so as to succinctly represent the most important aspects of the document, has garnered much research over the years. In this paper, we cast extractive speech summarization as an ad-hoc information retrieval (IR) problem and investigate various language modeling (LM) methods for important sentence selection. The main contributions of this paper are four-fold. First, we explore a novel sentence modeling paradigm built on top of the notion of relevance, where the relationship between a candidate summary sentence and a spoken document to be summarized is discovered through different granularities of context for relevance modeling. Second, not only lexical but also topical cues inherent in the spoken document are exploited for sentence modeling. Third, we propose a novel clarity measure for use in important sentence selection, which can help quantify the thematic specificity of each individual sentence that is deemed to be a crucial indicator orthogonal to the relevance measure provided by the LM-based methods. Fourth, in an attempt to lessen summarization performance degradation caused by imperfect speech recognition, we investigate making use of different levels of index features for LM-based sentence modeling, including words, subword-level units, and their combination. Experiments on broadcast news summarization seem to demonstrate the performance merits of our methods when compared to several existing well-developed and/or state-of-the-art methods.

KW - Clarity measure

KW - KL divergence

KW - language modeling

KW - relevance modeling

KW - speech summarization

UR - http://www.scopus.com/inward/record.url?scp=84928124743&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84928124743&partnerID=8YFLogxK

U2 - 10.1109/TASLP.2015.2414820

DO - 10.1109/TASLP.2015.2414820

M3 - Article

AN - SCOPUS:84928124743

VL - 23

SP - 957

EP - 969

JO - IEEE Transactions on Audio, Speech and Language Processing

JF - IEEE Transactions on Audio, Speech and Language Processing

SN - 1558-7916

IS - 6

M1 - 7063924

ER -