A discriminative HMM/N-gram-based retrieval approach for mandarin spoken documents

Abstract

__________________________________________________________________________________________In recent years, statistical modeling approaches have steadily gained in popularity in the field of information retrieval. This article presents an HMM/N-gram-based retrieval approach for Mandarin spoken documents. The underlying characteristics and the various structures of this approach were extensively investigated and analyzed. The retrieval capabilities were verified by tests with word-and syllable-level indexing features and comparisons to the conventional vector-space model approach. To further improve the discrimination capabilities of the HMMs, both the expectation-maximization (EM) and minimum classification error (MCE) training algorithms were introduced in training. Fusion of information via indexing word-and syllable-level features was also investigated. The spoken document retrieval experiments were performed on the Topic Detection and Tracking Corpora (TDT-2 and TDT-3). Very encouraging retrieval performance was obtained. INTRODUCTIONOver the past three decades, statistical modeling approaches for speech and language processing have been studied extensively. Among the approaches, hidden Markov modeling (HMM) for speech recognition is undoubtedly the most prevalent and effective [Jelinek 1997]. In this approach, a set of statistical phoneme-or word-level HMMs was trained beforehand with a labeled speech corpus; the probability of the test speech utterance with respect to the HMMs was then evaluated on the HMM network to find the optimal phoneme or word sequence with the maximum likelihood. This statistical paradigm was first introduced for the information retrieval problem by BBN Technologies [Miller et al., 1999] and by Ponte and Croft [1998] and Song and Croft [1999], indicating very good potential, and was then extended in a number of

Spoken Document Retrieval Leveraging Unsupervised and Supervised Topic Modeling Techniques

“…We use the Mandarin Chinese collection of the TDT corpora for the retrospective retrieval task [9], such that the statistics for the entire document collection is obtainable. The Chinese news stories (text) from Xinhua News Agency are used as our test queries (or training query exemplars) and training corpus for all topic models (excluding test query set and training query exemplars).…”

“…On the other hand, subword-level indexing features behave more robustly against the Chinese word tokenization ambiguity, homophone ambiguity, open vocabulary problem, and speech recognition errors; hence, subword-based retrieval enhances recall. Accordingly, there is good reason to fuse the information obtained from indexing the features of different levels [9].…”

“…Accordingly, documents can be ranked in decreasing order of the probability that each document model generates the query. Yet, there have been a number of studies to further extend ULM in an attempt to account for the dependency between terms based on n-grams of various orders, or some grammar structures, mostly leading to only mild gains or even spoiled results [9], [12].…”

Statistical lattice-based spoken document retrieval

“…This method was further improved upon in Song and Croft [1999]. Chen et al [2004] applied Song and Croft's method to Mandarin SDR using 1-best ASR transcripts. In this task, it was also shown to outperform tf · idf (with logarithmically adjusted document and query term frequencies).…”

Word Topic Models for Spoken Document Retrieval and Transcription

“…For example, the n-gram modeling (especially the bigram and trigram modeling) approach, which determines the probability of a word given the preceding n-1 word history, is most prominently used [Jelinek and Mercer 1980;Rosenfeld 2000;Bellegarda 2004]. This statistical paradigm was first introduced for the information retrieval (IR) problems by Ponte and Croft [1998], Song and Croft [1999], and Miller et al [1999], indicating very good potential, and was then extended in a number of publications [Berger and Lafferty 1999;Hoffmann 1999;Lafferty and Zhai 2001;Chen et al 2004b]. In these approaches, the relevance measure between a query Q and a document D is expressed as P (D |Q ); that is, the probability that D is relevant given that the query Q is posed.…”