Skip to main content

Ye-yi Wang

Followers

3

Following

1

Co-author

1

Public Views

University of Bristol

Radosław Zyzik

Jesuit University Ignatianum in Kraków

University of Sarajevo

Viacheslav Kuleshov

Stockholm University

Carnegie Mellon University

Michael Walpole

The University of New South Wales

University of Hawaii at Manoa

Bengt Liljebladh

University of Gothenburg

Stephane Galland

Université de Technologie de Belfort-Montbéliard (UTBM)

Svetlana Petrova

Bergische University Wuppertal (Germany)

Interests

Uploads

Papers by Ye-yi Wang

The voice-rate dialog system for consumer ratings

Voice-Rate is an automated dialog system which provides access to over one million ratings of pro... more Voice-Rate is an automated dialog system which provides access to over one million ratings of products and businesses. By calling a toll-free number, consumers can access ratings for products, national businesses such as airlines, and local businesses such as restaurants. Voice-Rate also has a facility for recording and analyzing ratings that are given over the phone. The service has been primed with ratings taken from a variety of web sources, and we are augmenting these with user ratings. Voice-Rate can be accessed by dialing 1-877-456-DATA.

Connectionist F-structure

A traditional transfer system in machine translation maps between language structures and an inte... more A traditional transfer system in machine translation maps between language structures and an intermediate representation. Our con-nectionist transfer system maps from f-structures of one language to f-structures of another language. It encodes the intermediate representation implicitly in neural networks' activation patterns. The system is learnable, therefore it does not need any eeort in hand-crafting the representation and mapping rules. Experiments show the system has good scalability and generalizability performance.

Combination of CFG and n-gram modeling in semantic grammar learning

8th European Conference on Speech Communication and Technology (Eurospeech 2003)

SGStudio is a grammar authoring tool that eases semantic grammar development. It is capable of in... more SGStudio is a grammar authoring tool that eases semantic grammar development. It is capable of integrating different information sources and learning from annotated examples to induct CFG rules. In this paper, we investigate a modification to its underlying model by replacing CFG rules with n-gram statistical models. The new model is a composite of HMM and CFG. The advantages of the new model include its built-in robust feature and its scalability to an n-gram classifier when the understanding does not involve slot filling. We devised a decoder for the model. Preliminary results show that the new model achieved 32% error reduction in high resolution understanding.

Robust language understanding in mipad

7th European Conference on Speech Communication and Technology (Eurospeech 2001)

MiPad is an application prototype for the study of conversational, multi-modal interface in Micro... more MiPad is an application prototype for the study of conversational, multi-modal interface in Microsoft Research. It has a Tap and Talk interface that allows users to effectively interact with a PDA device. The major Spoken Language Understanding (SLU) engine component behind MiPad is a robust chart parser. This paper discusses some novel features of the parser that enable it to take full advantage of the Tap and Talk interface and better support semantic based analysis. It also describes some implementation issues so that these new features can be accommodated without slowing down the parser. The new implementation speeds up the parser by a factor of three, making it more suitable for a SLU server.

Mipad: a next generation PDA prototype

6th International Conference on Spoken Language Processing (ICSLP 2000)

MiPad is one of the application prototypes in a project codenamed Dr Who. As a wireless Personal ... more MiPad is one of the application prototypes in a project codenamed Dr Who. As a wireless Personal Digital Assistant (PDA), MiPad fully integrates continuous speech recognition (CSR) and spoken language understanding (SLU) to enable users to accomplish many common tasks using a multimodal interface and wireless technologies. It tries to solve the problem of pecking with tiny styluses or typing on minuscule keyboards in today's PDAs or smart phones. It also avoids the problem of being a cellular telephone that depends on speech-only interaction. MiPad incorporates a built-in microphone that activates whenever a field is selected. As a user taps the screen or uses a built-in roller to navigate, the tapping action narrows the number of possible instructions for spoken language processing. MiPad currently runs on a Windows CE Pocket PC with a Windows 2000 Server where speech recognition is performed. The Dr Who CSR engine has a 64k word vocabulary with a unified context-free grammar and ngram language model. The Dr Who SLU engine is based on a robust chart parser and a plan-based dialog manager. This paper discusses MiPad's design, implementation work in progress, and preliminary user study in comparison to the existing pen-based PDA interface.

MiPad: a multimodal interaction prototype

2001 IEEE International Conference on Acoustics, Speech, and Signal Processing. Proceedings (Cat. No.01CH37221)

Use and acquisition of semantic language model

Demonstration Papers at HLT-NAACL 2004 on XX - HLT-NAACL '04, 2004

Semantic language model is a technique that utilizes the semantic structure of an utterance to be... more Semantic language model is a technique that utilizes the semantic structure of an utterance to better rank the likelihood of words composing the sentence. When used in a conversational system, one can dynamically integrate the dialog state and domain semantics into the semantic language model to better guide the speech recognizer executing the decoding process. We describe one such application that employs semantic language model to cope with spontaneous speech in a robust manner. The semantic language model, though can be manually crafted without data, can benefit significantly from data driven machine learning techniques. An example based approach is also described here to demonstrate a viable approach.

An introduction to voice search

IEEE Signal Processing Magazine, 2008

oice search is the technology underlying many spoken dialog systems (SDSs) that provide users wit... more oice search is the technology underlying many spoken dialog systems (SDSs) that provide users with the information they request with a spoken query. The information normally exists in a large database, and the query has to be compared with a field in the database to obtain the relevant information. The contents of the field, such as business or product names, are often unstructured text. For example, directory assistance (DA) [1] is one of the most popular voice search applications, in which users issue a spoken query and an automated system returns the phone number and address information of a business or an individual. Other voice search applications include music/video management [2], business and product reviews [3], stock price quotes, and conference information systems [4], [5]. Figure 1 shows the typical architecture of a voice search system, where a user's utterance is first recognized with an automatic speech recognizer (ASR) that utilizes an acoustic model (AM), pronunciation model (PM), and language model (LM). The m-best results from the ASR are passed to a search component to obtain the n-best semantic interpretations; i.e., a list of up to n entries in the database. The interpretations are passed to a dialog manager (DM) subsequently. The DM utilizes confidence measures, which indicate the certainty of the interpretations, to decide how to present the n-best results. If the system has high confidence on a few entries, it directly presents them to the user. Otherwise, a disambiguation module is exploited to interact with the user to understand what he actually needs. VOICE SEARCH AND OTHER SPOKEN DIALOG TECHNOLOGIES SDSs are often chronologically categorized into three generations: informational, transactional, and problem solving [6], [7] (earlier command and control speech applications in the 1980s are not considered as SDSs in this categorization). The first-generation SDSs focus on providing users with the information they request, such as flight status and weather information. The second-generation SDSs conduct transactions automatically with users; e.g., to book air flight tickets or perform bank balance transfers. The third-generation SDSs are often used in customer support by interacting with callers to diagnose the problems they are experiencing with a device or a service.

An Integrative and Discriminative Technique for Spoken Utterance Classification

IEEE Transactions on Audio, Speech, and Language Processing, 2008

Traditional methods of spoken utterance classification (SUC) adopt two independently trained phas... more Traditional methods of spoken utterance classification (SUC) adopt two independently trained phases. In the first phase, an automatic speech recognition (ASR) module returns the most likely sentence for the observed acoustic signal. In the second phase, a semantic classifier transforms the resulting sentence into the most likely semantic class. Since the two phases are isolated from each other, such traditional SUC systems are suboptimal. In this paper, we present a novel integrative and discriminative learning technique for SUC to alleviate this problem, and thereby, reduce the semantic classification error rate (CER). Our approach revolves around the effective use of the-best lists generated by the ASR module to reduce semantic classification errors. The-best list sentences are first rescored using all the available knowledge sources. Then, the sentence that is most likely to helps reduce the CER are extracted from the-best lists as well as those sentences that are most likely to increase the CER. These sentences are used to discriminatively train the language and semantic-classifier models to minimize the overall semantic CER. Our experiments resulted in a reduction of CER from its initial value of 4.92% to 4.04% in the standard ATIS task.

Multi-task learning for spoken language understanding with shared slots

Interspeech 2011, 2011

This paper addresses the problem of learning multiple spoken language understanding (SLU) tasks t... more This paper addresses the problem of learning multiple spoken language understanding (SLU) tasks that have overlapping sets of slots. In such a scenario, it is possible to achieve better slot filling performance by learning multiple tasks simultaneously, as opposed to learning them independently. We focus on presenting a number of simple multi-task learning algorithms for slot filling systems based on semi-Markov CRFs, assuming the knowledge of shared slots. Furthermore, we discuss an intradomain clustering method that automatically discovers shared slots from training data. The effectiveness of our proposed approaches is demonstrated in an SLU application that involves three different yet related tasks.

Connectionist F-structure transfer

Current Issues in Linguistic Theory, 1997

A traditional transfer system in machine translation maps between language structures and an inte... more A traditional transfer system in machine translation maps between language structures and an intermediate representation. Our connectionist transfer system maps from f-structures of one language to f-structures of another language. It encodes the intermediate representation implicitly in neural networks' activation patterns. The system is learnable, therefore it does not need any e ort in handcrafting the representation and mapping rules. Experiments show the system has good scalability and generalizability performance.

INTERSPEECH 2007 The Voice-Rate Dialog System for Consumer Ratings

Voice-Rate is an experimental dialog system that makes product and business ratings available to ... more Voice-Rate is an experimental dialog system that makes product and business ratings available to consumers via a tollfree phone number. By calling Voice-Rate, users can access the ratings of more than one million products, a quarter million local businesses (restaurants), and three thousand national businesses. This paper describes the Voice Rate system, and solutions to three key technical challenges: robust name-matching, efficient disambiguation, and review synthesis for telephone playback. Voice-Rate can be accessed by calling 1-877-456-DATA (toll-free) within the U.S. Index Terms: dialog systems, consumer ratings, speech recognition, disambiguation, review summarization

Discriminative models for spoken language understanding

Interspeech 2006, 2006

This paper studies several discriminative models for spoken language understanding (SLU). While a... more This paper studies several discriminative models for spoken language understanding (SLU). While all of them fall into the conditional model framework, different optimization criteria lead to conditional random fields, perceptron, minimum classification error and large margin models. The paper discusses the relationship amongst these models and compares them in terms of accuracy, training speed and robustness.

Statistical spoken language understanding: from generative model to conditional model

Spoken Language Understanding (SLU) addresses the problem of extracting semantic meaning conveyed... more Spoken Language Understanding (SLU) addresses the problem of extracting semantic meaning conveyed in a user's utterance. Traditionally the problem is solved with a knowledge-based approach. In the past decade many data-driven statistical models have been proposed, all of them are in the generative framework. In our previous study, we have introduced a HMM/CFG composite model. It is a generative model that integrates knowledge-based approach in a statistical learning framework. We have investigated similar integration of prior knowledge and statistical learning in the framework of conditional models recently. This extended summary describes our experiences and presents some preliminary results, which shows a 17% slot error rate reduction over the generative model.

Semantic Frame-Based Spoken Language Understanding

Spoken Language Understanding, 2011

In the United States, the study of the frame-based SLU started in the 1970's in the DARPA Speech ... more In the United States, the study of the frame-based SLU started in the 1970's in the DARPA Speech Understanding Research (SUR) and then the Resource Management (RM) tasks. At this early stage, natural language understanding (NLU) techniques like finite state machine (FSM) and augmented transition networks (ATNs) were applied for SLU . The study of SLU surged in the 90's, with the DARPA sponsored Air Travel Information System (ATIS) evaluations ). Multiple research labs from both academia and industry, including AT&T, BBN, Carnegie Mellon University, MIT and SRI, developed systems that attempted to understand users' spontaneous spoken queries for air travel information (including flight information, ground transportation information, airport service information, etc.) and then obtain the answers from a standard database. ATIS is This is a Book Title Name of the Author/Editor c

XII. A meteorological journal, principally relating to atmospheric electricity; kept at Knightsbridge, from the 9th of May, 1789, to the 8th May, 1790

A description of the instrument for collecting atmospheric electricity, used in the following jou... more

Decoding Algorithm In Statistical Machine Translation

Proceedings of the eighth conference on …, 1997

Decoding algorithm is a crucial part in statistical machine translation. We describe a stack deco... more Decoding algorithm is a crucial part in statistical machine translation. We describe a stack decoding algorithm in this paper. We present the hypothesis scoring method and the heuristics used in our algorithm. We report several techniques deployed to improve the performance of the decoder. We also introduce a simplified model to moderate the sparse data problem and to speed up the decoding process. We evaluate and compare these techniques/models in our statistical machine translation system.

Semi-Supervised Learning of Semantic Classes for Query Understanding: From the Web and for the Web

Proceeding of the 18th …, 2009

Understanding intents from search queries can improve a user's search experience and boost a site... more Understanding intents from search queries can improve a user's search experience and boost a site's advertising profits. Query tagging via statistical sequential labeling models has been shown to perform well, but annotating the training set for supervised learning requires substantial human effort. Domain-specific knowledge, such as semantic class lexicons, reduces the amount of needed manual annotations, but much human effort is still required to maintain these as search topics evolve over time. This paper investigates semi-supervised learning algorithms that leverage structured data (HTML lists) from the Web to automatically generate semantic-class lexicons, which are used to improve query tagging performance-even with far less training data. We focus our study on understanding the correct objectives for the semi-supervised lexicon learning algorithms that are crucial for the success of query tagging. Prior work on lexicon acquisition has largely focused on the precision of the lexicons, but we show that precision is not important if the lexicons are used for query tagging. A more adequate criterion should emphasize a trade-off between maximizing the recall of semantic class instances in the data, and minimizing the confusability. This ensures that the similar levels of precision and recall are observed on both training and test set, hence prevents over-fitting the lexicon features. Experimental results on retail product queries show that enhancing a query tagger with lexicons learned with this objective reduces word level tagging errors by up to 25% compared to the baseline tagger that does not use any lexicon features. In contrast, lexicons obtained through a precision-centric learning algorithm even degrade the performance of a tagger compared to the baseline. Furthermore, the proposed method outperforms one in which semantic class lexicons have been extracted from a database.

Learning with click graph for query intent classification

ACM Transactions on Information Systems, 2010

Topical query classification, as one step toward understanding users&amp;amp;amp;#x27; search... more Topical query classification, as one step toward understanding users&amp;amp;amp;#x27; search intent, is gaining increasing attention in information retrieval. Previous works on this subject primarily focused on enrichment of query features, for example, by augmenting queries with search engine results. In this work, we investigate a completely orthogonal approach—instead of improving feature representation, we aim at drastically increasing the amount of

Lexicon modeling for query understanding

Lexicons are important resources for semantic tagging. However, commonly used lexicons collected ... more Lexicons are important resources for semantic tagging. However, commonly used lexicons collected from entity databases suffer from multiple problems, such as ambiguity, limited coverage and lack of relative importance. In this work we present a lexicon modeling technique that automatically expands the lexicon and assigns weights to its elements. For lexicon expansion, we use a generative model to extract patterns from query logs using known lexicon seeds, and discover new lexicon elements using the learned patterns. For lexicon weighting, we propose two approaches based on generative and discriminative models to learn the relative importance of lexicon elements from user click statistics. Experiments on text queries in multiple domains show that our lexicon modeling technique can significantly improve semantic tagging performance.

The voice-rate dialog system for consumer ratings

Voice-Rate is an automated dialog system which provides access to over one million ratings of pro... more Voice-Rate is an automated dialog system which provides access to over one million ratings of products and businesses. By calling a toll-free number, consumers can access ratings for products, national businesses such as airlines, and local businesses such as restaurants. Voice-Rate also has a facility for recording and analyzing ratings that are given over the phone. The service has been primed with ratings taken from a variety of web sources, and we are augmenting these with user ratings. Voice-Rate can be accessed by dialing 1-877-456-DATA.

Connectionist F-structure

A traditional transfer system in machine translation maps between language structures and an inte... more A traditional transfer system in machine translation maps between language structures and an intermediate representation. Our con-nectionist transfer system maps from f-structures of one language to f-structures of another language. It encodes the intermediate representation implicitly in neural networks' activation patterns. The system is learnable, therefore it does not need any eeort in hand-crafting the representation and mapping rules. Experiments show the system has good scalability and generalizability performance.

Combination of CFG and n-gram modeling in semantic grammar learning

8th European Conference on Speech Communication and Technology (Eurospeech 2003)

SGStudio is a grammar authoring tool that eases semantic grammar development. It is capable of in... more SGStudio is a grammar authoring tool that eases semantic grammar development. It is capable of integrating different information sources and learning from annotated examples to induct CFG rules. In this paper, we investigate a modification to its underlying model by replacing CFG rules with n-gram statistical models. The new model is a composite of HMM and CFG. The advantages of the new model include its built-in robust feature and its scalability to an n-gram classifier when the understanding does not involve slot filling. We devised a decoder for the model. Preliminary results show that the new model achieved 32% error reduction in high resolution understanding.

Robust language understanding in mipad

7th European Conference on Speech Communication and Technology (Eurospeech 2001)

MiPad is an application prototype for the study of conversational, multi-modal interface in Micro... more MiPad is an application prototype for the study of conversational, multi-modal interface in Microsoft Research. It has a Tap and Talk interface that allows users to effectively interact with a PDA device. The major Spoken Language Understanding (SLU) engine component behind MiPad is a robust chart parser. This paper discusses some novel features of the parser that enable it to take full advantage of the Tap and Talk interface and better support semantic based analysis. It also describes some implementation issues so that these new features can be accommodated without slowing down the parser. The new implementation speeds up the parser by a factor of three, making it more suitable for a SLU server.

Mipad: a next generation PDA prototype

6th International Conference on Spoken Language Processing (ICSLP 2000)

MiPad is one of the application prototypes in a project codenamed Dr Who. As a wireless Personal ... more MiPad is one of the application prototypes in a project codenamed Dr Who. As a wireless Personal Digital Assistant (PDA), MiPad fully integrates continuous speech recognition (CSR) and spoken language understanding (SLU) to enable users to accomplish many common tasks using a multimodal interface and wireless technologies. It tries to solve the problem of pecking with tiny styluses or typing on minuscule keyboards in today's PDAs or smart phones. It also avoids the problem of being a cellular telephone that depends on speech-only interaction. MiPad incorporates a built-in microphone that activates whenever a field is selected. As a user taps the screen or uses a built-in roller to navigate, the tapping action narrows the number of possible instructions for spoken language processing. MiPad currently runs on a Windows CE Pocket PC with a Windows 2000 Server where speech recognition is performed. The Dr Who CSR engine has a 64k word vocabulary with a unified context-free grammar and ngram language model. The Dr Who SLU engine is based on a robust chart parser and a plan-based dialog manager. This paper discusses MiPad's design, implementation work in progress, and preliminary user study in comparison to the existing pen-based PDA interface.

MiPad: a multimodal interaction prototype

2001 IEEE International Conference on Acoustics, Speech, and Signal Processing. Proceedings (Cat. No.01CH37221)

Use and acquisition of semantic language model

Demonstration Papers at HLT-NAACL 2004 on XX - HLT-NAACL '04, 2004

Semantic language model is a technique that utilizes the semantic structure of an utterance to be... more Semantic language model is a technique that utilizes the semantic structure of an utterance to better rank the likelihood of words composing the sentence. When used in a conversational system, one can dynamically integrate the dialog state and domain semantics into the semantic language model to better guide the speech recognizer executing the decoding process. We describe one such application that employs semantic language model to cope with spontaneous speech in a robust manner. The semantic language model, though can be manually crafted without data, can benefit significantly from data driven machine learning techniques. An example based approach is also described here to demonstrate a viable approach.

An introduction to voice search

IEEE Signal Processing Magazine, 2008

oice search is the technology underlying many spoken dialog systems (SDSs) that provide users wit... more oice search is the technology underlying many spoken dialog systems (SDSs) that provide users with the information they request with a spoken query. The information normally exists in a large database, and the query has to be compared with a field in the database to obtain the relevant information. The contents of the field, such as business or product names, are often unstructured text. For example, directory assistance (DA) [1] is one of the most popular voice search applications, in which users issue a spoken query and an automated system returns the phone number and address information of a business or an individual. Other voice search applications include music/video management [2], business and product reviews [3], stock price quotes, and conference information systems [4], [5]. Figure 1 shows the typical architecture of a voice search system, where a user's utterance is first recognized with an automatic speech recognizer (ASR) that utilizes an acoustic model (AM), pronunciation model (PM), and language model (LM). The m-best results from the ASR are passed to a search component to obtain the n-best semantic interpretations; i.e., a list of up to n entries in the database. The interpretations are passed to a dialog manager (DM) subsequently. The DM utilizes confidence measures, which indicate the certainty of the interpretations, to decide how to present the n-best results. If the system has high confidence on a few entries, it directly presents them to the user. Otherwise, a disambiguation module is exploited to interact with the user to understand what he actually needs. VOICE SEARCH AND OTHER SPOKEN DIALOG TECHNOLOGIES SDSs are often chronologically categorized into three generations: informational, transactional, and problem solving [6], [7] (earlier command and control speech applications in the 1980s are not considered as SDSs in this categorization). The first-generation SDSs focus on providing users with the information they request, such as flight status and weather information. The second-generation SDSs conduct transactions automatically with users; e.g., to book air flight tickets or perform bank balance transfers. The third-generation SDSs are often used in customer support by interacting with callers to diagnose the problems they are experiencing with a device or a service.

An Integrative and Discriminative Technique for Spoken Utterance Classification

IEEE Transactions on Audio, Speech, and Language Processing, 2008

Traditional methods of spoken utterance classification (SUC) adopt two independently trained phas... more Traditional methods of spoken utterance classification (SUC) adopt two independently trained phases. In the first phase, an automatic speech recognition (ASR) module returns the most likely sentence for the observed acoustic signal. In the second phase, a semantic classifier transforms the resulting sentence into the most likely semantic class. Since the two phases are isolated from each other, such traditional SUC systems are suboptimal. In this paper, we present a novel integrative and discriminative learning technique for SUC to alleviate this problem, and thereby, reduce the semantic classification error rate (CER). Our approach revolves around the effective use of the-best lists generated by the ASR module to reduce semantic classification errors. The-best list sentences are first rescored using all the available knowledge sources. Then, the sentence that is most likely to helps reduce the CER are extracted from the-best lists as well as those sentences that are most likely to increase the CER. These sentences are used to discriminatively train the language and semantic-classifier models to minimize the overall semantic CER. Our experiments resulted in a reduction of CER from its initial value of 4.92% to 4.04% in the standard ATIS task.

Multi-task learning for spoken language understanding with shared slots

Interspeech 2011, 2011

This paper addresses the problem of learning multiple spoken language understanding (SLU) tasks t... more This paper addresses the problem of learning multiple spoken language understanding (SLU) tasks that have overlapping sets of slots. In such a scenario, it is possible to achieve better slot filling performance by learning multiple tasks simultaneously, as opposed to learning them independently. We focus on presenting a number of simple multi-task learning algorithms for slot filling systems based on semi-Markov CRFs, assuming the knowledge of shared slots. Furthermore, we discuss an intradomain clustering method that automatically discovers shared slots from training data. The effectiveness of our proposed approaches is demonstrated in an SLU application that involves three different yet related tasks.

Connectionist F-structure transfer

Current Issues in Linguistic Theory, 1997

A traditional transfer system in machine translation maps between language structures and an inte... more A traditional transfer system in machine translation maps between language structures and an intermediate representation. Our connectionist transfer system maps from f-structures of one language to f-structures of another language. It encodes the intermediate representation implicitly in neural networks' activation patterns. The system is learnable, therefore it does not need any e ort in handcrafting the representation and mapping rules. Experiments show the system has good scalability and generalizability performance.

INTERSPEECH 2007 The Voice-Rate Dialog System for Consumer Ratings

Voice-Rate is an experimental dialog system that makes product and business ratings available to ... more Voice-Rate is an experimental dialog system that makes product and business ratings available to consumers via a tollfree phone number. By calling Voice-Rate, users can access the ratings of more than one million products, a quarter million local businesses (restaurants), and three thousand national businesses. This paper describes the Voice Rate system, and solutions to three key technical challenges: robust name-matching, efficient disambiguation, and review synthesis for telephone playback. Voice-Rate can be accessed by calling 1-877-456-DATA (toll-free) within the U.S. Index Terms: dialog systems, consumer ratings, speech recognition, disambiguation, review summarization

Discriminative models for spoken language understanding

Interspeech 2006, 2006

This paper studies several discriminative models for spoken language understanding (SLU). While a... more This paper studies several discriminative models for spoken language understanding (SLU). While all of them fall into the conditional model framework, different optimization criteria lead to conditional random fields, perceptron, minimum classification error and large margin models. The paper discusses the relationship amongst these models and compares them in terms of accuracy, training speed and robustness.

Statistical spoken language understanding: from generative model to conditional model

Spoken Language Understanding (SLU) addresses the problem of extracting semantic meaning conveyed... more Spoken Language Understanding (SLU) addresses the problem of extracting semantic meaning conveyed in a user's utterance. Traditionally the problem is solved with a knowledge-based approach. In the past decade many data-driven statistical models have been proposed, all of them are in the generative framework. In our previous study, we have introduced a HMM/CFG composite model. It is a generative model that integrates knowledge-based approach in a statistical learning framework. We have investigated similar integration of prior knowledge and statistical learning in the framework of conditional models recently. This extended summary describes our experiences and presents some preliminary results, which shows a 17% slot error rate reduction over the generative model.

Semantic Frame-Based Spoken Language Understanding

Spoken Language Understanding, 2011

In the United States, the study of the frame-based SLU started in the 1970's in the DARPA Speech ... more In the United States, the study of the frame-based SLU started in the 1970's in the DARPA Speech Understanding Research (SUR) and then the Resource Management (RM) tasks. At this early stage, natural language understanding (NLU) techniques like finite state machine (FSM) and augmented transition networks (ATNs) were applied for SLU . The study of SLU surged in the 90's, with the DARPA sponsored Air Travel Information System (ATIS) evaluations ). Multiple research labs from both academia and industry, including AT&T, BBN, Carnegie Mellon University, MIT and SRI, developed systems that attempted to understand users' spontaneous spoken queries for air travel information (including flight information, ground transportation information, airport service information, etc.) and then obtain the answers from a standard database. ATIS is This is a Book Title Name of the Author/Editor c

XII. A meteorological journal, principally relating to atmospheric electricity; kept at Knightsbridge, from the 9th of May, 1789, to the 8th May, 1790

A description of the instrument for collecting atmospheric electricity, used in the following jou... more

Decoding Algorithm In Statistical Machine Translation

Proceedings of the eighth conference on …, 1997

Decoding algorithm is a crucial part in statistical machine translation. We describe a stack deco... more Decoding algorithm is a crucial part in statistical machine translation. We describe a stack decoding algorithm in this paper. We present the hypothesis scoring method and the heuristics used in our algorithm. We report several techniques deployed to improve the performance of the decoder. We also introduce a simplified model to moderate the sparse data problem and to speed up the decoding process. We evaluate and compare these techniques/models in our statistical machine translation system.

Semi-Supervised Learning of Semantic Classes for Query Understanding: From the Web and for the Web

Proceeding of the 18th …, 2009

Understanding intents from search queries can improve a user's search experience and boost a site... more Understanding intents from search queries can improve a user's search experience and boost a site's advertising profits. Query tagging via statistical sequential labeling models has been shown to perform well, but annotating the training set for supervised learning requires substantial human effort. Domain-specific knowledge, such as semantic class lexicons, reduces the amount of needed manual annotations, but much human effort is still required to maintain these as search topics evolve over time. This paper investigates semi-supervised learning algorithms that leverage structured data (HTML lists) from the Web to automatically generate semantic-class lexicons, which are used to improve query tagging performance-even with far less training data. We focus our study on understanding the correct objectives for the semi-supervised lexicon learning algorithms that are crucial for the success of query tagging. Prior work on lexicon acquisition has largely focused on the precision of the lexicons, but we show that precision is not important if the lexicons are used for query tagging. A more adequate criterion should emphasize a trade-off between maximizing the recall of semantic class instances in the data, and minimizing the confusability. This ensures that the similar levels of precision and recall are observed on both training and test set, hence prevents over-fitting the lexicon features. Experimental results on retail product queries show that enhancing a query tagger with lexicons learned with this objective reduces word level tagging errors by up to 25% compared to the baseline tagger that does not use any lexicon features. In contrast, lexicons obtained through a precision-centric learning algorithm even degrade the performance of a tagger compared to the baseline. Furthermore, the proposed method outperforms one in which semantic class lexicons have been extracted from a database.

Learning with click graph for query intent classification

ACM Transactions on Information Systems, 2010

Topical query classification, as one step toward understanding users&amp;amp;amp;#x27; search... more Topical query classification, as one step toward understanding users&amp;amp;amp;#x27; search intent, is gaining increasing attention in information retrieval. Previous works on this subject primarily focused on enrichment of query features, for example, by augmenting queries with search engine results. In this work, we investigate a completely orthogonal approach—instead of improving feature representation, we aim at drastically increasing the amount of

Lexicon modeling for query understanding

Lexicons are important resources for semantic tagging. However, commonly used lexicons collected ... more Lexicons are important resources for semantic tagging. However, commonly used lexicons collected from entity databases suffer from multiple problems, such as ambiguity, limited coverage and lack of relative importance. In this work we present a lexicon modeling technique that automatically expands the lexicon and assigns weights to its elements. For lexicon expansion, we use a generative model to extract patterns from query logs using known lexicon seeds, and discover new lexicon elements using the learned patterns. For lexicon weighting, we propose two approaches based on generative and discriminative models to learn the relative importance of lexicon elements from user click statistics. Experiments on text queries in multiple domains show that our lexicon modeling technique can significantly improve semantic tagging performance.