guanglu wan - Academia.edu (original) (raw)

Papers by guanglu wan

Lecture Notes in Computer Science, Dec 31, 2022

arXiv (Cornell University), Mar 30, 2022

Utilizing text-only data with an external language model (ELM) in end-to-end RNN-Transducer (RNN-... more Utilizing text-only data with an external language model (ELM) in end-to-end RNN-Transducer (RNN-T) for speech recognition is challenging. Recently, a class of methods such as density ratio (DR) and internal language model estimation (ILME) have been developed, outperforming the classic shallow fusion (SF) method. The basic idea behind these methods is that RNN-T posterior should first subtract the implicitly learned internal language model (ILM) prior, in order to integrate the ELM. While recent studies suggest that RNN-T only learns some low-order language model information, the DR method uses a well-trained neural language model with full context, which may be inappropriate for the estimation of ILM and deteriorate the integration performance. Based on the DR method, we propose a loworder density ratio method (LODR) by replacing the estimation with a low-order weak language model. Extensive empirical experiments are conducted on both in-domain and cross-domain scenarios on English LibriSpeech & Tedlium-2 and Chinese WenetSpeech & AISHELL-1 datasets. It is shown that LODR consistently outperforms SF in all tasks, while performing generally close to ILME and better than DR in most tests.

Given a long text, the summarization system aims to obtain a shorter highlight while keeping impo... more Given a long text, the summarization system aims to obtain a shorter highlight while keeping important information on the original text. For customer service, the summaries of most dialogues between an agent and a user focus on several fixed key points, such as user's question, user's purpose, the agent's solution, and so on. Traditional extractive methods are difficult to extract all predefined key points exactly. Furthermore, there is a lack of large-scale and high-quality extractive summarization datasets containing key points. In order to solve the above challenges, we propose a Distant Supervision based Machine Reading Comprehension model for extractive Summarization (DSMRC-S). DSMRC-S transforms the summarization task into the machine reading comprehension problem, to fetch key points from the original text exactly according to the predefined questions. In addition, a distant supervision method is proposed to alleviate the lack of eligible extractive summarization datasets. We conduct experiments on a large-scale summarization dataset collected in customer service scenarios, and the results show that the proposed DSMRC-S outperforms the strong baseline methods by 4 points on ROUGE-L.

arXiv (Cornell University), May 12, 2022

Though widely used in industry, traditional task-oriented dialogue systems suffer from three bott... more Though widely used in industry, traditional task-oriented dialogue systems suffer from three bottlenecks: (i) difficult ontology construction (e.g., intents and slots); (ii) poor controllability and interpretability; (iii) annotation-hungry. In this paper, we propose to represent utterance with a simpler concept named Dialogue Action, upon which we construct a tree-structured TaskFlow and further build task-oriented chatbot with TaskFlow as core component. A framework is presented to automatically construct TaskFlow from large-scale dialogues and deploy online. Our experiments on realworld after-sale customer services show TaskFlow can satisfy the major needs, as well as reduce the developer burden effectively. CCS CONCEPTS • Computing methodologies → Discourse, dialogue and pragmatics; • Information systems → Query intent.

arXiv (Cornell University), Jun 27, 2023

With the extensive accumulation of conversational data on the Internet, emotion recognition in co... more With the extensive accumulation of conversational data on the Internet, emotion recognition in conversations (ERC) has received increasing attention. Previous efforts of this task mainly focus on leveraging contextual and speaker-specific features, or integrating heterogeneous external commonsense knowledge. Among them, some heavily rely on future contexts, which, however, are not always available in real-life scenarios. This fact inspires us to generate pseudo future contexts to improve ERC. Specifically, for an utterance, we generate its future context with pre-trained language models, potentially containing extra beneficial knowledge in a conversational form homogeneous with the historical ones. These characteristics make pseudo future contexts easily fused with historical contexts and historical speaker-specific contexts, yielding a conceptually simple framework systematically integrating multicontexts. Experimental results on four ERC datasets demonstrate our method's superiority. Further in-depth analyses reveal that pseudo future contexts can rival real ones to some extent, especially in relatively context-independent conversations.

arXiv (Cornell University), Apr 3, 2023

End-to-end generation-based approaches have been investigated and applied in task-oriented dialog... more End-to-end generation-based approaches have been investigated and applied in task-oriented dialogue systems. However, in industrial scenarios, existing methods face the bottlenecks of reliability (e.g., domain-inconsistent responses, repetition problem, etc) and efficiency (e.g., long computation time, etc). In this paper, we propose a task-oriented dialogue system via action-level generation. Specifically, we first construct dialogue actions from large-scale dialogues and represent each natural language (NL) response as a sequence of dialogue actions. Further, we train a Sequence-to-Sequence model which takes the dialogue history as the input and outputs a sequence of dialogue actions. The generated dialogue actions are transformed into verbal responses. Experimental results show that our light-weighted method achieves competitive performance, and has the advantage of reliability and efficiency. CCS CONCEPTS • Computing methodologies → Discourse, dialogue and pragmatics.

arXiv (Cornell University), Nov 25, 2022

Event detection (ED) identifies and classifies event triggers from unstructured texts, serving as... more Event detection (ED) identifies and classifies event triggers from unstructured texts, serving as a fundamental task for information extraction. Despite the remarkable progress achieved in the past several years, most research efforts focus on detecting events from formal texts (e.g., news articles, Wikipedia documents, financial announcements). Moreover, the texts in each dataset are either from a single source or multiple yet relatively homogeneous sources. With massive amounts of user-generated text accumulating on the Web and inside enterprises, identifying meaningful events in these informal texts, usually from multiple heterogeneous sources, has become a problem of significant practical value. As a pioneering exploration that expands event detection to the scenarios involving informal and heterogeneous texts, we propose a new largescale Chinese event detection dataset based on user reviews, text conversations, and phone conversations in a leading e-commerce platform for food service. We carefully investigate the proposed dataset's textual informality and multi-source heterogeneity characteristics by inspecting data samples quantitatively and qualitatively. Extensive experiments with state-of-the-art event detection methods verify the unique challenges posed by these characteristics, indicating that multi-source informal event detection remains an open problem and requires further efforts. Our benchmark and code are released at https://github.c om/myeclipse/MUSIED.

Proceedings of the 32nd ACM International Conference on Information and Knowledge Management

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing

Event detection (ED) identifies and classifies event triggers from unstructured texts, serving as... more Event detection (ED) identifies and classifies event triggers from unstructured texts, serving as a fundamental task for information extraction. Despite the remarkable progress achieved in the past several years, most research efforts focus on detecting events from formal texts (e.g., news articles, Wikipedia documents, financial announcements). Moreover, the texts in each dataset are either from a single source or multiple yet relatively homogeneous sources. With massive amounts of user-generated text accumulating on the Web and inside enterprises, identifying meaningful events in these informal texts, usually from multiple heterogeneous sources, has become a problem of significant practical value. As a pioneering exploration that expands event detection to the scenarios involving informal and heterogeneous texts, we propose a new large-scale Chinese event detection dataset based on user reviews, text conversations, and phone conversations in a leading e-commerce platform for food service. We carefully investigate the proposed dataset's textual informality and multi-source heterogeneity characteristics by inspecting data samples quantitatively and qualitatively. Extensive experiments with state-of-the-art event detection methods verify the unique challenges posed by these characteristics, indicating that multisource informal event detection remains an open problem and requires further efforts. Our benchmark and code are released at https: //github.com/myeclipse/MUSIED.

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)

Current neural semantic parsers mostly take supervised approaches, which require a considerable a... more Current neural semantic parsers mostly take supervised approaches, which require a considerable amount of expensive training data. As a result, minimizing supervision requirements has been one of the key challenges in semantic parsing. In this paper, we propose a Retrieval as Ambiguous Supervision framework, which can effectively collect high-coverage ambiguous supervisions (i.e., the parse candidates of an utterance) via a pre-trained language modelsbased retrieval system. Then, by assuming candidates will contain the correct ones, the zeroshot task can be converted into an ambiguously supervised task. To improve the precision and coverage of such ambiguous supervision, we propose a confidence-driven self-training algorithm, in which a semantic parser is learned and exploited to disambiguate candidates iteratively. Experimental results show that our approach significantly outperforms the state-of-the-art zero-shot semantic parsing methods.

Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval

End-to-end generation-based approaches have been investigated and applied in task-oriented dialog... more End-to-end generation-based approaches have been investigated and applied in task-oriented dialogue systems. However, in industrial scenarios, existing methods face the bottlenecks of reliability (e.g., domain-inconsistent responses, repetition problem, etc) and efficiency (e.g., long computation time, etc). In this paper, we propose a task-oriented dialogue system via action-level generation. Specifically, we first construct dialogue actions from large-scale dialogues and represent each natural language (NL) response as a sequence of dialogue actions. Further, we train a Sequence-to-Sequence model which takes the dialogue history as the input and outputs a sequence of dialogue actions. The generated dialogue actions are transformed into verbal responses. Experimental results show that our light-weighted method achieves competitive performance, and has the advantage of reliability and efficiency. CCS CONCEPTS • Computing methodologies → Discourse, dialogue and pragmatics.

ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

The CTC model has been widely applied to many application scenarios because of its simple structu... more The CTC model has been widely applied to many application scenarios because of its simple structure, excellent performance, and fast inference speed. There are many peaks in the probability distribution predicted by the CTC models, and each peak represents a non-blank token. The recognition latency of CTC models can be reduced by encouraging the model to predict peaks earlier. Existing methods to reduce latency require modifying the transition relationship between tokens in the forward-backward algorithm, and the gradient calculation. Some of these methods even depend on the forced alignment results provided by other pretrained models. The above methods are complex to implement. To reduce the peak latency, we propose a simple and novel method named peak-first regularization, which utilizes a frame-wise knowledge distillation function to force the probability distribution of the CTC model to shift left along the time axis instead of directly modifying the calculation process of CTC loss and gradients. All the experiments are conducted on a Chinese Mandarin dataset AISHELL-1. We have verified the effectiveness of the proposed regularization on both streaming and non-streaming CTC models respectively. The results show that the proposed method can reduce the average peak latency by about 100 to 200 milliseconds with almost no degradation of recognition accuracy.

Cornell University - arXiv, Dec 6, 2022

Probabilistic linear discriminant analysis (PLDA) is commonly used in speaker verification system... more Probabilistic linear discriminant analysis (PLDA) is commonly used in speaker verification systems to score the similarity of speaker embeddings. Recent studies improved the performance of PLDA in domain-matched conditions by diagonalizing its covariance. We suspect such brutal pruning approach could eliminate its capacity in modeling dimension correlation of speaker embeddings, leading to inadequate performance with domain adaptation. This paper explores two alternative covariance regularization approaches, namely, interpolated PLDA and sparse PLDA, to tackle the problem. The interpolated PLDA incorporates the prior knowledge from cosine scoring to interpolate the covariance of PLDA. The sparse PLDA introduces a sparsity penalty to update the covariance. Experimental results demonstrate that both approaches outperform diagonal regularization noticeably with domain adaptation. In addition, in-domain data can be significantly reduced when training sparse PLDA for domain adaptation.

Cornell University - arXiv, Dec 6, 2022

Very deep models for speaker recognition (SR) have demonstrated remarkable performance improvemen... more Very deep models for speaker recognition (SR) have demonstrated remarkable performance improvement in recent research. However, it is impractical to deploy these models for ondevice applications with constrained computational resources. On the other hand, lightweight models are highly desired in practice despite their sub-optimal performance. This research aims to improve lightweight SR models through large-scale label-free knowledge distillation (KD). Existing KD approaches for SR typically require speaker labels to learn task-specific knowledge, due to the inefficiency of conventional loss for distillation. To address the inefficiency problem and achieve labelfree KD, we propose to employ the contrastive loss from selfsupervised learning for distillation. Extensive experiments are conducted on a collection of public speech datasets from diverse sources. Results on lightweight SR models show that the proposed approach of label-free KD with contrastive loss consistently outperforms both conventional distillation methods and self-supervised learning methods by a significant margin.

Cornell University - arXiv, Nov 6, 2022

The CTC model has been widely applied to many application scenarios because of its simple structu... more The CTC model has been widely applied to many application scenarios because of its simple structure, excellent performance, and fast inference speed. There are many peaks in the probability distribution predicted by the CTC models, and each peak represents a non-blank token. The recognition latency of CTC models can be reduced by encouraging the model to predict peaks earlier. Existing methods to reduce latency require modifying the transition relationship between tokens in the forward-backward algorithm, and the gradient calculation. Some of these methods even depend on the forced alignment results provided by other pretrained models. The above methods are complex to implement. To reduce the peak latency, we propose a simple and novel method named peak-first regularization, which utilizes a frame-wise knowledge distillation function to force the probability distribution of the CTC model to shift left along the time axis instead of directly modifying the calculation process of CTC loss and gradients. All the experiments are conducted on a Chinese Mandarin dataset AISHELL-1. We have verified the effectiveness of the proposed regularization on both streaming and non-streaming CTC models respectively. The results show that the proposed method can reduce the average peak latency by about 100 to 200 milliseconds with almost no degradation of recognition accuracy.

Interspeech 2022

Utilizing text-only data with an external language model (ELM) in end-to-end RNN-Transducer (RNN-... more Utilizing text-only data with an external language model (ELM) in end-to-end RNN-Transducer (RNN-T) for speech recognition is challenging. Recently, a class of methods such as density ratio (DR) and internal language model estimation (ILME) have been developed, outperforming the classic shallow fusion (SF) method. The basic idea behind these methods is that RNN-T posterior should first subtract the implicitly learned internal language model (ILM) prior, in order to integrate the ELM. While recent studies suggest that RNN-T only learns some low-order language model information, the DR method uses a well-trained neural language model with full context, which may be inappropriate for the estimation of ILM and deteriorate the integration performance. Based on the DR method, we propose a loworder density ratio method (LODR) by replacing the estimation with a low-order weak language model. Extensive empirical experiments are conducted on both in-domain and cross-domain scenarios on English LibriSpeech & Tedlium-2 and Chinese WenetSpeech & AISHELL-1 datasets. It is shown that LODR consistently outperforms SF in all tasks, while performing generally close to ILME and better than DR in most tests.

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022)

Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval

Though widely used in industry, traditional task-oriented dialogue systems suffer from three bott... more Though widely used in industry, traditional task-oriented dialogue systems suffer from three bottlenecks: (i) difficult ontology construction (e.g., intents and slots); (ii) poor controllability and interpretability; (iii) annotation-hungry. In this paper, we propose to represent utterance with a simpler concept named Dialogue Action, upon which we construct a tree-structured TaskFlow and further build task-oriented chatbot with TaskFlow as core component. A framework is presented to automatically construct TaskFlow from large-scale dialogues and deploy online. Our experiments on realworld after-sale customer services show TaskFlow can satisfy the major needs, as well as reduce the developer burden effectively. CCS CONCEPTS • Computing methodologies → Discourse, dialogue and pragmatics; • Information systems → Query intent.

Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval

Dialogue topic segmentation is a challenging task in which dialogues are split into segments with... more Dialogue topic segmentation is a challenging task in which dialogues are split into segments with pre-defined topics. Existing works on topic segmentation adopt a two-stage paradigm, including text segmentation and segment labeling. However, such methods tend to focus on the local context in segmentation, and the inter-segment dependency is not well captured. Besides, the ambiguity and labeling noise in dialogue segment bounds bring further challenges to existing models. In this work, we propose the Parallel Extraction Network with Neighbor Smoothing (PEN-NS) to address the above issues. Specifically, we propose the parallel extraction network to perform segment extractions, optimizing the bipartite matching cost of segments to capture inter-segment dependency. Furthermore, we propose neighbor smoothing to handle the segment-bound noise and ambiguity. Experiments on a dialoguebased and a document-based topic segmentation dataset show that PEN-NS outperforms state-the-of-art models significantly. CCS CONCEPTS • Computing methodologies → Discourse, dialogue and pragmatics; Information extraction. * The work was done while the author was an intern at Meituan.

State-of-art speaker verification (SV) systems use a back-end model to score the similarity of sp... more State-of-art speaker verification (SV) systems use a back-end model to score the similarity of speaker embeddings extracted from a neural network model. The commonly used back-end models are the cosine scoring and the probabilistic linear discriminant analysis (PLDA) scoring. With the recently developed neural embeddings, the theoretically more appealing PLDA approach is found to have no advantage against or even be inferior the simple cosine scoring in terms of SV system performance. This paper presents an investigation on the relation between the two scoring approaches, aiming to explain the above counter-intuitive observation. It is shown that the cosine scoring is essentially a special case of PLDA scoring. In other words, by properly setting the parameters of PLDA, the two back-ends become equivalent. As a consequence, the cosine scoring not only inherits the basic assumptions for the PLDA but also introduces additional assumptions on the properties of input embeddings. Experiments show that the dimensional independence assumption required by the cosine scoring contributes most to the performance gap between the two methods under the domain-matched condition. When there is severe domain mismatch and the dimensional independence assumption does not hold, the PLDA would perform better than the cosine for domain adaptation.

Lecture Notes in Computer Science, Dec 31, 2022

arXiv (Cornell University), Mar 30, 2022

Utilizing text-only data with an external language model (ELM) in end-to-end RNN-Transducer (RNN-... more Utilizing text-only data with an external language model (ELM) in end-to-end RNN-Transducer (RNN-T) for speech recognition is challenging. Recently, a class of methods such as density ratio (DR) and internal language model estimation (ILME) have been developed, outperforming the classic shallow fusion (SF) method. The basic idea behind these methods is that RNN-T posterior should first subtract the implicitly learned internal language model (ILM) prior, in order to integrate the ELM. While recent studies suggest that RNN-T only learns some low-order language model information, the DR method uses a well-trained neural language model with full context, which may be inappropriate for the estimation of ILM and deteriorate the integration performance. Based on the DR method, we propose a loworder density ratio method (LODR) by replacing the estimation with a low-order weak language model. Extensive empirical experiments are conducted on both in-domain and cross-domain scenarios on English LibriSpeech & Tedlium-2 and Chinese WenetSpeech & AISHELL-1 datasets. It is shown that LODR consistently outperforms SF in all tasks, while performing generally close to ILME and better than DR in most tests.

Given a long text, the summarization system aims to obtain a shorter highlight while keeping impo... more Given a long text, the summarization system aims to obtain a shorter highlight while keeping important information on the original text. For customer service, the summaries of most dialogues between an agent and a user focus on several fixed key points, such as user's question, user's purpose, the agent's solution, and so on. Traditional extractive methods are difficult to extract all predefined key points exactly. Furthermore, there is a lack of large-scale and high-quality extractive summarization datasets containing key points. In order to solve the above challenges, we propose a Distant Supervision based Machine Reading Comprehension model for extractive Summarization (DSMRC-S). DSMRC-S transforms the summarization task into the machine reading comprehension problem, to fetch key points from the original text exactly according to the predefined questions. In addition, a distant supervision method is proposed to alleviate the lack of eligible extractive summarization datasets. We conduct experiments on a large-scale summarization dataset collected in customer service scenarios, and the results show that the proposed DSMRC-S outperforms the strong baseline methods by 4 points on ROUGE-L.

arXiv (Cornell University), May 12, 2022

Though widely used in industry, traditional task-oriented dialogue systems suffer from three bott... more Though widely used in industry, traditional task-oriented dialogue systems suffer from three bottlenecks: (i) difficult ontology construction (e.g., intents and slots); (ii) poor controllability and interpretability; (iii) annotation-hungry. In this paper, we propose to represent utterance with a simpler concept named Dialogue Action, upon which we construct a tree-structured TaskFlow and further build task-oriented chatbot with TaskFlow as core component. A framework is presented to automatically construct TaskFlow from large-scale dialogues and deploy online. Our experiments on realworld after-sale customer services show TaskFlow can satisfy the major needs, as well as reduce the developer burden effectively. CCS CONCEPTS • Computing methodologies → Discourse, dialogue and pragmatics; • Information systems → Query intent.

arXiv (Cornell University), Jun 27, 2023

With the extensive accumulation of conversational data on the Internet, emotion recognition in co... more With the extensive accumulation of conversational data on the Internet, emotion recognition in conversations (ERC) has received increasing attention. Previous efforts of this task mainly focus on leveraging contextual and speaker-specific features, or integrating heterogeneous external commonsense knowledge. Among them, some heavily rely on future contexts, which, however, are not always available in real-life scenarios. This fact inspires us to generate pseudo future contexts to improve ERC. Specifically, for an utterance, we generate its future context with pre-trained language models, potentially containing extra beneficial knowledge in a conversational form homogeneous with the historical ones. These characteristics make pseudo future contexts easily fused with historical contexts and historical speaker-specific contexts, yielding a conceptually simple framework systematically integrating multicontexts. Experimental results on four ERC datasets demonstrate our method's superiority. Further in-depth analyses reveal that pseudo future contexts can rival real ones to some extent, especially in relatively context-independent conversations.

arXiv (Cornell University), Apr 3, 2023

End-to-end generation-based approaches have been investigated and applied in task-oriented dialog... more End-to-end generation-based approaches have been investigated and applied in task-oriented dialogue systems. However, in industrial scenarios, existing methods face the bottlenecks of reliability (e.g., domain-inconsistent responses, repetition problem, etc) and efficiency (e.g., long computation time, etc). In this paper, we propose a task-oriented dialogue system via action-level generation. Specifically, we first construct dialogue actions from large-scale dialogues and represent each natural language (NL) response as a sequence of dialogue actions. Further, we train a Sequence-to-Sequence model which takes the dialogue history as the input and outputs a sequence of dialogue actions. The generated dialogue actions are transformed into verbal responses. Experimental results show that our light-weighted method achieves competitive performance, and has the advantage of reliability and efficiency. CCS CONCEPTS • Computing methodologies → Discourse, dialogue and pragmatics.

arXiv (Cornell University), Nov 25, 2022

Event detection (ED) identifies and classifies event triggers from unstructured texts, serving as... more Event detection (ED) identifies and classifies event triggers from unstructured texts, serving as a fundamental task for information extraction. Despite the remarkable progress achieved in the past several years, most research efforts focus on detecting events from formal texts (e.g., news articles, Wikipedia documents, financial announcements). Moreover, the texts in each dataset are either from a single source or multiple yet relatively homogeneous sources. With massive amounts of user-generated text accumulating on the Web and inside enterprises, identifying meaningful events in these informal texts, usually from multiple heterogeneous sources, has become a problem of significant practical value. As a pioneering exploration that expands event detection to the scenarios involving informal and heterogeneous texts, we propose a new largescale Chinese event detection dataset based on user reviews, text conversations, and phone conversations in a leading e-commerce platform for food service. We carefully investigate the proposed dataset's textual informality and multi-source heterogeneity characteristics by inspecting data samples quantitatively and qualitatively. Extensive experiments with state-of-the-art event detection methods verify the unique challenges posed by these characteristics, indicating that multi-source informal event detection remains an open problem and requires further efforts. Our benchmark and code are released at https://github.c om/myeclipse/MUSIED.

Proceedings of the 32nd ACM International Conference on Information and Knowledge Management

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing

Event detection (ED) identifies and classifies event triggers from unstructured texts, serving as... more Event detection (ED) identifies and classifies event triggers from unstructured texts, serving as a fundamental task for information extraction. Despite the remarkable progress achieved in the past several years, most research efforts focus on detecting events from formal texts (e.g., news articles, Wikipedia documents, financial announcements). Moreover, the texts in each dataset are either from a single source or multiple yet relatively homogeneous sources. With massive amounts of user-generated text accumulating on the Web and inside enterprises, identifying meaningful events in these informal texts, usually from multiple heterogeneous sources, has become a problem of significant practical value. As a pioneering exploration that expands event detection to the scenarios involving informal and heterogeneous texts, we propose a new large-scale Chinese event detection dataset based on user reviews, text conversations, and phone conversations in a leading e-commerce platform for food service. We carefully investigate the proposed dataset's textual informality and multi-source heterogeneity characteristics by inspecting data samples quantitatively and qualitatively. Extensive experiments with state-of-the-art event detection methods verify the unique challenges posed by these characteristics, indicating that multisource informal event detection remains an open problem and requires further efforts. Our benchmark and code are released at https: //github.com/myeclipse/MUSIED.

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)

Current neural semantic parsers mostly take supervised approaches, which require a considerable a... more Current neural semantic parsers mostly take supervised approaches, which require a considerable amount of expensive training data. As a result, minimizing supervision requirements has been one of the key challenges in semantic parsing. In this paper, we propose a Retrieval as Ambiguous Supervision framework, which can effectively collect high-coverage ambiguous supervisions (i.e., the parse candidates of an utterance) via a pre-trained language modelsbased retrieval system. Then, by assuming candidates will contain the correct ones, the zeroshot task can be converted into an ambiguously supervised task. To improve the precision and coverage of such ambiguous supervision, we propose a confidence-driven self-training algorithm, in which a semantic parser is learned and exploited to disambiguate candidates iteratively. Experimental results show that our approach significantly outperforms the state-of-the-art zero-shot semantic parsing methods.

Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval

End-to-end generation-based approaches have been investigated and applied in task-oriented dialog... more End-to-end generation-based approaches have been investigated and applied in task-oriented dialogue systems. However, in industrial scenarios, existing methods face the bottlenecks of reliability (e.g., domain-inconsistent responses, repetition problem, etc) and efficiency (e.g., long computation time, etc). In this paper, we propose a task-oriented dialogue system via action-level generation. Specifically, we first construct dialogue actions from large-scale dialogues and represent each natural language (NL) response as a sequence of dialogue actions. Further, we train a Sequence-to-Sequence model which takes the dialogue history as the input and outputs a sequence of dialogue actions. The generated dialogue actions are transformed into verbal responses. Experimental results show that our light-weighted method achieves competitive performance, and has the advantage of reliability and efficiency. CCS CONCEPTS • Computing methodologies → Discourse, dialogue and pragmatics.

ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

The CTC model has been widely applied to many application scenarios because of its simple structu... more The CTC model has been widely applied to many application scenarios because of its simple structure, excellent performance, and fast inference speed. There are many peaks in the probability distribution predicted by the CTC models, and each peak represents a non-blank token. The recognition latency of CTC models can be reduced by encouraging the model to predict peaks earlier. Existing methods to reduce latency require modifying the transition relationship between tokens in the forward-backward algorithm, and the gradient calculation. Some of these methods even depend on the forced alignment results provided by other pretrained models. The above methods are complex to implement. To reduce the peak latency, we propose a simple and novel method named peak-first regularization, which utilizes a frame-wise knowledge distillation function to force the probability distribution of the CTC model to shift left along the time axis instead of directly modifying the calculation process of CTC loss and gradients. All the experiments are conducted on a Chinese Mandarin dataset AISHELL-1. We have verified the effectiveness of the proposed regularization on both streaming and non-streaming CTC models respectively. The results show that the proposed method can reduce the average peak latency by about 100 to 200 milliseconds with almost no degradation of recognition accuracy.

Cornell University - arXiv, Dec 6, 2022

Probabilistic linear discriminant analysis (PLDA) is commonly used in speaker verification system... more Probabilistic linear discriminant analysis (PLDA) is commonly used in speaker verification systems to score the similarity of speaker embeddings. Recent studies improved the performance of PLDA in domain-matched conditions by diagonalizing its covariance. We suspect such brutal pruning approach could eliminate its capacity in modeling dimension correlation of speaker embeddings, leading to inadequate performance with domain adaptation. This paper explores two alternative covariance regularization approaches, namely, interpolated PLDA and sparse PLDA, to tackle the problem. The interpolated PLDA incorporates the prior knowledge from cosine scoring to interpolate the covariance of PLDA. The sparse PLDA introduces a sparsity penalty to update the covariance. Experimental results demonstrate that both approaches outperform diagonal regularization noticeably with domain adaptation. In addition, in-domain data can be significantly reduced when training sparse PLDA for domain adaptation.

Cornell University - arXiv, Dec 6, 2022

Very deep models for speaker recognition (SR) have demonstrated remarkable performance improvemen... more Very deep models for speaker recognition (SR) have demonstrated remarkable performance improvement in recent research. However, it is impractical to deploy these models for ondevice applications with constrained computational resources. On the other hand, lightweight models are highly desired in practice despite their sub-optimal performance. This research aims to improve lightweight SR models through large-scale label-free knowledge distillation (KD). Existing KD approaches for SR typically require speaker labels to learn task-specific knowledge, due to the inefficiency of conventional loss for distillation. To address the inefficiency problem and achieve labelfree KD, we propose to employ the contrastive loss from selfsupervised learning for distillation. Extensive experiments are conducted on a collection of public speech datasets from diverse sources. Results on lightweight SR models show that the proposed approach of label-free KD with contrastive loss consistently outperforms both conventional distillation methods and self-supervised learning methods by a significant margin.

Cornell University - arXiv, Nov 6, 2022

The CTC model has been widely applied to many application scenarios because of its simple structu... more The CTC model has been widely applied to many application scenarios because of its simple structure, excellent performance, and fast inference speed. There are many peaks in the probability distribution predicted by the CTC models, and each peak represents a non-blank token. The recognition latency of CTC models can be reduced by encouraging the model to predict peaks earlier. Existing methods to reduce latency require modifying the transition relationship between tokens in the forward-backward algorithm, and the gradient calculation. Some of these methods even depend on the forced alignment results provided by other pretrained models. The above methods are complex to implement. To reduce the peak latency, we propose a simple and novel method named peak-first regularization, which utilizes a frame-wise knowledge distillation function to force the probability distribution of the CTC model to shift left along the time axis instead of directly modifying the calculation process of CTC loss and gradients. All the experiments are conducted on a Chinese Mandarin dataset AISHELL-1. We have verified the effectiveness of the proposed regularization on both streaming and non-streaming CTC models respectively. The results show that the proposed method can reduce the average peak latency by about 100 to 200 milliseconds with almost no degradation of recognition accuracy.

Interspeech 2022

Utilizing text-only data with an external language model (ELM) in end-to-end RNN-Transducer (RNN-... more Utilizing text-only data with an external language model (ELM) in end-to-end RNN-Transducer (RNN-T) for speech recognition is challenging. Recently, a class of methods such as density ratio (DR) and internal language model estimation (ILME) have been developed, outperforming the classic shallow fusion (SF) method. The basic idea behind these methods is that RNN-T posterior should first subtract the implicitly learned internal language model (ILM) prior, in order to integrate the ELM. While recent studies suggest that RNN-T only learns some low-order language model information, the DR method uses a well-trained neural language model with full context, which may be inappropriate for the estimation of ILM and deteriorate the integration performance. Based on the DR method, we propose a loworder density ratio method (LODR) by replacing the estimation with a low-order weak language model. Extensive empirical experiments are conducted on both in-domain and cross-domain scenarios on English LibriSpeech & Tedlium-2 and Chinese WenetSpeech & AISHELL-1 datasets. It is shown that LODR consistently outperforms SF in all tasks, while performing generally close to ILME and better than DR in most tests.

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022)

Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval

Though widely used in industry, traditional task-oriented dialogue systems suffer from three bott... more Though widely used in industry, traditional task-oriented dialogue systems suffer from three bottlenecks: (i) difficult ontology construction (e.g., intents and slots); (ii) poor controllability and interpretability; (iii) annotation-hungry. In this paper, we propose to represent utterance with a simpler concept named Dialogue Action, upon which we construct a tree-structured TaskFlow and further build task-oriented chatbot with TaskFlow as core component. A framework is presented to automatically construct TaskFlow from large-scale dialogues and deploy online. Our experiments on realworld after-sale customer services show TaskFlow can satisfy the major needs, as well as reduce the developer burden effectively. CCS CONCEPTS • Computing methodologies → Discourse, dialogue and pragmatics; • Information systems → Query intent.

Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval

Dialogue topic segmentation is a challenging task in which dialogues are split into segments with... more Dialogue topic segmentation is a challenging task in which dialogues are split into segments with pre-defined topics. Existing works on topic segmentation adopt a two-stage paradigm, including text segmentation and segment labeling. However, such methods tend to focus on the local context in segmentation, and the inter-segment dependency is not well captured. Besides, the ambiguity and labeling noise in dialogue segment bounds bring further challenges to existing models. In this work, we propose the Parallel Extraction Network with Neighbor Smoothing (PEN-NS) to address the above issues. Specifically, we propose the parallel extraction network to perform segment extractions, optimizing the bipartite matching cost of segments to capture inter-segment dependency. Furthermore, we propose neighbor smoothing to handle the segment-bound noise and ambiguity. Experiments on a dialoguebased and a document-based topic segmentation dataset show that PEN-NS outperforms state-the-of-art models significantly. CCS CONCEPTS • Computing methodologies → Discourse, dialogue and pragmatics; Information extraction. * The work was done while the author was an intern at Meituan.

State-of-art speaker verification (SV) systems use a back-end model to score the similarity of sp... more State-of-art speaker verification (SV) systems use a back-end model to score the similarity of speaker embeddings extracted from a neural network model. The commonly used back-end models are the cosine scoring and the probabilistic linear discriminant analysis (PLDA) scoring. With the recently developed neural embeddings, the theoretically more appealing PLDA approach is found to have no advantage against or even be inferior the simple cosine scoring in terms of SV system performance. This paper presents an investigation on the relation between the two scoring approaches, aiming to explain the above counter-intuitive observation. It is shown that the cosine scoring is essentially a special case of PLDA scoring. In other words, by properly setting the parameters of PLDA, the two back-ends become equivalent. As a consequence, the cosine scoring not only inherits the basic assumptions for the PLDA but also introduces additional assumptions on the properties of input embeddings. Experiments show that the dimensional independence assumption required by the cosine scoring contributes most to the performance gap between the two methods under the domain-matched condition. When there is severe domain mismatch and the dimensional independence assumption does not hold, the PLDA would perform better than the cosine for domain adaptation.