Jiahuan Pei

Wentao Deng, Jiahuan Pei, Zhaochun Ren et al.

Answer selection in open-domain dialogues aims to select an accurate answer from candidates. Recent success of answer selection models hinges on training with large amounts of labeled data. However, collecting large-scale labeled data is labor-intensive and time-consuming. In this paper, we introduce the predicted intent labels to calibrate answer labels in a self-training paradigm. Specifically, we propose the intent-calibrated self-training (ICAST) to improve the quality of pseudo answer labels through the intent-calibrated answer selection paradigm, in which we employ pseudo intent labels to help improve pseudo answer labels. We carry out extensive experiments on two benchmark datasets with open-domain dialogues. The experimental results show that ICAST outperforms baselines consistently with 1%, 5% and 10% labeled data. Specifically, it improves 2.06% and 1.00% of F1 score on the two datasets, compared with the strongest baseline with only 5% labeled data.

Xueqing Peng, Zhuohan Xie, Yupeng Cao et al.

As AI agents improve, the central question is no longer whether they can solve isolated well-defined financial tasks, but whether they can reliably carry out financial professional work. Existing financial benchmarks offer only a partial view of this ability, as they primarily evaluate static competencies such as question answering, retrieval, summarization, and classification. We introduce Herculean, the first skilled benchmark for agentic financial intelligence spanning four representative workflows, including Trading, Hedging, Market Insights, and Auditing. Each workflow is instantiated as a standardized MCP-based skill environment with its own tools, interaction dynamics, constraints, and success criteria, enabling consistent end-to-end assessment of heterogeneous agent systems. Across frontier agents, we find agents perform relatively well on Trading and Market Insights, but struggle substantially on Hedging and Auditing, where long-horizon coordination, state consistency, and structured verification are critical. Overall, our results point to a key gap in current agents in turning financial reasoning into dependable workflow execution in high-stakes financial workflows.

Weijia Zhang, Mohammad Aliannejadi, Jiahuan Pei et al.

Large language models (LLMs) often generate content with unsupported or unverifiable content, known as "hallucinations." To address this, retrieval-augmented LLMs are employed to include citations in their content, grounding the content in verifiable sources. Despite such developments, manually assessing how well a citation supports the associated statement remains a major challenge. Previous studies tackle this challenge by leveraging faithfulness metrics to estimate citation support automatically. However, they limit this citation support estimation to a binary classification scenario, neglecting fine-grained citation support in practical scenarios. To investigate the effectiveness of faithfulness metrics in fine-grained scenarios, we propose a comparative evaluation framework that assesses the metric effectiveness in distinguishing citations between three-category support levels: full, partial, and no support. Our framework employs correlation analysis, classification evaluation, and retrieval evaluation to measure the alignment between metric scores and human judgments comprehensively. Our results indicate no single metric consistently excels across all evaluations, highlighting the complexity of accurately evaluating fine-grained support levels. Particularly, we find that the best-performing metrics struggle to distinguish partial support from full or no support. Based on these findings, we provide practical recommendations for developing more effective metrics.

Xin Sun, Yue Su, Yifan Mo et al.

Building trustworthy AI systems for mental health support is a shared priority across stakeholders from multiple disciplines. However, "trustworthy" remains loosely defined and inconsistently operationalized. AI research often focuses on technical criteria (e.g., robustness, explainability, and safety), while therapeutic practitioners emphasize therapeutic fidelity (e.g., appropriateness, empathy, and long-term user outcomes). To bridge the fragmented landscape, we propose a three-layer trust framework, covering human-oriented, AI-oriented, and interaction-oriented trust, integrating the viewpoints of key stakeholders (e.g., practitioners, researchers, regulators). Using this framework, we systematically review existing AI-driven research in mental health domain and examine evaluation practices for ``trustworthy'' ranging from automatic metrics to clinically validated approaches. We highlight critical gaps between what NLP currently measures and what real-world mental health contexts require, and outline a research agenda for building socio-technically aligned and genuinely trustworthy AI for mental health support.

Xin Sun, Xiao Tang, Abdallah El Ali et al.

Recent advancements in large language models (LLMs) have shown promise in generating psychotherapeutic dialogues, particularly in the context of motivational interviewing (MI). However, the inherent lack of transparency in LLM outputs presents significant challenges given the sensitive nature of psychotherapy. Applying MI strategies, a set of MI skills, to generate more controllable therapeutic-adherent conversations with explainability provides a possible solution. In this work, we explore the alignment of LLMs with MI strategies by first prompting the LLMs to predict the appropriate strategies as reasoning and then utilizing these strategies to guide the subsequent dialogue generation. We seek to investigate whether such alignment leads to more controllable and explainable generations. Multiple experiments including automatic and human evaluations are conducted to validate the effectiveness of MI strategies in aligning psychotherapy dialogue generation. Our findings demonstrate the potential of LLMs in producing strategically aligned dialogues and suggest directions for practical applications in psychotherapeutic settings.

Kai Golan Hashiloni, Daniel Fadlon, Lior Livyatan et al.

Idioms pose a fundamental challenge for language models, as their meaning cannot be inferred from surface form alone. Understanding such expressions, therefore, requires semantic abstraction beyond lexical overlap. We introduce IdioLink, a retrieval benchmark designed to test whether models can link idiomatic expressions to conceptually equivalent meanings expressed in literal or paraphrased forms. IdioLink comprises 10,700 documents and 2,140 queries, spanning 107 idioms with both literal and figurative uses. Each document and query is annotated with spans that convey the core meaning. Evaluating strong embedding baselines (e.g., BGE, E5, Contriever, and Qwen), we show that current models struggle to retrieve equivalent meanings across divergent surface realizations, relying instead on topical and shallow semantic cues. IdioLink exposes key gaps in idiom-aware semantic retrieval and provides a challenging testbed for future models.

Qingyu Meng, Min Chen, Dingming Liu et al.

Large language models (LLMs) can generate fluent dialogue, but prior works lack situational grounding, dynamic strategy control, and evaluation aligned with clinical standards in motivational interviewing (MI). We introduce StoryMI, a multi-LLM agent framework for controllable MI dialogue generation, where questionnaire-based client profiles are expanded into situational stories that provide narrative context for the dialogue. Therapist and client agents generate MI-coded utterances guided by MI codes selected by the interaction agent, while an interaction agent dynamically coordinates exchanges to control MI strategies during a multi-turn conversation. We propose a two-level evaluation protocol: lexical metrics and MI-specific measures of macro-level counseling strategies, alongside LLM-as-judge and human expert assessments. We construct a dataset of 6K simulated MI dialogues grounded in 1K questionnaire-story pairs, covering 12 MI codes and 13 symptom domains, and benchmark six open- and closed-source LLMs. Our results show that situational grounding and macro-level control can improve MI adherence and clinical plausibility, demonstrating the effectiveness of a structured multi-agent workflow for psychotherapy dialogue generation. We provide code and data for reproducibility.

Jiahuan Pei, Fanghua Ye, Xin Sun et al.

Large language models (LLMs) have advanced virtual educators and learners, bridging NLP with AI4Education. Existing work often lacks scalability and fails to leverage diverse, large-scale course content, with limited frameworks for assessing pedagogic quality. To this end, we propose WikiHowAgent, a multi-agent workflow leveraging LLMs to simulate interactive teaching-learning conversations. It integrates teacher and learner agents, an interaction manager, and an evaluator to facilitate procedural learning and assess pedagogic quality. We introduce a dataset of 114,296 teacher-learner conversations grounded in 14,287 tutorials across 17 domains and 727 topics. Our evaluation protocol combines computational and rubric-based metrics with human judgment alignment. Results demonstrate the workflow's effectiveness in diverse setups, offering insights into LLM capabilities across domains. Our datasets and implementations are fully open-sourced.

Zhiqi Gao, Guo Zhu, Huarui Luo et al.

Standardized patients (SPs) play a central role in clinical communication training but are costly, difficult to scale, and inconsistent. Large language model (LLM) based AI standardized patients (AI-SPs) promise flexible, on-demand practice, yet learners often report that they talk like a patient but feel different. We interviewed 12 clinical-year medical students and conducted three co-design workshops to examine how learners experience constraints of SP encounters and what they expect from AI-SPs. We identified six learner-centered needs, translated them into AI-SP design requirements, and synthesized a conceptual workflow. Our findings position AI-SPs as tools for deliberate practice and show that instructional usability, rather than conversational realism alone, drives learner trust, engagement, and educational value.

Wentao Deng, Jiahuan Pei, Zhiwei Xu et al.

A multi-agent system (MAS) enhances its capacity to solve complex natural language processing (NLP) tasks through collaboration among multiple agents, where consensus-seeking serves as a fundamental mechanism. However, existing consensus-seeking approaches typically rely on voting mechanisms to judge consensus, overlooking contradictions in system-internal beliefs that destabilize the consensus. Moreover, these methods often involve agents updating their results through indiscriminate collaboration with every other agent. Such uniform interaction fails to identify the optimal collaborators for each agent, hindering the emergence of a stable consensus. To address these challenges, we provide a theoretical framework for selecting optimal collaborators that maximize consensus stability. Based on the theorems, we propose the Belief-Calibrated Consensus Seeking (BCCS) framework to facilitate stable consensus via selecting optimal collaborators and calibrating the consensus judgment by system-internal beliefs. Experimental results on the MATH and MMLU benchmark datasets demonstrate that the proposed BCCS framework outperforms the best existing results by 2.23% and 3.95% of accuracy on challenging tasks, respectively. Our code and data are available at https://github.com/dengwentao99/BCCS.

Xiao Wang, Jiahuan Pei, Diancheng Shui et al.

Legal judgment prediction offers a compelling method to aid legal practitioners and researchers. However, the research question remains relatively under-explored: Should multiple defendants and charges be treated separately in LJP? To address this, we introduce a new dataset namely multi-person multi-charge prediction (MPMCP), and seek the answer by evaluating the performance of several prevailing legal large language models (LLMs) on four practical legal judgment scenarios: (S1) single defendant with a single charge, (S2) single defendant with multiple charges, (S3) multiple defendants with a single charge, and (S4) multiple defendants with multiple charges. We evaluate the dataset across two LJP tasks, i.e., charge prediction and penalty term prediction. We have conducted extensive experiments and found that the scenario involving multiple defendants and multiple charges (S4) poses the greatest challenges, followed by S2, S3, and S1. The impact varies significantly depending on the model. For example, in S4 compared to S1, InternLM2 achieves approximately 4.5% lower F1-score and 2.8% higher LogD, while Lawformer demonstrates around 19.7% lower F1-score and 19.0% higher LogD. Our dataset and code are available at https://github.com/lololo-xiao/MultiJustice-MPMCP.

Jiahuan Pei, Irene Viola, Haochen Huang et al.

Autonomous artificial intelligence (AI) agents have emerged as promising protocols for automatically understanding the language-based environment, particularly with the exponential development of large language models (LLMs). However, a fine-grained, comprehensive understanding of multimodal environments remains under-explored. This work designs an autonomous workflow tailored for integrating AI agents seamlessly into extended reality (XR) applications for fine-grained training. We present a demonstration of a multimodal fine-grained training assistant for LEGO brick assembly in a pilot XR environment. Specifically, we design a cerebral language agent that integrates LLM with memory, planning, and interaction with XR tools and a vision-language agent, enabling agents to decide their actions based on past experiences. Furthermore, we introduce LEGO-MRTA, a multimodal fine-grained assembly dialogue dataset synthesized automatically in the workflow served by a commercial LLM. This dataset comprises multimodal instruction manuals, conversations, XR responses, and vision question answering. Last, we present several prevailing open-resource LLMs as benchmarks, assessing their performance with and without fine-tuning on the proposed dataset. We anticipate that the broader impact of this workflow will advance the development of smarter assistants for seamless user interaction in XR environments, fostering research in both AI and HCI communities.

Pengjie Ren, Chengshun Shi, Shiguang Wu et al.

Parameter-efficient fine-tuning (PEFT) is a popular method for tailoring pre-trained large language models (LLMs), especially as the models' scale and the diversity of tasks increase. Low-rank adaptation (LoRA) is based on the idea that the adaptation process is intrinsically low-dimensional, i.e., significant model changes can be represented with relatively few parameters. However, decreasing the rank encounters challenges with generalization errors for specific tasks when compared to full-parameter fine-tuning. We present MELoRA, a mini-ensemble low-rank adapters that uses fewer trainable parameters while maintaining a higher rank, thereby offering improved performance potential. The core idea is to freeze original pretrained weights and train a group of mini LoRAs with only a small number of parameters. This can capture a significant degree of diversity among mini LoRAs, thus promoting better generalization ability. We conduct a theoretical analysis and empirical studies on various NLP tasks. Our experimental results show that, compared to LoRA, MELoRA achieves better performance with 8 times fewer trainable parameters on natural language understanding tasks and 36 times fewer trainable parameters on instruction following tasks, which demonstrates the effectiveness of MELoRA.

Zhiqi Gao, Huarui Luo, Guo Zhu et al.

AI standardized patients are becoming a setting for professional training in clinical consultation. This paper presents MedEasy, a multi-agent system that organizes virtual-patient practice through patient dialogue, clinical actions, decision submission, documentation, and feedback. We first conducted a formative study with 12 clinical-year medical students through interviews and three co-design workshops. The findings informed a staged workflow, structured case records, action-contingent findings, and trajectory-based review. We then conducted an evaluative user study with a separate cohort of 12 clinical-year medical students, with each participant completing two counterbalanced cases. Learners interpreted MedEasy as a connected consultation environment. They used patient responses, examination findings, available actions, and feedback together to judge whether the represented case remained coherent. They valued repeatable practice and recorded review, while questioning missing actions and feedback criteria. The paper contributes design implications for AI-supported professional training systems that use case-specific standards to connect situated practice.

Yifan Mo, Xiao Fu, Yue Su et al.

This work investigates the ability of large language models (LLMs) to generate mathematical equations from scientific texts. Prior work faces challenges in unstructured grounding, multi-equation dependency, and humanaligned evaluation. To this end, we construct a dataset of AI research papers, pairing contextual passages with ground-truth equations and variable descriptions. We develop an explainable equation generation workflow and evaluate it across diverse open- and closed-source LLM backbones. We introduce an evaluation protocol combining automatic metrics, LLM-based rubrics, and human judgments to assess accuracy, explainability, and human-LLM alignment. Results indicate that LLMs perform moderately on lexical- and syntactic-based similarity, while struggling with semantic accuracy. Comparisons between LLM-based evaluations and human judgments reveal limited alignment, highlighting challenges in using LLMs to assess equation quality. These findings offer insights for improving equation generation models and developing more reliable evaluation methods for scientific text. We provide code and data for reproducibility.

Xin Sun, Jan de Wit, Zhuying Li et al.

Chatbots or conversational agents (CAs) are increasingly used to improve access to digital psychotherapy. Many current systems rely on rigid, rule-based designs, heavily dependent on expert-crafted dialogue scripts for guiding therapeutic conversations. Although advances in large language models (LLMs) offer potential for more flexible interactions, their lack of controllability and explanability poses challenges in high-stakes contexts like psychotherapy. To address this, we conducted two studies in this work to explore how aligning LLMs with expert-crafted scripts can enhance psychotherapeutic chatbot performance. In Study 1 (N=43), an online experiment with a within-subjects design, we compared rule-based, pure LLM, and LLMs aligned with expert-crafted scripts via fine-tuning and prompting. Results showed that aligned LLMs significantly outperformed the other types of chatbots in empathy, dialogue relevance, and adherence to therapeutic principles. Building on findings, we proposed ``Script-Strategy Aligned Generation (SSAG)'', a more flexible alignment approach that reduces reliance on fully scripted content while maintaining LLMs' therapeutic adherence and controllability. In a 10-day field Study 2 (N=21), SSAG achieved comparable therapeutic effectiveness to full-scripted LLMs while requiring less than 40\% of expert-crafted dialogue content. Beyond these results, this work advances LLM applications in psychotherapy by providing a controllable and scalable solution, reducing reliance on expert effort. By enabling domain experts to align LLMs through high-level strategies rather than full scripts, SSAG supports more efficient co-development and expands access to a broader context of psychotherapy.

Albert Tang, Yifan Mo, Jie Li et al.

The double empathy problem frames communication difficulties between neurodivergent and neurotypical individuals as arising from mutual misunderstanding, yet most interventions focus on autistic individuals. We present NeuroWise, a multi-agent LLM-based coaching system that supports neurotypical users through stress visualization, interpretation of internal experiences, and contextual guidance. In a between-subjects study (N=30), NeuroWise was rated as helpful by all participants and showed a significant condition-time effect on deficit-based attributions (p=0.02): NeuroWise users reduced deficit framing, while baseline users shifted toward blaming autistic "deficits" after difficult interactions. NeuroWise users also completed conversations more efficiently (37% fewer turns, p=0.03). These findings suggest that AI-based interpretation can support attributional change by helping users recognize communication challenges as mutual.

Haochen Huang, Jiahuan Pei, Mohammad Aliannejadi et al.

Vision-language models (VLMs) are facing the challenges of understanding and following multimodal assembly instructions, particularly when fine-grained spatial reasoning and precise object state detection are required. In this work, we explore LEGO Co-builder, a hybrid benchmark combining real-world LEGO assembly logic with programmatically generated multimodal scenes. The dataset captures stepwise visual states and procedural instructions, allowing controlled evaluation of instruction-following, object detection, and state detection. We introduce a unified framework and assess leading VLMs such as GPT-4o, Gemini, and Qwen-VL, under zero-shot and fine-tuned settings. Our results reveal that even advanced models like GPT-4o struggle with fine-grained assembly tasks, with a maximum F1 score of just 40.54\% on state detection, highlighting gaps in fine-grained visual understanding. We release the benchmark, codebase, and generation pipeline to support future research on multimodal assembly assistants grounded in real-world workflows.

Xiaoyu Shen, Rexhina Blloshmi, Dawei Zhu et al.

Retrieval-augmented generation has gained popularity as a framework to enhance large language models with external knowledge. However, its effectiveness hinges on the retrieval robustness of the model. If the model lacks retrieval robustness, its performance is constrained by the accuracy of the retriever, resulting in significant compromises when the retrieved context is irrelevant. In this paper, we evaluate the "implicit" retrieval robustness of various large language models, instructing them to directly output the final answer without explicitly judging the relevance of the retrieved context. Our findings reveal that fine-tuning on a mix of gold and distracting context significantly enhances the model's robustness to retrieval inaccuracies, while still maintaining its ability to extract correct answers when retrieval is accurate. This suggests that large language models can implicitly handle relevant or irrelevant retrieved context by learning solely from the supervision of the final answer in an end-to-end manner. Introducing an additional process for explicit relevance judgment can be unnecessary and disrupts the end-to-end approach.

Weijia Zhang, Mohammad Aliannejadi, Yifei Yuan et al.

Large language models (LLMs) often produce unsupported or unverifiable content, known as "hallucinations." To mitigate this, retrieval-augmented LLMs incorporate citations, grounding the content in verifiable sources. Despite such developments, manually assessing how well a citation supports the associated statement remains a major challenge. Previous studies use faithfulness metrics to estimate citation support automatically but are limited to binary classification, overlooking fine-grained citation support in practical scenarios. To investigate the effectiveness of faithfulness metrics in fine-grained scenarios, we propose a comparative evaluation framework that assesses the metric effectiveness in distinguishing citations between three-category support levels: full, partial, and no support. Our framework employs correlation analysis, classification evaluation, and retrieval evaluation to measure the alignment between metric scores and human judgments comprehensively. Our results show no single metric consistently excels across all evaluations, revealing the complexity of assessing fine-grained support. Based on the findings, we provide practical recommendations for developing more effective metrics.

Jiahuan Pei, Cheng Wang, György Szarvas

Transformers are state-of-the-art in a wide range of NLP tasks and have also been applied to many real-world products. Understanding the reliability and certainty of transformer model predictions is crucial for building trustable machine learning applications, e.g., medical diagnosis. Although many recent transformer extensions have been proposed, the study of the uncertainty estimation of transformer models is under-explored. In this work, we propose a novel way to enable transformers to have the capability of uncertainty estimation and, meanwhile, retain the original predictive performance. This is achieved by learning a hierarchical stochastic self-attention that attends to values and a set of learnable centroids, respectively. Then new attention heads are formed with a mixture of sampled centroids using the Gumbel-Softmax trick. We theoretically show that the self-attention approximation by sampling from a Gumbel distribution is upper bounded. We empirically evaluate our model on two text classification tasks with both in-domain (ID) and out-of-domain (OOD) datasets. The experimental results demonstrate that our approach: (1) achieves the best predictive performance and uncertainty trade-off among compared methods; (2) exhibits very competitive (in most cases, improved) predictive performance on ID datasets; (3) is on par with Monte Carlo dropout and ensemble methods in uncertainty estimation on OOD datasets.

Benyou Wang, Qianqian Xie, Jiahuan Pei et al.

Pre-trained language models (PLMs) have been the de facto paradigm for most natural language processing (NLP) tasks. This also benefits biomedical domain: researchers from informatics, medicine, and computer science (CS) communities propose various PLMs trained on biomedical datasets, e.g., biomedical text, electronic health records, protein, and DNA sequences for various biomedical tasks. However, the cross-discipline characteristics of biomedical PLMs hinder their spreading among communities; some existing works are isolated from each other without comprehensive comparison and discussions. It expects a survey that not only systematically reviews recent advances of biomedical PLMs and their applications but also standardizes terminology and benchmarks. In this paper, we summarize the recent progress of pre-trained language models in the biomedical domain and their applications in biomedical downstream tasks. Particularly, we discuss the motivations and propose a taxonomy of existing biomedical PLMs. Their applications in biomedical downstream tasks are exhaustively discussed. At last, we illustrate various limitations and future trends, which we hope can provide inspiration for the future research of the research community.

Guojun Yan, Jiahuan Pei, Pengjie Ren et al.

Medical dialogue systems (MDSs) aim to assist doctors and patients with a range of professional medical services, i.e., diagnosis, treatment and consultation. The development of MDSs is hindered because of a lack of resources. In particular. (1) there is no dataset with large-scale medical dialogues that covers multiple medical services and contains fine-grained medical labels (i.e., intents, actions, slots, values), and (2) there is no set of established benchmarks for MDSs for multi-domain, multi-service medical dialogues. In this paper, we present ReMeDi, a set of resource for medical dialogues. ReMeDi consists of two parts, the ReMeDi dataset and the ReMeDi benchmarks. The ReMeDi dataset contains 96,965 conversations between doctors and patients, including 1,557 conversations with fine-gained labels. It covers 843 types of diseases, 5,228 medical entities, and 3 specialties of medical services across 40 domains. To the best of our knowledge, the ReMeDi dataset is the only medical dialogue dataset that covers multiple domains and services, and has fine-grained medical labels. The second part of the ReMeDi resources consists of a set of state-of-the-art models for (medical) dialogue generation. The ReMeDi benchmark has the following methods: (1) pretrained models (i.e., BERT-WWM, BERT-MED, GPT2, and MT5) trained, validated, and tested on the ReMeDi dataset, and (2) a self-supervised contrastive learning(SCL) method to expand the ReMeDi dataset and enhance the training of the state-of-the-art pretrained models. We describe the creation of the ReMeDi dataset, the ReMeDi benchmarking methods, and establish experimental results using the ReMeDi benchmarking methods on the ReMeDi dataset for future research to compare against. With this paper, we share the dataset, implementations of the benchmarks, and evaluation scripts.

Jiahuan Pei, Pengjie Ren, Christof Monz et al.

Dialogue response generation (DRG) is a critical component of task-oriented dialogue systems (TDSs). Its purpose is to generate proper natural language responses given some context, e.g., historical utterances, system states, etc. State-of-the-art work focuses on how to better tackle DRG in an end-to-end way. Typically, such studies assume that each token is drawn from a single distribution over the output vocabulary, which may not always be optimal. Responses vary greatly with different intents, e.g., domains, system actions. We propose a novel mixture-of-generators network (MoGNet) for DRG, where we assume that each token of a response is drawn from a mixture of distributions. MoGNet consists of a chair generator and several expert generators. Each expert is specialized for DRG w.r.t. a particular intent. The chair coordinates multiple experts and combines the output they have generated to produce more appropriate responses. We propose two strategies to help the chair make better decisions, namely, a retrospective mixture-of-generators (RMoG) and prospective mixture-of-generators (PMoG). The former only considers the historical expert-generated responses until the current time step while the latter also considers possible expert-generated responses in the future by encouraging exploration. In order to differentiate experts, we also devise a global-and-local (GL) learning scheme that forces each expert to be specialized towards a particular intent using a local loss and trains the chair and all experts to coordinate using a global loss. We carry out extensive experiments on the MultiWOZ benchmark dataset. MoGNet significantly outperforms state-of-the-art methods in terms of both automatic and human evaluations, demonstrating its effectiveness for DRG.

Jiahuan Pei, Pengjie Ren, Maarten de Rijke

End-to-end Task-oriented Dialogue Systems (TDSs) have attracted a lot of attention for their superiority (e.g., in terms of global optimization) over pipeline modularized TDSs. Previous studies on end-to-end TDSs use a single-module model to generate responses for complex dialogue contexts. However, no model consistently outperforms the others in all cases. We propose a neural Modular Task-oriented Dialogue System(MTDS) framework, in which a few expert bots are combined to generate the response for a given dialogue context. MTDS consists of a chair bot and several expert bots. Each expert bot is specialized for a particular situation, e.g., one domain, one type of action of a system, etc. The chair bot coordinates multiple expert bots and adaptively selects an expert bot to generate the appropriate response. We further propose a Token-level Mixture-of-Expert (TokenMoE) model to implement MTDS, where the expert bots predict multiple tokens at each timestamp and the chair bot determines the final generated token by fully taking into consideration the outputs of all expert bots. Both the chair bot and the expert bots are jointly trained in an end-to-end fashion. To verify the effectiveness of TokenMoE, we carry out extensive experiments on a benchmark dataset. Compared with the baseline using a single-module model, our TokenMoE improves the performance by 8.1% of inform rate and 0.8% of success rate.