The latency-quality tradeoff is a fundamental constraint in open-domain dialogue AI systems, since comprehensive knowledge access necessitates prohibitive response delays. Contemporary approaches offer two inadequate solutions: lightweight instruct models achieve sub-second latency but lack reasoning depth, while tool-augmented ReAct agents enhance factuality through external knowledge at the cost of synchronous execution that blocks interaction during retrieval processes. PMFR is thus proposed, with a temporal decoupling framework that fundamentally resolves the contradiction through asynchronous knowledge orchestration. PMFR employs three coordinated components: (1) a Knowledge Adequacy Evaluator for real-time sufficiency assessment, (2) a Lightweight Response Generator for immediate user interaction, and (3) an Asynchronous Knowledge Refinement Agent for background knowledge enhancement. This architecture maintains continuous conversational flow while progressively enriching knowledge coverage through intelligent triggering mechanisms. Evaluation results on TopiOCQA demonstrate PMFR outperforms brute-force scaling: PMFR achieves 95.3% latency reduction (23.38s -> 1.09s) while preserving response quality comparable to heavyweight synchronous baselines (GEval-C: 0.613 vs. 0.620).

We study the limitations of Large Language Models (LLMs) for the task of response generation in human-machine dialogue. Several techniques have been proposed in the literature for different dialogue types (e.g., Open-Domain). However, the evaluations of these techniques have been limited in terms of base LLMs, dialogue types and evaluation metrics. In this work, we extensively analyze different LLM adaptation techniques when applied to different dialogue types. We have selected two base LLMs, Llama-2 and Mistral, and four dialogue types Open-Domain, Knowledge-Grounded, Task-Oriented, and Question Answering. We evaluate the performance of in-context learning and fine-tuning techniques across datasets selected for each dialogue type. We assess the impact of incorporating external knowledge to ground the generation in both scenarios of Retrieval-Augmented Generation (RAG) and gold knowledge. We adopt consistent evaluation and explainability criteria for automatic metrics and human evaluation protocols. Our analysis shows that there is no universal best-technique for adapting large language models as the efficacy of each technique depends on both the base LLM and the specific type of dialogue. Last but not least, the assessment of the best adaptation technique should include human evaluation to avoid false expectations and outcomes derived from automatic metrics.

In this paper we propose a study of linguistic portability strategies of large pre-trained language models (PLMs) used for open-domain dialogue systems in a high-resource language for this task. In particular the target low-resource language (L_T) will be simulated with French, as it lacks of task-specific resources and allows our human evaluation, when the source language (L_S) is English. For obvious reasons, recent works using such models for open-domain dialogue are mostly developed in English. Yet building specific PLMs for each possible target language supposes collecting new datasets and is costly. For this reason, trying to leverage all existing resources (PLMs and data) in both L_S and L_T , we wish to assess the performance achievable in L_T with different approaches. The first two approaches evaluate the usage of Neural Machine Translation (NMT) at different levels: TrainOnTarget where a L_S dataset is translated before fine-tuning in L_T and TestOnSource where a L_S model is coupled with NMT modules during inference. Then, the advent of BLOOM [2], the world first open-access multilingual large PLM, allow researchers to develop new approaches aiming to leverage not only the model's full accessibility but also its multilingualism and translation abilities. In this context the task is learned in L_S first and adapted to L_T using the MAD-X Adapter architecture [16]. In the two sets of experiments models are evaluated in spoken dialogue conditions with human and the strategies can be compared in terms of perceived interaction quality.

The long-standing one-to-many problem of gold standard responses in open-domain dialogue systems presents challenges for automatic evaluation metrics. Though prior works have demonstrated some success by applying powerful Large Language Models (LLMs), existing approaches still struggle with the one-to-many problem, and exhibit subpar performance in domain-specific scenarios. We assume the commonsense reasoning biases within LLMs may hinder their performance in domainspecific evaluations. To address both issues, we propose a novel framework SLIDE (Small and Large Integrated for Dialogue Evaluation), that leverages both a small, specialised model (SLM), and LLMs for the evaluation of open domain dialogues. Our approach introduces several techniques: (1) Contrastive learning to differentiate between robust and non-robust response embeddings; (2) A novel metric for semantic sensitivity that combines embedding cosine distances with similarity learned through neural networks, and (3) a strategy for incorporating the evaluation results from both the SLM and LLMs. Our empirical results demonstrate that our approach achieves state-of-the-art performance in both the classification and evaluation tasks, and additionally the SLIDE evaluator exhibits better correlation with human judgements. Our code is available at https:// github.com/hegehongcha/SLIDE-ACL2024.

Most existing dialogue corpora and models have been designed to fit into 2 predominant categories : task-oriented dialogues portray functional goals, such as making a restaurant reservation or booking a plane ticket, while chit-chat/open-domain dialogues focus on holding a socially engaging talk with a user. However, humans tend to seamlessly switch between modes and even use chitchat to enhance task-oriented conversations. To bridge this gap, new datasets have recently been created, blending both communication modes into conversation examples. The approaches used tend to rely on adding chit-chat snippets to pre-existing, human-generated task-oriented datasets. Given the tendencies observed in humans, we wonder however if the latter do not \textit{already} hold chit-chat sequences. By using topic modeling and searching for topics which are most similar to a set of keywords related to social talk, we explore the training sets of Schema-Guided Dialogues and MultiWOZ. Our study shows that sequences related to social talk are indeed naturally present, motivating further research on ways chitchat is combined into task-oriented dialogues.

In Natural Language Generation (NLG) tasks, for any input, multiple communicative goals are plausible, and any goal can be put into words, or produced, in multiple ways. We characterise the extent to which human production varies lexically, syntactically, and semantically across four NLG tasks, connecting human production variability to aleatoric or data uncertainty. We then inspect the space of output strings shaped by a generation system's predicted probability distribution and decoding algorithm to probe its uncertainty. For each test input, we measure the generator's calibration to human production variability. Following this instance-level approach, we analyse NLG models and decoding strategies, demonstrating that probing a generator with multiple samples and, when possible, multiple references, provides the level of detail necessary to gain understanding of a model's representation of uncertainty. Code available at https://github.com/dmg-illc/nlg-uncertainty-probes.

Recent work suggests that interpolating between the weights of two specialized language models can transfer knowledge between tasks in a way that multi-task learning cannot. However, very few have explored interpolation between more than two models, where each has a distinct knowledge base. In this paper, we introduce Derivative Free Weight-space Ensembling (DFWE), a new few-sample task transfer approach for open-domain dialogue. Our framework creates a set of diverse expert language models trained using a predefined set of source tasks. Next, we finetune each of the expert models on the target task, approaching the target task from several distinct knowledge bases. Finally, we linearly interpolate between the model weights using a gradient-free-optimization algorithm, to efficiently find a good interpolation weighting. We demonstrate the effectiveness of the method on FETA-Friends outperforming the standard pretrain-finetune approach.

With the increasing popularity of online chatting, stickers are becoming important in our online communication. Selecting appropriate stickers in open-domain dialogue requires a comprehensive understanding of both dialogues and stickers, as well as the relationship between the two types of modalities. To tackle these challenges, we propose a multitask learning method comprised of three auxiliary tasks to enhance the understanding of dialogue history, emotion and semantic meaning of stickers. Extensive experiments conducted on a recent challenging dataset show that our model can better combine the multimodal information and achieve significantly higher accuracy over strong baselines. Ablation study further verifies the effectiveness of each auxiliary task. Our code is available at \url{https://github.com/nonstopfor/Sticker-Selection}

There has recently been an explosion of work on spoken dialogue systems, along with an increased interest in open-domain systems that engage in casual conversations on popular topics such as movies, books and music. These systems aim to socially engage, entertain, and even empathize with their users. Since the achievement of such social goals is hard to measure, recent research has used dialogue length or human ratings as evaluation metrics, and developed methods for automatically calculating novel metrics, such as coherence, consistency, relevance and engagement. Here we develop a PARADISE model for predicting the performance of Athena, a dialogue system that has participated in thousands of conversations with real users, while competing as a finalist in the Alexa Prize. We use both user ratings and dialogue length as metrics for dialogue quality, and experiment with predicting these metrics using automatic features that are both system dependent and independent. Our goal is to learn a general objective function that can be used to optimize the dialogue choices of any Alexa Prize system in real time and evaluate its performance. Our best model for predicting user ratings gets an R$^2$ of .136 with a DistilBert model, and the best model for predicting length with system independent features gets an R$^2$ of .865, suggesting that conversation length may be a more reliable measure for automatic training of dialogue systems.

The goal of building intelligent dialogue systems has largely been \textit{separately} pursued under two paradigms: task-oriented dialogue (TOD) systems, which perform goal-oriented functions, and open-domain dialogue (ODD) systems, which focus on non-goal-oriented chitchat. The two dialogue modes can potentially be intertwined together seamlessly in the same conversation, as easily done by a friendly human assistant. Such ability is desirable in conversational agents, as the integration makes them more accessible and useful. Our paper addresses this problem of fusing TODs and ODDs in multi-turn dialogues. Based on the popular TOD dataset MultiWOZ, we build a new dataset FusedChat, by rewriting the existing TOD turns and adding new ODD turns. This procedure constructs conversation sessions containing exchanges from both dialogue modes. It features inter-mode contextual dependency, i.e., the dialogue turns from the two modes depend on each other. Rich dependency patterns including co-reference and ellipsis are features. The new dataset, with 60k new human-written ODD turns and 5k re-written TOD turns, offers a benchmark to test a dialogue model's ability to perform inter-mode conversations. This is a more challenging task since the model has to determine the appropriate dialogue mode and generate the response based on the inter-mode context. But such models would better mimic human-level conversation capabilities. We evaluate baseline models on this task, including \textit{classification-based} two-stage models and \textit{two-in-one} fused models. We publicly release FusedChat and the baselines to propel future work on inter-mode dialogue systems https://github.com/tomyoung903/FusedChat.