Transformer-based ranking models, such as MonoBERT, are central to Information Retrieval; yet their inner workings remain largely opaque. This hinders not only our understanding of the systems implementing them, but also our ability to improve them. To alleviate this limitation, we introduce IR Lens, a new interpretability tool tailored to cross-encoders based on two key components: 1) Neuron Integrated Gradients to expose the contributions of model parts at multiple levels, and 2) targeted ablations to support hypothesis tracking. With its interactive graphical interface, IR Lens enables IR practitioners to explore, analyze, and manipulate neuron-level mechanisms in cross-encoders, facilitating a deeper understanding of neural ranking models. By extending the reach of existing interpretability methods, we believe IR Lens has the potential to support the improvement of cross-encoders.

Learned Sparse IR models, such as SPLADE, offer an excellent efficiency-effectiveness tradeoff. However, they rely on the underlying backbone vocabulary, which might hinder performance (polysemicity and synonymy) and pose a challenge for multi-lingual and multi-modal usages. To solve this limitation, we propose to replace the backbone vocabulary with a latent space of semantic concepts learned using Sparse Auto-Encoders (SAE). Throughout this paper, we study the compatibility of these 2 concepts, explore training approaches, and analyze the differences between our SAE-SPLADE model and traditional SPLADE models. Our experiments demonstrate that SAE-SPLADE achieves retrieval performance comparable to SPLADE on both in-domain and out-of-domain tasks while offering improved efficiency.

Large Language Models (LLMs), when used for conditional text generation, often produce hallucinations, i.e., information that is unfaithful or not grounded in the input context. This issue arises in typical conditional text generation tasks, such as text summarization and data-to-text generation, where the goal is to produce fluent text based on contextual input. When fine-tuned on specific domains, LLMs struggle to provide faithful answers to a given context, often adding information or generating errors. One underlying cause of this issue is that LLMs rely on statistical patterns learned from their training data. This reliance can interfere with the model's ability to stay faithful to a provided context, leading to the generation of ungrounded information. We build upon this observation and introduce a novel self-supervised method for generating a training set of unfaithful samples. We then refine the model using a training process that encourages the generation of grounded outputs over unfaithful ones, drawing on preference-based training. Our approach leads to significantly more grounded text generation, outperforming existing self-supervised techniques in faithfulness, as evaluated through automatic metrics, LLM-based assessments, and human evaluations.

In information retrieval (IR), providing appropriate clarifications to better understand users' information needs is crucial for building a proactive search-oriented dialogue system. Due to the strong in-context learning ability of large language models (LLMs), recent studies investigate prompting methods to generate clarifications using few-shot or Chain of Thought (CoT) prompts. However, vanilla CoT prompting does not distinguish the characteristics of different information needs, making it difficult to understand how LLMs resolve ambiguities in user queries. In this work, we focus on the concept of ambiguity for clarification, seeking to model and integrate ambiguities in the clarification process. Following the reasoning and acting paradigm, we propose a new prompting scheme Ambiguity Type-Chain of Thought (AT-CoT), which enhances the reasoning abilities of LLMs by limiting CoT to first predict ambiguity types that can be interpreted as actions, then generate clarifications correspondingly. Experiments are conducted on various datasets containing human-annotated clarifying questions to compare AT-CoT with multiple baselines. We also perform user simulation to implicitly measure the quality of generated clarifications under various IR scenarios. Our codes are available at: https://github.com/anfutang/ClarifyingAmbiguities/.

International audience

Florian Le Bronnec, Song Duong, Mathieu Ravaut, Alexandre Allauzen, Nancy Chen, Vincent Guigue, Alberto Lumbreras, Laure Soulier, Patrick Gallinari. Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Co-speech gestures are fundamental for communication. The advent of recent deep learning techniques has facilitated the creation of lifelike, synchronous co-speech gestures for Embodied Conversational Agents. "In-the-wild" datasets, aggregating video content from platforms like YouTube via human pose detection technologies, provide a feasible solution by offering 2D skeletal sequences aligned with speech. Concurrent developments in lifting models enable the conversion of these 2D sequences into 3D gesture databases. However, it is important to note that the 3D poses estimated from the 2D extracted poses are, in essence, approximations of the ground-truth, which remains in the 2D domain. This distinction raises questions about the impact of gesture representation dimensionality on the quality of generated motions. Our study examines the effect of using either 2D or 3D joint coordinates as training data on the performance of speech-to-gesture deep generative models.

Neural data-to-text systems lack the control and factual accuracy required to generate useful and insightful summaries of multidimensional data.We propose a solution in the form of data views, where each view describes an entity and its attributes along specific dimensions.A sequence of views can then be used as a high-level schema for document planning, with the neural model handling the complexities of micro-planning and surface realization.We show that our view-based system retains factual accuracy while offering high-level control of output that can be tailored based on user preference or other norms within the domain.

Foundation models are first pre-trained on vast unsupervised datasets and then fine-tuned on labeled data. Reinforcement learning, notably from human feedback (RLHF), can further align the network with the intended usage. Yet the imperfections in the proxy reward may hinder the training and lead to suboptimal results; the diversity of objectives in real-world tasks and human opinions exacerbate the issue. This paper proposes embracing the heterogeneity of diverse rewards by following a multi-policy strategy. Rather than focusing on a single a priori reward, we aim for Pareto-optimal generalization across the entire space of preferences. To this end, we propose rewarded soup, first specializing multiple networks independently (one for each proxy reward) and then interpolating their weights linearly. This succeeds empirically because we show that the weights remain linearly connected when fine-tuned on diverse rewards from a shared pre-trained initialization. We demonstrate the effectiveness of our approach for text-to-text (summarization, Q&A, helpful assistant, review), text-image (image captioning, text-to-image generation, visual grounding), and control (locomotion) tasks. We hope to enhance the alignment of deep models, and how they interact with the world in all its diversity.

When users initiate search sessions, their query are often ambiguous or might lack of context; this resulting in non-efficient document ranking. Multiple approaches have been proposed by the Information Retrieval community to add context and retrieve documents aligned with users' intents. While some work focus on query disambiguation using users' browsing history, a recent line of work proposes to interact with users by asking clarification questions or/and proposing clarification panels. However, these approaches count either a limited number (i.e., 1) of interactions with user or log-based interactions. In this paper, we propose and evaluate a fully simulated query clarification framework allowing multi-turn interactions between IR systems and user agents.

Abstract Data-to-Text Generation (DTG) is a subfield of Natural Language Generation aiming at transcribing structured data in natural language descriptions. The field has been recently boosted by the use of neural-based generators which exhibit on one side great syntactic skills without the need of hand-crafted pipelines; on the other side, the quality of the generated text reflects the quality of the training data, which in realistic settings only offer imperfectly aligned structure-text pairs. Consequently, state-of-art neural models include misleading statements –usually called hallucinations—in their outputs. The control of this phenomenon is today a major challenge for DTG, and is the problem addressed in the paper. Previous work deal with this issue at the instance level: using an alignment score for each table-reference pair. In contrast, we propose a finer-grained approach, arguing that hallucinations should rather be treated at the word level. Specifically, we propose a Multi-Branch Decoder which is able to leverage word-level labels to learn the relevant parts of each training instance. These labels are obtained following a simple and efficient scoring procedure based on co-occurrence analysis and dependency parsing. Extensive evaluations, via automated metrics and human judgment on the standard WikiBio benchmark, show the accuracy of our alignment labels and the effectiveness of the proposed Multi-Branch Decoder. Our model is able to reduce and control hallucinations, while keeping fluency and coherence in generated texts. Further experiments on a degraded version of ToTTo show that our model could be successfully used on very noisy settings.

Clement Rebuffel, Thomas Scialom, Laure Soulier, Benjamin Piwowarski, Sylvain Lamprier, Jacopo Staiano, Geoffrey Scoutheeten, Patrick Gallinari. Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. 2021.

In this paper, we propose a new unsupervised method for learning word embedding with raw characters as input representations, bypassing the problems arising from the use of a dictionary.To achieve this purpose, we translate the distributional hypothesis into a unsupervised metric learning objective, which allows to consider only an encoder instead of an encoder-decoder architecture.We propose to use a convolutional neural network with bilinear product blocks and residual connections to encode co-occurrences patterns.We show the effectiveness of our approach by comparing it with classical word embedding methods such as fastText and GloVe on several benchmarks.

In language generation models conditioned by structured data, the classical training via maximum likelihood almost always leads models to pick up on dataset divergence (i.e., hallucinations or omissions), and to incorporate them erroneously in their own generations at inference.In this work, we build ontop of previous Reinforcement Learning based approaches and show that a model-agnostic framework relying on the recently introduced PARENT metric is efficient at reducing both hallucinations and omissions.Evaluations on the widely used WikiBIO and WebNLG benchmarks demonstrate the effectiveness of this framework compared to state-of-the-art models.

Patrick Bordes, Eloi Zablocki, Laure Soulier, Benjamin Piwowarski, Patrick Gallinari. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

Zero-Shot Learning (ZSL) aims at classifying unlabeled objects by leveraging auxiliary knowledge, such as semantic representations. A limitation of previous approaches is that only intrinsic properties of objects, e.g. their visual appearance, are taken into account while their context, e.g. the surrounding objects in the image, is ignored. Following the intuitive principle that objects tend to be found in certain contexts but not others, we propose a new and challenging approach, context-aware ZSL, that leverages semantic representations in a new way to model the conditional likelihood of an object to appear in a given context. Finally, through extensive experiments conducted on Visual Genome, we show that contextual information can substantially improve the standard ZSL approach and is robust to unbalanced classes.

An intensive recent research work investigated the combined use of hand-curated knowledge resources and corpus-driven resources to learn effective text representations. The overall learning process could be run by online revising the learning objective or by offline refining an original learned representation. The differentiated impact of each of the learning approaches on the quality of the learned representations has not been studied so far in the literature. This article focuses on the design of comparable offline vs. online knowledge-enhanced document representation learning models and the comparison of their effectiveness using a set of standard IR and NLP downstream tasks. The results of quantitative and qualitative analyses show that (1) offline vs. online learning approaches have dissimilar result trends regarding the task as well as the dataset distribution counts with regard to domain application; (2) while considering external knowledge resources is undoubtedly beneficial, the way used to express relational constraints could affect semantic inference effectiveness. The findings of this work present opportunities for the design of future representation learning models, but also for providing insights about the evaluation of such models.

Representing the semantics of words is a long-standing problem for the natural language processing community. Most methods compute word semantics given their textual context in large corpora. More recently, researchers attempted to integrate perceptual and visual features. Most of these works consider the visual appearance of objects to enhance word representations but they ignore the visual environment and context in which objects appear. We propose to unify text-based techniques with vision-based techniques by simultaneously leveraging textual and visual context to learn multimodal word embeddings. We explore various choices for what can serve as a visual context and present an end-to-end method to integrate visual context elements in a multimodal skip-gram model. We provide experiments and extensive analysis of the obtained results.

Designing powerful tools that support cooking activities has rapidly gained popularity due to the massive amounts of available data, as well as recent advances in machine learning that are capable of analyzing them. In this paper, we propose a cross-modal retrieval model aligning visual and textual data (like pictures of dishes and their recipes) in a shared representation space. We describe an effective learning scheme, capable of tackling large-scale problems, and validate it on the Recipe1M dataset containing nearly 1 million picture-recipe pairs. We show the effectiveness of our approach regarding previous state-of-the-art models and present qualitative results over computational cooking use cases.

The workshop on the evaluation of collaborative information retrieval and seeking (ECol) is held in conjunction with the ACM SIGIR Conference on Human Information Interaction & Retrieval (CHIIR) in Oslo, Norway. To make the workshop active and the participant pro-active, we released datasets and tools so as to help researchers contributing to the formalization of evaluation frameworks for challenging collaborative tasks. The workshop is split into two parts. First, a presentation session. Then, the afternoon is devoted to group discussion addressing challenges of evaluating and designing models for social and collaborative search.

In this paper, we specifically consider the challenging task of solving a question posted on Twitter. The latter generally remains unanswered and most of the replies, if any, are only from members of the questioner's neighborhood. As outlined in previous work related to community Q&A, we believe that question-answering is a collaborative process and that the relevant answer to a question post is an aggregation of answer nuggets posted by a group of relevant users. Thus, the problem of identifying the relevant answer turns into the problem of identifying the right group of users who would provide useful answers and would possibly be willing to collaborate together in the long-term. Accordingly, we present a novel method, called CRAQ, that is built on the collaboration paradigm and formulated as a group entropy optimization problem. To optimize the quality of the group, an information gain measure is used to select the most likely ``informative" users according to topical and collaboration likelihood predictive features. Crowd-based experiments performed on two crisis-related Twitter datasets demonstrate the effectiveness of our collaborative-based answering approach.

Collaborative Information Seeking/Retrieval (CIS/CIR) has given rise to several challenges in terms of search behavior analysis, retrieval model formalization as well as interface design. However, the major issue of evaluation in CIS/CIR is still underexplored. The goal of this workshop is to investigate the evaluation challenges in CIS/CIR with the hope of building standardized evaluation frameworks, methodologies, and task specifications that would foster and grow the research area (in a collaborative fashion).

Collaborative information retrieval systems often rely on division of labor policies. Such policies allow work to be divided among collaborators with the aim of preventing redundancy and optimizing the synergic effects of collaboration. Most of the underlying methods achieve these goals by the means of explicit vs. implicit role-based mediation. In this paper, we investigate whether and how different factors, such as users' behavior, search strategies, and effectiveness, are related to role assignment within a collaborative exploratory search. Our main findings suggest that: (1) spontaneous and cohesive implicit roles might emerge during the collaborative search session implying users with no prior roles, and that these implicit roles favor the search precision, (2) role drift might occur alongside the search session performed by users with prior-assigned roles.

Most of the previous approaches surrounding collaborative information retrieval (CIR) provide either a user-based mediation, in which the system only supports users' collaborative activities, or a system-based mediation, in which the system plays an active part in balancing user roles, re-ranking results, and distributing them to optimize overall retrieval performance. In this paper, we propose to combine both of these approaches by a role mining methodology that learns from users' actions about the retrieval strategy they adapt. This hybrid method aims at showing how users are different and how to use these differences for suggesting roles. The core of the method is expressed as an algorithm that (1) monitors users' actions in a CIR setting; (2) discovers differences among the collaborators along certain dimensions; and (3) suggests appropriate roles to make the most out of individual skills and optimize IR performance. Our approach is empirically evaluated and relies on two different laboratory studies involving 70 pairs of users. Our experiments show promising results that highlight how role mining could optimize the collaboration within a search session. The contributions of this work include a new algorithm for mining user roles in collaborative IR, an evaluation methodology, and a new approach to improve IR performance with the operationalization of user-driven system-mediated collaboration.

Collaborative information retrieval involves retrieval settings in which a group of users collaborates to satisfy the same underlying need. One core issue of collaborative IR models involves either supporting collaboration with adapted tools or developing IR models for a multiple-user context and providing a ranked list of documents adapted for each collaborator. In this paper, we introduce the first document-ranking model supporting collaboration between two users characterized by roles relying on different domain expertise levels. Specifically, we propose a two-step ranking model: we first compute a document-relevance score, taking into consideration domain expertise-based roles. We introduce specificity and novelty factors into language-model smoothing, and then we assign, via an Expectation–Maximization algorithm, documents to the best-suited collaborator. Our experiments employ a simulation-based framework of collaborative information retrieval and show the significant effectiveness of our model at different search levels.

A new challenge, accessing multiple relevant entities, arises from the availability of linked heterogeneous data. In this article, we address more specifically the problem of accessing relevant entities, such as publications and authors within a bibliographic network, given an information need. We propose a novel algorithm, called BibRank, that estimates a joint relevance of documents and authors within a bibliographic network. This model ranks each type of entity using a score propagation algorithm with respect to the query topic and the structure of the underlying bi‐type information entity network. Evidence sources, namely content‐based and network‐based scores, are both used to estimate the topical similarity between connected entities. For this purpose, authorship relationships are analyzed through a language model‐based score on the one hand and on the other hand, non topically related entities of the same type are detected through marginal citations. The article reports the results of experiments using the Bibrank algorithm for an information retrieval task. The CiteSeerX bibliographic data set forms the basis for the topical query automatic generation and evaluation. We show that a statistically significant improvement over closely related ranking models is achieved.