LongRecall. The completeness of machine-generated text, ensuring that it captures all relevant information, is crucial in domains such as medicine and law and in tasks like list-based question answering (QA), where omissions can have serious consequences. However, existing recall metrics often depend on lexical overlap, leading to errors with unsubstantiated entities and paraphrased answers, while LLM-as-a-Judge methods with long holistic prompts capture broader semantics but remain prone to misalignment and hallucinations without structured verification. We introduce LongRecall, a general three-stage recall evaluation framework that decomposes answers into self-contained facts, successively narrows plausible candidate matches through lexical and semantic filtering, and verifies their alignment through structured entailment checks. This design reduces false positives and false negatives while accommodating diverse phrasings and contextual variations, serving as a foundational building block for systematic recall assessment. We evaluate LongRecall on three challenging long-form QA benchmarks using both human annotations and LLM-based judges, demonstrating substantial improvements in recall accuracy over strong lexical and LLM-as-a-Judge baselines.

Retrieval-Augmented Generation (RAG) grounds Large Language Model (LLM) output by leveraging external knowledge sources to reduce factual hallucinations. However, prior work lacks a comprehensive evaluation of different language families, making it challenging to evaluate LLM robustness against errors in external retrieved knowledge. To overcome this, we establish **NoMIRACL**, a human-annotated dataset for evaluating LLM robustness in RAG across 18 typologically diverse languages. NoMIRACL includes both a non-relevant and a relevant subset. Queries in the non-relevant subset contain passages judged as non-relevant, whereas queries in the relevant subset include at least a single judged relevant passage. We measure relevance assessment using: (i) *hallucination rate*, measuring model tendency to hallucinate when the answer is not present in passages in the non-relevant subset, and (ii) *error rate*, measuring model inaccuracy to recognize relevant passages in the relevant subset. In our work, we observe that most models struggle to balance the two capacities. Models such as LLAMA-2 and Orca-2 achieve over 88% hallucination rate on the non-relevant subset. Mistral and LLAMA-3 hallucinate less but can achieve up to a 74.9% error rate on the relevant subset. Overall, GPT-4 is observed to provide the best tradeoff on both subsets, highlighting future work necessary to improve LLM robustness. NoMIRACL dataset and evaluation code are available at: https://github.com/project-miracl/nomiracl.

Instruction-tuned large language models (LLMs) have been shown to be viable surrogates for the widely used, albeit overly rigid, lexical matching metrics in evaluating question answering (QA) models. However, these LLM-based evaluation methods are invariably based on proprietary LLMs. Despite their remarkable capabilities, proprietary LLMs are costly and subject to internal changes that can affect their output, which inhibits the reproducibility of their results and limits the widespread adoption of LLM-based evaluation. In this demo, we aim to use publicly available LLMs for standardizing LLM-based QA evaluation. However, open-source LLMs lag behind their proprietary counterparts. We overcome this gap by adopting chain-of-thought prompting with self-consistency to build a reliable evaluation framework. We demonstrate that our evaluation framework, based on 750M and 7B open LLMs, correlates competitively with human judgment, compared to most recent GPT-3 and GPT-4 models.

The zero-shot effectiveness of neural retrieval models is often evaluated on the BEIR benchmark-a combination of different IR evaluation datasets. Interestingly, previous studies found that particularly on the BEIR subset Touche 2020, an argument retrieval task, neural retrieval models are considerably less effective than BM25. Still, so far, no further investigation has been conducted on what makes argument retrieval so "special". To more deeply analyze the respective potential limits of neural retrieval models, we run a reproducibility study on the Touche 2020 data. In our study, we focus on two experiments: (i) a black-box evaluation (i.e., no model retraining), incorporating a theoretical exploration using retrieval axioms, and (ii) a data denoising evaluation involving post-hoc relevance judgments. Our black-box evaluation reveals an inherent bias of neural models towards retrieving short passages from the Touche 2020 data, and we also find that quite a few of the neural models' results are unjudged in the Touche 2020 data. As many of the short Touche passages are not argumentative and thus non-relevant per se, and as the missing judgments complicate fair comparison, we denoise the Touche 2020 data by excluding very short passages (less than 20 words) and by augmenting the unjudged data with post-hoc judgments following the Touche guidelines. On the denoised data, the effectiveness of the neural models improves by up to 0.52 in nDCG@10, but BM25 is still more effective. Our code and the augmented Touche 2020 dataset are available at https://github.com/castorini/touche- error-analysis.

BEIR is a benchmark dataset originally designed for zero-shot evaluation of retrieval models across 18 different domain/task combinations. In recent years, we have witnessed the growing popularity of models based on representation learning, which naturally begs the question: How effective are these models when presented with queries and documents that differ from the training data? While BEIR was designed to answer this question, our work addresses two shortcomings that prevent the benchmark from achieving its full potential: First, the sophistication of modern neural methods and the complexity of current software infrastructure create barriers to entry for newcomers. To this end, we provide reproducible reference implementations that cover learned dense and sparse models. Second, comparisons on BEIR are performed by reducing scores from heterogeneous datasets into a single average that is difficult to interpret. To remedy this, we present meta-analyses focusing on effect sizes across datasets that are able to accurately quantify model differences. By addressing both shortcomings, our work facilitates future explorations in a range of interesting research questions.

Unjudged documents or holes in information retrieval benchmarks are considered non-relevant in evaluation, yielding no gains in measuring effectiveness. However, these missing judgments may inadvertently introduce biases into the evaluation as their prevalence for a retrieval model is heavily contingent on the pooling process. Thus, filling holes becomes crucial in ensuring reliable and accurate evaluation. Collecting human judgment for all documents is cumbersome and impractical. In this paper, we aim at leveraging large language models (LLMs) to automatically label unjudged documents. Our goal is to instruct an LLM using detailed instructions to assign fine-grained relevance judgments to holes. To this end, we systematically simulate scenarios with varying degrees of holes by randomly dropping relevant documents from the relevance judgment in TREC DL tracks. Our experiments reveal a strong correlation between our LLM-based method and ground-truth relevance judgments. Based on our simulation experiments conducted on three TREC DL datasets, in the extreme scenario of retaining only 10 Kendall tau correlation of 0.87 and 0.92 on an average for Vicuña-7B and GPT-3.5 Turbo respectively.

BEIR is a benchmark dataset for zero-shot evaluation of information retrieval models across 18 different domain/task combinations. In recent years, we have witnessed the growing popularity of a representation learning approach to building retrieval models, typically using pretrained transformers in a supervised setting. This naturally begs the question: How effective are these models when presented with queries and documents that differ from the training data? Examples include searching in different domains (e.g., medical or legal text) and with different types of queries (e.g., keywords vs. well-formed questions). While BEIR was designed to answer these questions, our work addresses two shortcomings that prevent the benchmark from achieving its full potential: First, the sophistication of modern neural methods and the complexity of current software infrastructure create barriers to entry for newcomers. To this end, we provide reproducible reference implementations that cover the two main classes of approaches: learned dense and sparse models. Second, there does not exist a single authoritative nexus for reporting the effectiveness of different models on BEIR, which has led to difficulty in comparing different methods. To remedy this, we present an official self-service BEIR leaderboard that provides fair and consistent comparisons of retrieval models. By addressing both shortcomings, our work facilitates future explorations in a range of interesting research questions that BEIR enables.

MIRACL is a multilingual dataset for ad hoc retrieval across 18 languages that collectively encompass over three billion native speakers around the world. This resource is designed to support monolingual retrieval tasks, where the queries and the corpora are in the same language. In total, we have gathered over 726k high-quality relevance judgments for 78k queries over Wikipedia in these languages, where all annotations have been performed by native speakers hired by our team. MIRACL covers languages that are both typologically close as well as distant from 10 language families and 13 sub-families, associated with varying amounts of publicly available resources. Extensive automatic heuristic verification and manual assessments were performed during the annotation process to control data quality. In total, MIRACL represents an investment of around five person-years of human annotator effort. Our goal is to spur research on improving retrieval across a continuum of languages, thus enhancing information access capabilities for diverse populations around the world, particularly those that have traditionally been underserved. MIRACL is available at http://miracl.ai/.

The rise of large language models (LLMs) had a transformative impact on search, ushering in a new era of search engines that are capable of generating search results in natural language text, imbued with citations for supporting sources. Building generative information-seeking models demands openly accessible datasets, which currently remain lacking. In this paper, we introduce a new dataset, HAGRID (Human-in-the-loop Attributable Generative Retrieval for Information-seeking Dataset) for building end-to-end generative information-seeking models that are capable of retrieving candidate quotes and generating attributed explanations. Unlike recent efforts that focus on human evaluation of black-box proprietary search engines, we built our dataset atop the English subset of MIRACL, a publicly available information retrieval dataset. HAGRID is constructed based on human and LLM collaboration. We first automatically collect attributed explanations that follow an in-context citation style using an LLM, i.e. GPT-3.5. Next, we ask human annotators to evaluate the LLM explanations based on two criteria: informativeness and attributability. HAGRID serves as a catalyst for the development of information-seeking models with better attribution capabilities.

Lexical matching remains the de facto evaluation method for open-domain question answering (QA). Unfortunately, lexical matching fails completely when a plausible candidate answer does not appear in the list of gold answers, which is increasingly the case as we shift from extractive to generative models. The recent success of large language models (LLMs) for QA aggravates lexical matching failures since candidate answers become longer, thereby making matching with the gold answers even more challenging. Without accurate evaluation, the true progress in open-domain QA remains unknown. In this paper, we conduct a thorough analysis of various open-domain QA models, including LLMs, by manually evaluating their answers on a subset of NQ-open, a popular benchmark. Our assessments reveal that while the true performance of all models is significantly underestimated, the performance of the InstructGPT (zero-shot) LLM increases by nearly +60%, making it on par with existing top models, and the InstructGPT (few-shot) model actually achieves a new state-of-the-art on NQ-open. We also find that more than 50% of lexical matching failures are attributed to semantically equivalent answers. We further demonstrate that regex matching ranks QA models consistent with human judgments, although still suffering from unnecessary strictness. Finally, we demonstrate that automated evaluation models are a reasonable surrogate for lexical matching in some circumstances, but not for long-form answers generated by LLMs. The automated models struggle in detecting hallucinations in LLM answers and are thus unable to evaluate LLMs. At this time, there appears to be no substitute for human evaluation.

Many recent QA models retrieve answers from passages, rather than whole documents, due to the limitations of deep learning models with limited context size. However, this approach ignores important document-level cues that can be crucial in answering questions. This paper reviews three open-domain QA benchmarks from a document-level perspective and finds that they are biased towards passage-level information. Out of 17,000 assessed questions, 82 were identified as requiring document-level reasoning and could not be answered by passage-based models. Document-level retrieval (BM25) outperformed both dense and sparse passage-level retrieval on these questions, highlighting the need for more evaluation of models' ability to understand documents, an often-overlooked challenge in open-domain QA.

The ever-increasing size of language models curtails their widespread availability to the community, thereby galvanizing many companies into offering access to large language models through APIs. One particular type, suitable for dense retrieval, is a semantic embedding service that builds vector representations of input text. With a growing number of publicly available APIs, our goal in this paper is to analyze existing offerings in realistic retrieval scenarios, to assist practitioners and researchers in finding suitable services according to their needs. Specifically, we investigate the capabilities of existing semantic embedding APIs on domain generalization and multilingual retrieval. For this purpose, we evaluate these services on two standard benchmarks, BEIR and MIRACL. We find that re-ranking BM25 results using the APIs is a budget-friendly approach and is most effective in English, in contrast to the standard practice of employing them as first-stage retrievers. For non-English retrieval, re-ranking still improves the results, but a hybrid model with BM25 works best, albeit at a higher cost. We hope our work lays the groundwork for evaluating semantic embedding APIs that are critical in search and more broadly, for information access.

The goal of information-seeking dialogue is to respond to seeker queries with natural language utterances that are grounded on knowledge sources. However, dialogue systems often produce unsupported utterances, a phenomenon known as hallucination. To mitigate this behavior, we adopt a data-centric solution and create FaithDial, a new benchmark for hallucination-free dialogues, by editing hallucinated responses in the Wizard of Wikipedia (WoW) benchmark. We observe that FaithDial is more faithful than WoW while also maintaining engaging conversations. We show that FaithDial can serve as training signal for: i) a hallucination critic, which discriminates whether an utterance is faithful or not, and boosts the performance by 12.8 F1 score on the BEGIN benchmark compared to existing datasets for dialogue coherence; ii) high-quality dialogue generation. We benchmark a series of state-of-the-art models and propose an auxiliary contrastive objective that achieves the highest level of faithfulness and abstractiveness based on several automated metrics. Further, we find that the benefits of FaithDial generalize to zero-shot transfer on other datasets, such as CMU-Dog and TopicalChat. Finally, human evaluation reveals that responses generated by models trained on FaithDial are perceived as more interpretable, cooperative, and engaging.

Data Augmentation (DA) is known to improve the generalizability of deep neural networks. Most existing DA techniques naively add a certain number of augmented samples without considering the quality and the added computational cost of these samples. To tackle this problem, a common strategy, adopted by several state-of-the-art DA methods, is to adaptively generate or re-weight augmented samples with respect to the task objective during training. However, these adaptive DA methods: (1) are computationally expensive and not sample-efficient, and (2) are designed merely for a specific setting. In this work, we present a universal DA technique, called Glitter, to overcome both issues. Glitter can be plugged into any DA method, making training sample-efficient without sacrificing performance. From a pre-generated pool of augmented samples, Glitter adaptively selects a subset of worst-case samples with maximal loss, analogous to adversarial DA. Without altering the training strategy, the task objective can be optimized on the selected subset. Our thorough experiments on the GLUE benchmark, SQuAD, and HellaSwag in three widely used training setups including consistency training, self-distillation and knowledge distillation reveal that Glitter is substantially faster to train and achieves a competitive performance, compared to strong baselines.

MIRACL (Multilingual Information Retrieval Across a Continuum of Languages) is a multilingual dataset we have built for the WSDM 2023 Cup challenge that focuses on ad hoc retrieval across 18 different languages, which collectively encompass over three billion native speakers around the world. These languages have diverse typologies, originate from many different language families, and are associated with varying amounts of available resources -- including what researchers typically characterize as high-resource as well as low-resource languages. Our dataset is designed to support the creation and evaluation of models for monolingual retrieval, where the queries and the corpora are in the same language. In total, we have gathered over 700k high-quality relevance judgments for around 77k queries over Wikipedia in these 18 languages, where all assessments have been performed by native speakers hired by our team. Our goal is to spur research that will improve retrieval across a continuum of languages, thus enhancing information access capabilities for diverse populations around the world, particularly those that have been traditionally underserved. This overview paper describes the dataset and baselines that we share with the community. The MIRACL website is live at http://miracl.ai/.

The paradigm of fine-tuning Pre-trained Language Models (PLMs) has been successful in Entity Matching (EM). Despite their remarkable performance, PLMs exhibit tendency to learn spurious correlations from training data. In this work, we aim at investigating whether PLM-based entity matching models can be trusted in real-world applications where data distribution is different from that of training. To this end, we design an evaluation benchmark to assess the robustness of EM models to facilitate their deployment in the real-world settings. Our assessments reveal that data imbalance in the training data is a key problem for robustness. We also find that data augmentation alone is not sufficient to make a model robust. As a remedy, we prescribe simple modifications that can improve the robustness of PLM-based EM models. Our experiments show that while yielding superior results for in-domain generalization, our proposed model significantly improves the model robustness, compared to state-of-the-art EM models.

The advent of multilingual language models has generated a resurgence of interest in cross-lingual information retrieval (CLIR), which is the task of searching documents in one language with queries from another. However, the rapid pace of progress has led to a confusing panoply of methods and reproducibility has lagged behind the state of the art. In this context, our work makes two important contributions: First, we provide a conceptual framework for organizing different approaches to cross-lingual retrieval using multi-stage architectures for mono-lingual retrieval as a scaffold. Second, we implement simple yet effective reproducible baselines in the Anserini and Pyserini IR toolkits for test collections from the TREC 2022 NeuCLIR Track, in Persian, Russian, and Chinese. Our efforts are built on a collaboration of the two teams that submitted the most effective runs to the TREC evaluation. These contributions provide a firm foundation for future advances.

There is essential information in the underlying structure of words and phrases in natural language questions, and this structure has been extensively studied. In this paper, we study one particular structure, referred to as frozen phrases, that is highly expected to transfer as a whole from questions to answer passages. Frozen phrases, if detected, can be helpful in open-domain Question Answering (QA) where identifying the localized context of a given input question is crucial. An interesting question is if frozen phrases can be accurately detected. We cast the problem as a sequence-labeling task and create synthetic data from existing QA datasets to train a model. We further plug this model into a sparse retriever that is made aware of the detected phrases. Our experiments reveal that detecting frozen phrases whose presence in answer documents are highly plausible yields significant improvements in retrievals as well as in the end-to-end accuracy of open-domain QA models.