Utilizing large language models (LLMs) for zero-shot document ranking is done in one of two ways: (1) prompt-based re-ranking methods, which require no further training but are only feasible for re-ranking a handful of candidate documents due to computational costs; and (2) unsupervised contrastive trained dense retrieval methods, which can retrieve relevant documents from the entire corpus but require a large amount of paired text data for contrastive training.In this paper, we propose PromptReps, which combines the advantages of both categories: no need for training and the ability to retrieve from the whole corpus. Our method only requires prompts to guide an LLM to generate query and document representations for effective document retrieval. Specifically, we prompt the LLMs to represent a given text using a single word, and then use the last token’s hidden states and the corresponding logits associated with the prediction of the next token to construct a hybrid document retrieval system. The retrieval system harnesses both dense text embedding and sparse bag-of-words representations given by the LLM.Our experimental evaluation on the MSMARCO, TREC deep learning and BEIR zero-shot document retrieval datasets illustrates that this simple prompt-based LLM retrieval method can achieve a similar or higher retrieval effectiveness than state-of-the-art LLM embedding methods that are trained with large amounts of unsupervised data, especially when using a larger LLM.

Diffusion models are the state of the art in text-to-image generation, but their perceptual variability remains understudied. In this paper, we examine how prompts affect image variability in black-box diffusion-based models. We propose W1KP, a human-calibrated measure of variability in a set of images, bootstrapped from existing image-pair perceptual distances. Current datasets do not cover recent diffusion models, thus we curate three test sets for evaluation. Our best perceptual distance outperforms nine baselines by up to 18 points in accuracy, and our calibration matches graded human judgements 78% of the time. Using W1KP, we study prompt reusability and show that Imagen prompts can be reused for 10-50 random seeds before new images become too similar to already generated images, while Stable Diffusion XL and DALL-E 3 can be reused 50-200 times. Lastly, we analyze 56 linguistic features of real prompts, finding that the prompt’s length, CLIP embedding norm, concreteness, and word senses influence variability most. As far as we are aware, we are the first to analyze diffusion variability from a visuolinguistic perspective. Our project page is at http://w1kp.com.

In the real world, documents are organized in different formats and varied modalities. Traditional retrieval pipelines require tailored document parsing techniques and content extraction modules to prepare input for indexing. This process is tedious, prone to errors, and has information loss. To this end, we propose Document Screenshot Embedding (DSE), a novel retrieval paradigm that regards document screenshots as a unified input format, which does not require any content extraction preprocess and preserves all the information in a document (e.g., text, image and layout). DSE leverages a large vision-language model to directly encode document screenshots into dense representations for retrieval. To evaluate our method, we first craft the dataset of Wiki-SS, a 1.3M Wikipedia web page screenshots as the corpus to answer the questions from the Natural Questions dataset. In such a text-intensive document retrieval setting, DSE shows competitive effectiveness compared to other text retrieval methods relying on parsing. For example, DSE outperforms BM25 by 17 points in top-1 retrieval accuracy. Additionally, in a mixed-modality task of slide retrieval, DSE significantly outperforms OCR text retrieval methods by over 15 points in nDCG@10. These experiments show that DSE is an effective document retrieval paradigm for diverse types of documents. Model checkpoints, code, and Wiki-SS collection will be released.

The rapid evolution of Large Language Models (LLMs) and conversational assistants necessitates dynamic, scalable, and configurable conversational datasets for training and evaluation.These datasets must accommodate diverse user interaction modes, including text and voice, each presenting unique modeling challenges. Knowledge Graphs (KGs), with their structured and evolving nature, offer an ideal foundation for current and precise knowledge.Although human-curated KG-based conversational datasets exist, they struggle to keep pace with the rapidly changing user information needs.We present ConvKGYarn, a scalable method for generating up-to-date and configurable conversational KGQA datasets. Qualitative psychometric analyses demonstrate ConvKGYarn’s effectiveness in producing high-quality data comparable to popular conversational KGQA datasets across various metrics.ConvKGYarn excels in adhering to human interaction configurations and operating at a significantly larger scale.We showcase ConvKGYarn’s utility by testing LLMs on diverse conversations — exploring model behavior on conversational KGQA sets with different configurations grounded in the same KG fact set.Our results highlight the ability of ConvKGYarn to improve KGQA foundations and evaluate parametric knowledge of LLMs, thus offering a robust solution to the constantly evolving landscape of conversational assistants.

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.

There has been a huge number of benchmarks proposed to evaluate how large language models (LLMs) behave for logic inference tasks. However, it remains an open question how to properly evaluate this ability. In this paper, we provide a systematic overview of prior works on the logical reasoning ability of LLMs for analyzing categorical syllogisms. We first investigate all the possible variations for categorical syllogisms from a purely logical perspective and then examine the underlying configurations (i.e., mood and figure) tested by existing datasets. Our results indicate that compared to template-based synthetic datasets, crowdsourcing approaches normally sacrifice the coverage of configurations (i.e., mood and figure) of categorical syllogisms for more language variations, thus bringing challenges to fully testing LLMs under different situations. We then summarize the findings and observations for the performance of LLMs to infer the validity of syllogisms from the current literature. The error rate breakdown analyses suggest that the interpretation of quantifiers seems to be the current bottleneck that limits the performance of the LLMs and is thus worth more attention. Finally, we discuss several points that might be worth considering when researchers plan to release categorical syllogism datasets. We hope our work will provide a timely review of the current literature regarding categorical syllogisms, and motivate more interdisciplinary research between communities, specifically computational linguists and logicians.

Large language models (LLMs) exhibit positional bias in how they use context, which especially affects listwise ranking. To address this, we propose permutation self-consistency, a form of self-consistency over the ranking list outputs of black-box LLMs. Our key idea is to marginalize out different list orders in the prompt to produce an order-independent ranking with less positional bias. First, given some input prompt, we repeatedly shuffle the list in the prompt and pass it through the LLM while holding the instructions the same. Next, we aggregate the resulting sample of rankings by computing the central ranking closest in distance to all of them, marginalizing out prompt order biases in the process. Theoretically, we prove the robustness of our method, showing convergence to the true ranking under random perturbations.Empirically, on five datasets in sorting and passage reranking, our approach improves scores from conventional inference by up to 34-52% for Mistral, 7-18% for GPT-3.5, 8-16% for LLaMA v2 (70B). Our code is at https://github.com/castorini/perm-sc.

There has been limited success for dense retrieval models in multilingual retrieval, due to uneven and scarce training data available across multiple languages. Synthetic training data generation is promising (e.g., InPars or Promptagator), but has been investigated only for English. Therefore, to study model capabilities across both cross-lingual and monolingual retrieval tasks, we develop **SWIM-IR**, a synthetic retrieval training dataset containing 33 (high to very-low resource) languages for fine-tuning multilingual dense retrievers without requiring any human supervision. To construct SWIM-IR, we propose SAP (summarize-then-ask prompting), where the large language model (LLM) generates a textual summary prior to the query generation step. SAP assists the LLM in generating informative queries in the target language. Using SWIM-IR, we explore synthetic fine-tuning of multilingual dense retrieval models and evaluate them robustly on three retrieval benchmarks: XOR-Retrieve (cross-lingual), MIRACL (monolingual) and XTREME-UP (cross-lingual). Our models, called SWIM-X, are competitive with human-supervised dense retrieval models, e.g., mContriever-X, finding that SWIM-IR can cheaply substitute for expensive human-labeled retrieval training data. SWIM-IR dataset and SWIM-X models are available at: https://github.com/google-research-datasets/SWIM-IR.

In text ranking, it is generally believed that the cross-encoders already gather sufficient token interaction information via the attention mechanism in the hidden layers. However, our results show that the cross-encoders can consistently benefit from additional token interaction in the similarity computation at the last layer. We introduce CELI (Cross-Encoder with Late Interaction), which incorporates a late interaction layer into the current cross-encoder models. This simple method brings 5% improvement on BEIR without compromising in-domain effectiveness or search latency. Extensive experiments show that this finding is consistent across different sizes of the cross-encoder models and the first-stage retrievers. Our findings suggest that boiling all information into the [CLS] token is a suboptimal use for cross-encoders, and advocate further studies to investigate its relevance score mechanism.

This paper examines the integration of images into Wikipedia articles by evaluating image–text retrieval tasks in multimedia content creation, focusing on developing retrieval-augmented tools to enhance the creation of high-quality multimedia articles. Despite ongoing research, the interplay between text and visuals, such as photos and diagrams, remains underexplored, limiting support for real-world applications. We introduce AToMiC, a dataset for long-form, knowledge-intensive image–text retrieval, detailing its task design, evaluation protocols, and relevance criteria.Our findings show that a hybrid approach combining a sparse retriever with a dense retriever achieves satisfactory effectiveness, with nDCG@10 scores around 0.4 for Image Suggestion and Image Promotion tasks, providing insights into the challenges of retrieval evaluation in an image–text interleaved article composition context.The AToMiC dataset is available at https://github.com/TREC-AToMiC/AToMiC.

The emerging citation-based QA systems are gaining more attention especially in generative AI search applications. The importance of extracted knowledge provided to these systems is vital from both accuracy (completeness of information) and efficiency (extracting the information in a timely manner). In this regard, citation-based QA systems are suffering from two shortcomings. First, they usually rely only on web as a source of extracted knowledge and adding other external knowledge sources can hamper the efficiency of the system. Second, web-retrieved contents are usually obtained by some simple heuristics such as fixed length or breakpoints which might lead to splitting information into pieces. To mitigate these issues, we propose our enhanced web and efficient knowledge graph (KG) retrieval solution (EWEK-QA) to enrich the content of the extracted knowledge fed to the system. This has been done through designing an adaptive web retriever and incorporating KGs triples in an efficient manner. We demonstrate the effectiveness of over the open-source state-of-the-art (SoTA) web-based and KG baseline models using a comprehensive set of quantitative and human evaluation experiments. Our model is able to: first, improve the web-retriever baseline in terms of extracting more relevant passages (>20%), the coverage of answer span (>25%) and self containment (>35%); second, obtain and integrate KG triples into its pipeline very efficiently (by avoiding any LLM calls) to outperform the web-only and KG-only SoTA baselines significantly in 7 quantitative QA tasks and our human evaluation.

Large language models (LLMs) as listwise rerankers have shown impressive zero-shot capabilities in various passage ranking tasks. Despite their success, there is still a gap in existing literature on their effectiveness in reranking low-resource languages. To address this, we investigate how LLMs function as listwise rerankers in cross-lingual information retrieval (CLIR) systems with queries in English and passages in four African languages: Hausa, Somali, Swahili, and Yoruba. We analyze and compare the effectiveness of monolingual reranking using either query or document translations. We also evaluate the effectiveness of LLMs when leveraging their own generated translations. To grasp the general picture, we examine the effectiveness of multiple LLMs — the proprietary models RankGPT-4 and RankGPT-3.5, along with the open-source model RankZephyr. While the document translation setting, i.e., both queries and documents are in English, leads to the best reranking effectiveness, our results indicate that for specific LLMs, reranking in the African language setting achieves competitive effectiveness with the cross-lingual setting, and even performs better when using the LLM’s own translations.

While dense retrieval has been shown to be effective and efficient across tasks and languages, it remains difficult to create effective fully zero-shot dense retrieval systems when no relevance labels are available. In this paper, we recognize the difficulty of zero-shot learning and encoding relevance. Instead, we propose to pivot through Hypothetical Document Embeddings (HyDE). Given a query, HyDE first zero-shot prompts an instruction-following language model (e.g., InstructGPT) to generate a hypothetical document. The document captures relevance patterns but is “fake” and may contain hallucinations. Then, an unsupervised contrastively learned encoder (e.g., Contriever) encodes the document into an embedding vector. This vector identifies a neighborhood in the corpus embedding space, from which similar real documents are retrieved based on vector similarity. This second step grounds the generated document to the actual corpus, with the encoder’s dense bottleneck filtering out the hallucinations. Our experiments show that HyDE significantly outperforms the state-of-the-art unsupervised dense retriever Contriever and shows strong performance comparable to fine-tuned retrievers across various tasks (e.g. web search, QA, fact verification) and in non-English languages (e.g., sw, ko, ja, bn).

Diffusion models are a milestone in text-to-image generation, but they remain poorly understood, lacking interpretability analyses. In this paper, we perform a text-image attribution analysis on Stable Diffusion, a recently open-sourced model. To produce attribution maps, we upscale and aggregate cross-attention maps in the denoising module, naming our method DAAM. We validate it by testing its segmentation ability on nouns, as well as its generalized attribution quality on all parts of speech, rated by humans. On two generated datasets, we attain a competitive 58.8-64.8 mIoU on noun segmentation and fair to good mean opinion scores (3.4-4.2) on generalized attribution. Then, we apply DAAM to study the role of syntax in the pixel space across head–dependent heat map interaction patterns for ten common dependency relations. We show that, for some relations, the head map consistently subsumes the dependent, while the opposite is true for others. Finally, we study several semantic phenomena, focusing on feature entanglement; we find that the presence of cohyponyms worsens generation quality by 9%, and descriptive adjectives attend too broadly. We are the first to interpret large diffusion models from a visuolinguistic perspective, which enables future research. Our code is at https://github.com/castorini/daam.

Multi-vector retrieval methods combine the merits of sparse (e.g. BM25) and dense (e.g. DPR) retrievers and have achieved state-of-the-art performance on various retrieval tasks. These methods, however, are orders of magnitude slower and need much more space to store their indices compared to their single-vector counterparts. In this paper, we unify different multi-vector retrieval models from a token routing viewpoint and propose conditional token interaction via dynamic lexical routing, namely CITADEL, for efficient and effective multi-vector retrieval.CITADEL learns to route different token vectors to the predicted lexical keys such that a query token vector only interacts with document token vectors routed to the same key. This design significantly reduces the computation cost while maintaining high accuracy. Notably, CITADEL achieves the same or slightly better performance than the previous state of the art, ColBERT-v2, on both in-domain (MS MARCO) and out-of-domain (BEIR) evaluations, while being nearly 40 times faster. Source code and data are available at https://github.com/facebookresearch/dpr-scale/tree/citadel.

Noticing the urgent need to provide tools for fast and user-friendly qualitative analysis of large-scale textual corpora of the modern NLP, we propose to turn to the mature and well-tested methods from the domain of Information Retrieval (IR) - a research field with a long history of tackling TB-scale document collections. We discuss how Pyserini - a widely used toolkit for reproducible IR research can be integrated with the Hugging Face ecosystem of open-source AI libraries and artifacts. We leverage the existing functionalities of both platforms while proposing novel features further facilitating their integration. Our goal is to give NLP researchers tools that will allow them to develop retrieval-based instrumentation for their data analytics needs with ease and agility. We include a Jupyter Notebook-based walk through the core interoperability features, available on GitHub: https://github.com/huggingface/gaia. We then demonstrate how the ideas we present can be operationalized to create a powerful tool for qualitative data analysis in NLP. We present GAIA Search - a search engine built following previously laid out principles, giving access to four popular large-scale text collections. GAIA serves a dual purpose of illustrating the potential of methodologies we discuss but also as a standalone qualitative analysis tool that can be leveraged by NLP researchers aiming to understand datasets prior to using them in training. GAIA is hosted live on Hugging Face Spaces: https://huggingface.co/spaces/spacerini/gaia.

The ever-increasing size of language models curtails their widespread access to the community, thereby galvanizing many companies and startups into offering access to large language models through APIs. One particular API, suitable for dense retrieval, is the semantic embedding API that builds vector representations of a given text. With a growing number of APIs at our disposal, in this paper, our goal is to analyze semantic embedding APIs in realistic retrieval scenarios in order to assist practitioners and researchers in finding suitable services according to their needs. Specifically, we wish to investigate the capabilities of existing APIs on domain generalization and multilingual retrieval. For this purpose, we evaluate the embedding APIs 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 on English, in contrast to the standard practice, i.e., 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 thoroughly evaluating APIs that are critical in search and more broadly, in information retrieval.

There exists a wide variety of efficiency methods for natural language processing (NLP) tasks, such as pruning, distillation, dynamic inference, quantization, etc. From a different perspective, we can consider an efficiency method as an operator applied on a model. Naturally, we may construct a pipeline of operators, i.e., to apply multiple efficiency methods on the model sequentially. In this paper, we study the plausibility of this idea, and more importantly, the commutativity and cumulativeness of efficiency operators. We make two interesting observations from our experiments: (1) The operators are commutative—the order of efficiency methods within the pipeline has little impact on the final results; (2) The operators are also cumulative—the final results of combining several efficiency methods can be estimated by combining the results of individual methods. These observations deepen our understanding of efficiency operators and provide useful guidelines for building them in real-world applications.

Deep neural networks (DNNs) are often used for text classification due to their high accuracy. However, DNNs can be computationally intensive, requiring millions of parameters and large amounts of labeled data, which can make them expensive to use, to optimize, and to transfer to out-of-distribution (OOD) cases in practice. In this paper, we propose a non-parametric alternative to DNNs that’s easy, lightweight, and universal in text classification: a combination of a simple compressor like gzip with a k-nearest-neighbor classifier. Without any training parameters, our method achieves results that are competitive with non-pretrained deep learning methods on six in-distribution datasets.It even outperforms BERT on all five OOD datasets, including four low-resource languages. Our method also excels in the few-shot setting, where labeled data are too scarce to train DNNs effectively.

Various techniques have been developed in recent years to improve dense retrieval (DR), such as unsupervised contrastive learning and pseudo-query generation. Existing DRs, however, often suffer from effectiveness tradeoffs between supervised and zero-shot retrieval, which some argue was due to the limited model capacity. We contradict this hypothesis and show that a generalizable DR can be trained to achieve high accuracy in both supervised and zero-shot retrieval without increasing model size. In particular, we systematically examine the contrastive learning of DRs, under the framework of Data Augmentation (DA). Our study shows that common DA practices such as query augmentation with generative models and pseudo-relevance label creation using a cross-encoder, are often inefficient and sub-optimal. We hence propose a new DA approach with diverse queries and sources of supervision to progressively train a generalizable DR. As a result, DRAGON, our Dense Retriever trained with diverse AuGmentatiON, is the first BERT-base-sized DR to achieve state-of-the-art effectiveness in both supervised and zero-shot evaluations and even competes with models using more complex late interaction.

In this study, we highlight the importance of enhancing the quality of pretraining data in multilingual language models. Existing web crawls have demonstrated quality issues, particularly in the context of low-resource languages. Consequently, we introduce a new multilingual pretraining corpus for 16 African languages, designed by carefully auditing existing pretraining corpora to understand and rectify prevalent quality issues. To compile this dataset, we undertake a rigorous examination of current data sources for thirteen languages within one of the most extensive multilingual web crawls, mC4, and extract cleaner data through meticulous auditing and improved web crawling strategies. Subsequently, we pretrain a new T5-based model on this dataset and evaluate its performance on multiple downstream tasks. Our model demonstrates better downstream effectiveness over existing pretrained models across four NLP tasks, underscoring the critical role data quality plays in pretraining language models in low-resource scenarios. Specifically, on cross-lingual QA evaluation, our new model is more than twice as effective as multilingual T5. All code, data and models are publicly available at https://github.com/castorini/AfriTeVa-keji.

The emerging paradigm of generative retrieval re-frames the classic information retrieval problem into a sequence-to-sequence modeling task, forgoing external indices and encoding an entire document corpus within a single Transformer. Although many different approaches have been proposed to improve the effectiveness of generative retrieval, they have only been evaluated on document corpora on the order of 100K in size. We conduct the first empirical study of generative retrieval techniques across various corpus scales, ultimately scaling up to the entire MS MARCO passage ranking task with a corpus of 8.8M passages and evaluating model sizes up to 11B parameters. We uncover several findings about scaling generative retrieval to millions of passages; notably, the central importance of using synthetic queries as document representations during indexing, the ineffectiveness of existing proposed architecture modifications when accounting for compute cost, and the limits of naively scaling model parameters with respect to retrieval performance. While we find that generative retrieval is competitive with state-of-the-art dual encoders on small corpora, scaling to millions of passages remains an important and unsolved challenge. We believe these findings will be valuable for the community to clarify the current state of generative retrieval, highlight the unique challenges, and inspire new research directions.

Multilingual information retrieval (MLIR) is a crucial yet challenging task due to the need for human annotations in multiple languages, making training data creation labor-intensive. In this paper, we introduce mAggretriever, which effectively leverages semantic and lexical features from pre-trained multilingual transformers (e.g., mBERT and XLM-R) for dense retrieval. To enhance training and inference efficiency, we employ approximate masked-language modeling prediction for computing lexical features, reducing 70–85% GPU memory requirement for mAggretriever fine-tuning. Empirical results demonstrate that mAggretriever, fine-tuned solely on English training data, surpasses existing state-of-the-art multilingual dense retrieval models that undergo further training on large-scale MLIR training data. Our code is available at url.

We present Spacerini, a tool that integrates the Pyserini toolkit for reproducible information retrieval research with Hugging Face to enable the seamless construction and deployment of interactive search engines. Spacerini makes state-of-the-art sparse and dense retrieval models more accessible to non-IR practitioners while minimizing deployment effort. This is useful for NLP researchers who want to better understand and validate their research by performing qualitative analyses of training corpora, for IR researchers who want to demonstrate new retrieval models integrated into the growing Pyserini ecosystem, and for third parties reproducing the work of other researchers. Spacerini is open source and includes utilities for loading, preprocessing, indexing, and deploying search engines locally and remotely. We demonstrate a portfolio of 13 search engines created with Spacerini for different use cases.

Pre-trained language models have been successful in many knowledge-intensive NLP tasks. However, recent work has shown that models such as BERT are not “structurally ready” to aggregate textual information into a [CLS] vector for dense passage retrieval (DPR). This “lack of readiness” results from the gap between language model pre-training and DPR fine-tuning. Previous solutions call for computationally expensive techniques such as hard negative mining, cross-encoder distillation, and further pre-training to learn a robust DPR model. In this work, we instead propose to fully exploit knowledge in a pre-trained language model for DPR by aggregating the contextualized token embeddings into a dense vector, which we call agg★. By concatenating vectors from the [CLS] token and agg★, our Aggretriever model substantially improves the effectiveness of dense retrieval models on both in-domain and zero-shot evaluations without introducing substantial training overhead. Code is available at https://github.com/castorini/dhr.

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/.

Pretrained language models represent the state of the art in NLP, but the successful construction of such models often requires large amounts of data and computational resources. Thus, the paucity of data for low-resource languages impedes the development of robust NLP capabilities for these languages. There has been some recent success in pretraining encoderonly models solely on a combination of lowresource African languages, exemplified by AfriBERTa. In this work, we extend the approach of “small data” pretraining to encoder– decoder models. We introduce AfriTeVa, a family of sequence-to-sequence models derived from T5 that are pretrained on 10 African languages from scratch. With a pretraining corpus of only around 1GB, we show that it is possible to achieve competitive downstream effectiveness for machine translation and text classification, compared to larger models trained on much more data. All the code and model checkpoints described in this work are publicly available at https://github.com/castorini/afriteva.

In information retrieval (IR), candidate set pruning has been commonly used to speed up two-stage relevance ranking. However, such an approach lacks accurate error control and often trades accuracy against computational efficiency in an empirical fashion, missing theoretical guarantees. In this paper, we propose the concept of certified error control of candidate set pruning for relevance ranking, which means that the test error after pruning is guaranteed to be controlled under a user-specified threshold with high probability. Both in-domain and out-of-domain experiments show that our method successfully prunes the first-stage retrieved candidate sets to improve the second-stage reranking speed while satisfying the pre-specified accuracy constraints in both settings. For example, on MS MARCO Passage v1, our method reduces the average candidate set size from 1000 to 27, increasing reranking speed by about 37 times, while keeping MRR@10 greater than a pre-specified value of 0.38 with about 90% empirical coverage. In contrast, empirical baselines fail to meet such requirements. Code and data are available at: https://github.com/alexlimh/CEC-Ranking.

Language diversity in NLP is critical in enabling the development of tools for a wide range of users.However, there are limited resources for building such tools for many languages, particularly those spoken in Africa.For search, most existing datasets feature few or no African languages, directly impacting researchers’ ability to build and improve information access capabilities in those languages.Motivated by this, we created AfriCLIRMatrix, a test collection for cross-lingual information retrieval research in 15 diverse African languages.In total, our dataset contains 6 million queries in English and 23 million relevance judgments automatically mined from Wikipedia inter-language links, covering many more African languages than any existing information retrieval test collection.In addition, we release BM25, dense retrieval, and sparse–dense hybrid baselines to provide a starting point for the development of future systems.We hope that these efforts can spur additional work in search for African languages.AfriCLIRMatrix can be downloaded at https://github.com/castorini/africlirmatrix.

End-to-end automatic speech recognition systems represent the state of the art, but they rely on thousands of hours of manually annotated speech for training, as well as heavyweight computation for inference. Of course, this impedes commercialization since most companies lack vast human and computational resources. In this paper, we explore training and deploying an ASR system in the label-scarce, compute-limited setting. To reduce human labor, we use a third-party ASR system as a weak supervision source, supplemented with labeling functions derived from implicit user feedback. To accelerate inference, we propose to route production-time queries across a pool of CUDA graphs of varying input lengths, the distribution of which best matches the traffic’s. Compared to our third-party ASR, we achieve a relative improvement in word-error rate of 8% and a speedup of 600%. Our system, called SpeechNet, currently serves 12 million queries per day on our voice-enabled smart television. To our knowledge, this is the first time a large-scale, Wav2vec-based deployment has been described in the academic literature.

The application of natural language processing (NLP) to cancer pathology reports has been focused on detecting cancer cases, largely ignoring precancerous cases. Improving the characterization of precancerous adenomas assists in developing diagnostic tests for early cancer detection and prevention, especially for colorectal cancer (CRC). Here we developed transformer-based deep neural network NLP models to perform the CRC phenotyping, with the goal of extracting precancerous lesion attributes and distinguishing cancer and precancerous cases. We achieved 0.914 macro-F1 scores for classifying patients into negative, non-advanced adenoma, advanced adenoma and CRC. We further improved the performance to 0.923 using an ensemble of classifiers for cancer status classification and lesion size named-entity recognition (NER). Our results demonstrated the potential of using NLP to leverage real-world health record data to facilitate the development of diagnostic tests for early cancer prevention.

With the recent success of dense retrieval methods based on bi-encoders, studies have applied this approach to various interesting downstream retrieval tasks with good efficiency and in-domain effectiveness.Recently, we have also seen the presence of dense retrieval models in Math Information Retrieval (MIR) tasks,but the most effective systems remain classic retrieval methods that consider hand-crafted structure features.In this work, we try to combine the best of both worlds: a well-defined structure search method for effective formula search and efficient bi-encoder dense retrieval models to capture contextual similarities.Specifically, we have evaluated two representative bi-encoder models for token-level and passage-level dense retrieval on recent MIR tasks.Our results show that bi-encoder models are highly complementary to existing structure search methods, and we are able to advance the state-of-the-art on MIR datasets.

In-context learning using large language models has recently shown surprising results for semantic parsing tasks such as Text-to-SQL translation.Prompting GPT-3 or Codex using several examples of question-SQL pairs can produce excellent results, comparable to state-of-the-art finetuning-based models.However, existing work primarily focuses on English datasets, and it is unknown whether large language models can serve as competitive semantic parsers for other languages.To bridge this gap, our work focuses on cross-lingual Text-to-SQL semantic parsing for translating non-English utterances into SQL queries based on an English schema.We consider a zero-shot transfer learning setting with the assumption that we do not have any labeled examples in the target language (but have annotated examples in English).This work introduces the XRICL framework, which learns to retrieve relevant English exemplars for a given query to construct prompts.We also include global translation exemplars for a target language to facilitate the translation process for large language models.To systematically evaluate our model, we construct two new benchmark datasets, XSpider and XKaggle-dbqa, which include questions in Chinese, Vietnamese, Farsi, and Hindi.Our experiments show that XRICL effectively leverages large pre-trained language models to outperform existing baselines.Data and code are publicly available at https://github.com/Impavidity/XRICL.

We focus on the cross-lingual Text-to-SQL semantic parsing task,where the parsers are expected to generate SQL for non-English utterances based on English database schemas.Intuitively, English translation as side information is an effective way to bridge the language gap,but noise introduced by the translation system may affect parser effectiveness.In this work, we propose a Representation Mixup Framework (Rex) for effectively exploiting translations in the cross-lingual Text-to-SQL task.Particularly, it uses a general encoding layer, a transition layer, and a target-centric layer to properly guide the information flow of the English translation.Experimental results on CSpider and VSpider show that our framework can benefit from cross-lingual training and improve the effectiveness of semantic parsers, achieving state-of-the-art performance.

The bi-encoder design of dense passage retriever (DPR) is a key factor to its success in open-domain question answering (QA), yet it is unclear how DPR’s question encoder and passage encoder individually contributes to overall performance, which we refer to as the encoder attribution problem. The problem is important as it helps us identify the factors that affect individual encoders to further improve overall performance. In this paper, we formulate our analysis under a probabilistic framework called encoder marginalization, where we quantify the contribution of a single encoder by marginalizing other variables. First, we find that the passage encoder contributes more than the question encoder to in-domain retrieval accuracy. Second, we demonstrate how to find the affecting factors for each encoder, where we train DPR with different amounts of data and use encoder marginalization to analyze the results. We find that positive passage overlap and corpus coverage of training data have big impacts on the passage encoder, while the question encoder is mainly affected by training sample complexity under this setting. Based on this framework, we can devise data-efficient training regimes: for example, we manage to train a passage encoder on SQuAD using 60% less training data without loss of accuracy.

Pretrained multilingual language models have been shown to work well on many languages for a variety of downstream NLP tasks. However, these models are known to require a lot of training data. This consequently leaves out a huge percentage of the world’s languages as they are under-resourced. Furthermore, a major motivation behind these models is that lower-resource languages benefit from joint training with higher-resource languages. In this work, we challenge this assumption and present the first attempt at training a multilingual language model on only low-resource languages. We show that it is possible to train competitive multilingual language models on less than 1 GB of text. Our model, named AfriBERTa, covers 11 African languages, including the first language model for 4 of these languages. Evaluations on named entity recognition and text classification spanning 10 languages show that our model outperforms mBERT and XLM-Rin several languages and is very competitive overall. Results suggest that our “small data” approach based on similar languages may sometimes work better than joint training on large datasets with high-resource languages. Code, data and models are released at https://github.com/keleog/afriberta.

We present Mr. TyDi, a multi-lingual benchmark dataset for mono-lingual retrieval in eleven typologically diverse languages, designed to evaluate ranking with learned dense representations. The goal of this resource is to spur research in dense retrieval techniques in non-English languages, motivated by recent observations that existing techniques for representation learning perform poorly when applied to out-of-distribution data. As a starting point, we provide zero-shot baselines for this new dataset based on a multi-lingual adaptation of DPR that we call “mDPR”. Experiments show that although the effectiveness of mDPR is much lower than BM25, dense representations nevertheless appear to provide valuable relevance signals, improving BM25 results in sparse–dense hybrids. In addition to analyses of our results, we also discuss future challenges and present a research agenda in multi-lingual dense retrieval. Mr. TyDi can be downloaded at https://github.com/castorini/mr.tydi.

Dense retrieval has shown great success for passage ranking in English. However, its effectiveness for non-English languages remains unexplored due to limitation in training resources. In this work, we explore different transfer techniques for document ranking from English annotations to non-English languages. Our experiments reveal that zero-shot model-based transfer using mBERT improves search quality. We find that weakly-supervised target language transfer is competitive compared to generation-based target language transfer, which requires translation models.

In selective prediction, a classifier is allowed to abstain from making predictions on low-confidence examples. Though this setting is interesting and important, selective prediction has rarely been examined in natural language processing (NLP) tasks. To fill this void in the literature, we study in this paper selective prediction for NLP, comparing different models and confidence estimators. We further propose a simple error regularization trick that improves confidence estimation without substantially increasing the computation budget. We show that recent pre-trained transformer models simultaneously improve both model accuracy and confidence estimation effectiveness. We also find that our proposed regularization improves confidence estimation and can be applied to other relevant scenarios, such as using classifier cascades for accuracy–efficiency trade-offs. Source code for this paper can be found at https://github.com/castorini/transformers-selective.

This work explores a framework for fact verification that leverages pretrained sequence-to-sequence transformer models for sentence selection and label prediction, two key sub-tasks in fact verification. Most notably, improving on previous pointwise aggregation approaches for label prediction, we take advantage of T5 using a listwise approach coupled with data augmentation. With this enhancement, we observe that our label prediction stage is more robust to noise and capable of verifying complex claims by jointly reasoning over multiple pieces of evidence. Experimental results on the FEVER task show that our system attains a FEVER score of 75.87% on the blind test set. This puts our approach atop the competitive FEVER leaderboard at the time of our work, scoring higher than the second place submission by almost two points in label accuracy and over one point in FEVER score.

The semantics of a text is manifested not only by what is read but also by what is not read. In this article, we will study how those implicit “not read” information such as end-of-paragraph () and end-of-sequence () affect the quality of text generation. Specifically, we find that the pre-trained language model GPT2 can generate better continuations by learning to generate the in the fine-tuning stage. Experimental results on English story generation show that can lead to higher BLEU scores and lower perplexity. We also conduct experiments on a self-collected Chinese essay dataset with Chinese-GPT2, a character level LM without and during pre-training. Experimental results show that the Chinese GPT2 can generate better essay endings with .

The goal of text ranking is to generate an ordered list of texts retrieved from a corpus in response to a query for a particular task. Although the most common formulation of text ranking is search, instances of the task can also be found in many text processing applications. This tutorial provides an overview of text ranking with neural network architectures known as transformers, of which BERT (Bidirectional Encoder Representations from Transformers) is the best-known example. These models produce high quality results across many domains, tasks, and settings. This tutorial, which is based on the preprint of a forthcoming book to be published by Morgan and & Claypool under the Synthesis Lectures on Human Language Technologies series, provides an overview of existing work as a single point of entry for practitioners who wish to deploy transformers for text ranking in real-world applications and researchers who wish to pursue work in this area. We cover a wide range of techniques, grouped into two categories: transformer models that perform reranking in multi-stage ranking architectures and learned dense representations that perform ranking directly.

It is well known that rerankers built on pretrained transformer models such as BERT have dramatically improved retrieval effectiveness in many tasks. However, these gains have come at substantial costs in terms of efficiency, as noted by many researchers. In this work, we show that it is possible to retain the benefits of transformer-based rerankers in a multi-stage reranking pipeline by first using feature-based learning-to-rank techniques to reduce the number of candidate documents under consideration without adversely affecting their quality in terms of recall. Applied to the MS MARCO passage and document ranking tasks, we are able to achieve the same level of effectiveness, but with up to 18× increase in efficiency. Furthermore, our techniques are orthogonal to other methods focused on accelerating transformer inference, and thus can be combined for even greater efficiency gains. A higher-level message from our work is that, even though pretrained transformers dominate the modern IR landscape, there are still important roles for “traditional” LTR techniques, and that we should not forget history.

We present an efficient training approach to text retrieval with dense representations that applies knowledge distillation using the ColBERT late-interaction ranking model. Specifically, we propose to transfer the knowledge from a bi-encoder teacher to a student by distilling knowledge from ColBERT’s expressive MaxSim operator into a simple dot product. The advantage of the bi-encoder teacher–student setup is that we can efficiently add in-batch negatives during knowledge distillation, enabling richer interactions between teacher and student models. In addition, using ColBERT as the teacher reduces training cost compared to a full cross-encoder. Experiments on the MS MARCO passage and document ranking tasks and data from the TREC 2019 Deep Learning Track demonstrate that our approach helps models learn robust representations for dense retrieval effectively and efficiently.

Fine-tuned pre-trained transformers achieve the state of the art in passage reranking. Unfortunately, how they make their predictions remains vastly unexplained, especially at the end-to-end, input-to-output level. Little known is how tokens, layers, and passages precisely contribute to the final prediction. In this paper, we address this gap by leveraging the recently developed information bottlenecks for attribution (IBA) framework. On BERT-based models for passage reranking, we quantitatively demonstrate the framework’s veracity in extracting attribution maps, from which we perform detailed, token-wise analysis about how predictions are made. Overall, we find that BERT still cares about exact token matching for reranking; the [CLS] token mainly gathers information for predictions at the last layer; top-ranked passages are robust to token removal; and BERT fine-tuned on MSMARCO has positional bias towards the start of the passage.

This work describes the adaptation of a pretrained sequence-to-sequence model to the task of scientific claim verification in the biomedical domain. We propose a system called VerT5erini that exploits T5 for abstract retrieval, sentence selection, and label prediction, which are three critical sub-tasks of claim verification. We evaluate our pipeline on SciFACT, a newly curated dataset that requires models to not just predict the veracity of claims but also provide relevant sentences from a corpus of scientific literature that support the prediction. Empirically, our system outperforms a strong baseline in each of the three sub-tasks. We further show VerT5erini’s ability to generalize to two new datasets of COVID-19 claims using evidence from the CORD-19 corpus.

The slow speed of BERT has motivated much research on accelerating its inference, and the early exiting idea has been proposed to make trade-offs between model quality and efficiency. This paper aims to address two weaknesses of previous work: (1) existing fine-tuning strategies for early exiting models fail to take full advantage of BERT; (2) methods to make exiting decisions are limited to classification tasks. We propose a more advanced fine-tuning strategy and a learning-to-exit module that extends early exiting to tasks other than classification. Experiments demonstrate improved early exiting for BERT, with better trade-offs obtained by the proposed fine-tuning strategy, successful application to regression tasks, and the possibility to combine it with other acceleration methods. Source code can be found at https://github.com/castorini/berxit.

In the paraphrase generation task, source sentences often contain phrases that should not be altered. Which phrases, however, can be context dependent and can vary by application. Our solution to this challenge is to provide the user with explicit tags that can be placed around any arbitrary segment of text to mean “don’t change me!” when generating a paraphrase; the model learns to explicitly copy these phrases to the output. The contribution of this work is a novel data generation technique using distant supervision that allows us to start with a pretrained sequence-to-sequence model and fine-tune a paraphrase generator that exhibits this behavior, allowing user-controllable paraphrase generation. Additionally, we modify the loss during fine-tuning to explicitly encourage diversity in model output. Our technique is language agnostic, and we report experiments in English and Chinese.

Multi-task dense retrieval models can be used to retrieve documents from a common corpus (e.g., Wikipedia) for different open-domain question-answering (QA) tasks. However, Karpukhin et al. (2020) shows that jointly learning different QA tasks with one dense model is not always beneficial due to corpus inconsistency. For example, SQuAD only focuses on a small set of Wikipedia articles while datasets like NQ and Trivia cover more entries, and joint training on their union can cause performance degradation. To solve this problem, we propose to train individual dense passage retrievers (DPR) for different tasks and aggregate their predictions during test time, where we use uncertainty estimation as weights to indicate how probable a specific query belongs to each expert’s expertise. Our method reaches state-of-the-art performance on 5 benchmark QA datasets, with up to 10% improvement in top-100 accuracy compared to a joint-training multi-task DPR on SQuAD. We also show that our method handles corpus inconsistency better than the joint-training DPR on a mixed subset of different QA datasets. Code and data are available at https://github.com/alexlimh/DPR_MUF.

In this paper, we address unsupervised chunking as a new task of syntactic structure induction, which is helpful for understanding the linguistic structures of human languages as well as processing low-resource languages. We propose a knowledge-transfer approach that heuristically induces chunk labels from state-of-the-art unsupervised parsing models; a hierarchical recurrent neural network (HRNN) learns from such induced chunk labels to smooth out the noise of the heuristics. Experiments show that our approach largely bridges the gap between supervised and unsupervised chunking.

Query auto completion (QAC) is the task of predicting a search engine user’s final query from their intermediate, incomplete query. In this paper, we extend QAC to the streaming voice search setting, where automatic speech recognition systems produce intermediate transcriptions as users speak. Naively applying existing methods fails because the intermediate transcriptions often don’t form prefixes or even substrings of the final transcription. To address this issue, we propose to condition QAC approaches on intermediate transcriptions to complete voice queries. We evaluate our models on a speech-enabled smart television with real-life voice search traffic, finding that this ASR-aware conditioning improves the completion quality. Our best method obtains an 18% relative improvement in mean reciprocal rank over previous methods.

This paper describes a compact and effective model for low-latency passage retrieval in conversational search based on learned dense representations. Prior to our work, the state-of-the-art approach uses a multi-stage pipeline comprising conversational query reformulation and information retrieval modules. Despite its effectiveness, such a pipeline often includes multiple neural models that require long inference times. In addition, independently optimizing each module ignores dependencies among them. To address these shortcomings, we propose to integrate conversational query reformulation directly into a dense retrieval model. To aid in this goal, we create a dataset with pseudo-relevance labels for conversational search to overcome the lack of training data and to explore different training strategies. We demonstrate that our model effectively rewrites conversational queries as dense representations in conversational search and open-domain question answering datasets. Finally, after observing that our model learns to adjust the L2 norm of query token embeddings, we leverage this property for hybrid retrieval and to support error analysis.

Recent work has shown that dense passage retrieval techniques achieve better ranking accuracy in open-domain question answering compared to sparse retrieval techniques such as BM25, but at the cost of large space and memory requirements. In this paper, we analyze the redundancy present in encoded dense vectors and show that the default dimension of 768 is unnecessarily large. To improve space efficiency, we propose a simple unsupervised compression pipeline that consists of principal component analysis (PCA), product quantization, and hybrid search. We further investigate other supervised baselines and find surprisingly that unsupervised PCA outperforms them in some settings. We perform extensive experiments on five question answering datasets and demonstrate that our best pipeline achieves good accuracy–space trade-offs, for example, 48× compression with less than 3% drop in top-100 retrieval accuracy on average or 96× compression with less than 4% drop. Code and data are available at http://pyserini.io/.

The task of semantic code search is to retrieve code snippets from a source code corpus based on an information need expressed in natural language. The semantic gap between natural language and programming languages has for long been regarded as one of the most significant obstacles to the effectiveness of keyword-based information retrieval (IR) methods. It is a common assumption that “traditional” bag-of-words IR methods are poorly suited for semantic code search: our work empirically investigates this assumption. Specifically, we examine the effectiveness of two traditional IR methods, namely BM25 and RM3, on the CodeSearchNet Corpus, which consists of natural language queries paired with relevant code snippets. We find that the two keyword-based methods outperform several pre-BERT neural models. We also compare several code-specific data pre-processing strategies and find that specialized tokenization improves effectiveness.

Pre-trained language models such as BERT have shown their effectiveness in various tasks. Despite their power, they are known to be computationally intensive, which hinders real-world applications. In this paper, we introduce early exiting BERT for document ranking. With a slight modification, BERT becomes a model with multiple output paths, and each inference sample can exit early from these paths. In this way, computation can be effectively allocated among samples, and overall system latency is significantly reduced while the original quality is maintained. Our experiments on two document ranking datasets demonstrate up to 2.5x inference speedup with minimal quality degradation. The source code of our implementation can be found at https://github.com/castorini/earlyexiting-monobert.

Researchers have proposed simple yet effective techniques for the retrieval problem based on using BERT as a relevance classifier to rerank initial candidates from keyword search. In this work, we tackle the challenge of fine-tuning these models for specific domains in a data and computationally efficient manner. Typically, researchers fine-tune models using corpus-specific labeled data from sources such as TREC. We first answer the question: How much data of this type do we need? Recognizing that the most computationally efficient training is no training, we explore zero-shot ranking using BERT models that have already been fine-tuned with the large MS MARCO passage retrieval dataset. We arrive at the surprising and novel finding that “some” labeled in-domain data can be worse than none at all.

We present Covidex, a search engine that exploits the latest neural ranking models to provide information access to the COVID-19 Open Research Dataset curated by the Allen Institute for AI. Our system has been online and serving users since late March 2020. The Covidex is the user application component of our three-pronged strategy to develop technologies for helping domain experts tackle the ongoing global pandemic. In addition, we provide robust and easy-to-use keyword search infrastructure that exploits mature fusion-based methods as well as standalone neural ranking models that can be incorporated into other applications. These techniques have been evaluated in the multi-round TREC-COVID challenge: Our infrastructure and baselines have been adopted by many participants, including some of the best systems. In round 3, we submitted the highest-scoring run that took advantage of previous training data and the second-highest fully automatic run. In rounds 4 and 5, we submitted the highest-scoring fully automatic runs.

Cydex is a platform that provides neural search infrastructure for domain-specific scholarly literature. The platform represents an abstraction of Covidex, our recently developed full-stack open-source search engine for the COVID-19 Open Research Dataset (CORD-19) from AI2. While Covidex takes advantage of the latest best practices for keyword search using the popular Lucene search library as well as state-of-the-art neural ranking models using T5, parts of the system were hard coded to only work with CORD-19. This paper describes our efforts to generalize Covidex into Cydex, which can be applied to scholarly literature in different domains. By decoupling corpus-specific configurations from the frontend implementation, we are able to demonstrate the generality of Cydex on two very different corpora: the ACL Anthology and a collection of hydrology abstracts. Our platform is entirely open source and available at cydex.ai.

While internalized “implicit knowledge” in pretrained transformers has led to fruitful progress in many natural language understanding tasks, how to most effectively elicit such knowledge remains an open question. Based on the text-to-text transfer transformer (T5) model, this work explores a template-based approach to extract implicit knowledge for commonsense reasoning on multiple-choice (MC) question answering tasks. Experiments on three representative MC datasets show the surprisingly good performance of our simple template, coupled with a logit normalization technique for disambiguation. Furthermore, we verify that our proposed template can be easily extended to other MC tasks with contexts such as supporting facts in open-book question answering settings. Starting from the MC task, this work initiates further research to find generic natural language templates that can effectively leverage stored knowledge in pretrained models.

Fine-tuned variants of BERT are able to achieve state-of-the-art accuracy on many natural language processing tasks, although at significant computational costs. In this paper, we verify BERT’s effectiveness for document classification and investigate the extent to which BERT-level effectiveness can be obtained by different baselines, combined with knowledge distillation—a popular model compression method. The results show that BERT-level effectiveness can be achieved by a single-layer LSTM with at least 40× fewer FLOPS and only ∼3% parameters. More importantly, this study analyzes the limits of knowledge distillation as we distill BERT’s knowledge all the way down to linear models—a relevant baseline for the task. We report substantial improvement in effectiveness for even the simplest models, as they capture the knowledge learnt by BERT.

We describe Howl, an open-source wake word detection toolkit with native support for open speech datasets such as Mozilla Common Voice (MCV) and Google Speech Commands (GSC). We report benchmark results of various models supported by our toolkit on GSC and our own freely available wake word detection dataset, built from MCV. One of our models is deployed in Firefox Voice, a plugin enabling speech interactivity for the Firefox web browser. Howl represents, to the best of our knowledge, the first fully productionized, open-source wake word detection toolkit with a web browser deployment target. Our codebase is at howl.ai.

Large-scale pre-trained language models such as BERT have brought significant improvements to NLP applications. However, they are also notorious for being slow in inference, which makes them difficult to deploy in real-time applications. We propose a simple but effective method, DeeBERT, to accelerate BERT inference. Our approach allows samples to exit earlier without passing through the entire model. Experiments show that DeeBERT is able to save up to ~40% inference time with minimal degradation in model quality. Further analyses show different behaviors in the BERT transformer layers and also reveal their redundancy. Our work provides new ideas to efficiently apply deep transformer-based models to downstream tasks. Code is available at https://github.com/castorini/DeeBERT.

In natural language processing, a recently popular line of work explores how to best report the experimental results of neural networks. One exemplar publication, titled “Show Your Work: Improved Reporting of Experimental Results” (Dodge et al., 2019), advocates for reporting the expected validation effectiveness of the best-tuned model, with respect to the computational budget. In the present work, we critically examine this paper. As far as statistical generalizability is concerned, we find unspoken pitfalls and caveats with this approach. We analytically show that their estimator is biased and uses error-prone assumptions. We find that the estimator favors negative errors and yields poor bootstrapped confidence intervals. We derive an unbiased alternative and bolster our claims with empirical evidence from statistical simulation. Our codebase is at https://github.com/castorini/meanmax.

A number of researchers have recently questioned the necessity of increasingly complex neural network (NN) architectures. In particular, several recent papers have shown that simpler, properly tuned models are at least competitive across several NLP tasks. In this work, we show that this is also the case for text generation from structured and unstructured data. We consider neural table-to-text generation and neural question generation (NQG) tasks for text generation from structured and unstructured data, respectively. Table-to-text generation aims to generate a description based on a given table, and NQG is the task of generating a question from a given passage where the generated question can be answered by a certain sub-span of the passage using NN models. Experimental results demonstrate that a basic attention-based seq2seq model trained with the exponential moving average technique achieves the state of the art in both tasks. Code is available at https://github.com/h-shahidi/2birds-gen.

The Neural Covidex is a search engine that exploits the latest neural ranking architectures to provide information access to the COVID-19 Open Research Dataset (CORD-19) curated by the Allen Institute for AI. It exists as part of a suite of tools we have developed to help domain experts tackle the ongoing global pandemic. We hope that improved information access capabilities to the scientific literature can inform evidence-based decision making and insight generation.

This work proposes the use of a pretrained sequence-to-sequence model for document ranking. Our approach is fundamentally different from a commonly adopted classification-based formulation based on encoder-only pretrained transformer architectures such as BERT. We show how a sequence-to-sequence model can be trained to generate relevance labels as “target tokens”, and how the underlying logits of these target tokens can be interpreted as relevance probabilities for ranking. Experimental results on the MS MARCO passage ranking task show that our ranking approach is superior to strong encoder-only models. On three other document retrieval test collections, we demonstrate a zero-shot transfer-based approach that outperforms previous state-of-the-art models requiring in-domain cross-validation. Furthermore, we find that our approach significantly outperforms an encoder-only architecture in a data-poor setting. We investigate this observation in more detail by varying target tokens to probe the model’s use of latent knowledge. Surprisingly, we find that the choice of target tokens impacts effectiveness, even for words that are closely related semantically. This finding sheds some light on why our sequence-to-sequence formulation for document ranking is effective. Code and models are available at pygaggle.ai.

We tackle the challenge of cross-lingual training of neural document ranking models for mono-lingual retrieval, specifically leveraging relevance judgments in English to improve search in non-English languages. Our work successfully applies multi-lingual BERT (mBERT) to document ranking and additionally compares against a number of alternatives: translating the training data, translating documents, multi-stage hybrids, and ensembles. Experiments on test collections in six different languages from diverse language families reveal many interesting findings: model-based relevance transfer using mBERT can significantly improve search quality in (non-English) mono-lingual retrieval, but other “low resource” approaches are competitive as well.

Pretrained transformers achieve the state of the art across tasks in natural language processing, motivating researchers to investigate their inner mechanisms. One common direction is to understand what features are important for prediction. In this paper, we apply information bottlenecks to analyze the attribution of each feature for prediction on a black-box model. We use BERT as the example and evaluate our approach both quantitatively and qualitatively. We show the effectiveness of our method in terms of attribution and the ability to provide insight into how information flows through layers. We demonstrate that our technique outperforms two competitive methods in degradation tests on four datasets. Code is available at https://github.com/bazingagin/IBA.

This paper explores the problem of ranking short social media posts with respect to user queries using neural networks. Instead of starting with a complex architecture, we proceed from the bottom up and examine the effectiveness of a simple, word-level Siamese architecture augmented with attention-based mechanisms for capturing semantic “soft” matches between query and post tokens. Extensive experiments on datasets from the TREC Microblog Tracks show that our simple models not only achieve better effectiveness than existing approaches that are far more complex or exploit a more diverse set of relevance signals, but are also much faster.

Neural network models for many NLP tasks have grown increasingly complex in recent years, making training and deployment more difficult. A number of recent papers have questioned the necessity of such architectures and found that well-executed, simpler models are quite effective. We show that this is also the case for document classification: in a large-scale reproducibility study of several recent neural models, we find that a simple BiLSTM architecture with appropriate regularization yields accuracy and F1 that are either competitive or exceed the state of the art on four standard benchmark datasets. Surprisingly, our simple model is able to achieve these results without attention mechanisms. While these regularization techniques, borrowed from language modeling, are not novel, to our knowledge we are the first to apply them in this context. Our work provides an open-source platform and the foundation for future work in document classification.

Consumers dissatisfied with the normal dispute resolution process provided by an e-commerce company’s customer service agents have the option of escalating their complaints by filing grievances with a government authority. This paper tackles the challenge of monitoring ongoing text chat dialogues to identify cases where the customer expresses such an intent, providing triage and prioritization for a separate pool of specialized agents specially trained to handle more complex situations. We describe a hybrid model that tackles this challenge by integrating recurrent neural networks with manually-engineered features. Experiments show that both components are complementary and contribute to overall recall, outperforming competitive baselines. A trial online deployment of our model demonstrates its business value in improving customer service.

We demonstrate an end-to-end question answering system that integrates BERT with the open-source Anserini information retrieval toolkit. In contrast to most question answering and reading comprehension models today, which operate over small amounts of input text, our system integrates best practices from IR with a BERT-based reader to identify answers from a large corpus of Wikipedia articles in an end-to-end fashion. We report large improvements over previous results on a standard benchmark test collection, showing that fine-tuning pretrained BERT with SQuAD is sufficient to achieve high accuracy in identifying answer spans.

Semantic similarity modeling is central to many NLP problems such as natural language inference and question answering. Syntactic structures interact closely with semantics in learning compositional representations and alleviating long-range dependency issues. How-ever, such structure priors have not been well exploited in previous work for semantic mod-eling. To examine their effectiveness, we start with the Pairwise Word Interaction Model, one of the best models according to a recent reproducibility study, then introduce components for modeling context and structure using multi-layer BiLSTMs and TreeLSTMs. In addition, we introduce residual connections to the deep convolutional neural network component of the model. Extensive evaluations on eight benchmark datasets show that incorporating structural information contributes to consistent improvements over strong baselines.

This paper applies BERT to ad hoc document retrieval on news articles, which requires addressing two challenges: relevance judgments in existing test collections are typically provided only at the document level, and documents often exceed the length that BERT was designed to handle. Our solution is to aggregate sentence-level evidence to rank documents. Furthermore, we are able to leverage passage-level relevance judgments fortuitously available in other domains to fine-tune BERT models that are able to capture cross-domain notions of relevance, and can be directly used for ranking news articles. Our simple neural ranking models achieve state-of-the-art effectiveness on three standard test collections.

Multilingual knowledge graphs (KGs), such as YAGO and DBpedia, represent entities in different languages. The task of cross-lingual entity alignment is to match entities in a source language with their counterparts in target languages. In this work, we investigate embedding-based approaches to encode entities from multilingual KGs into the same vector space, where equivalent entities are close to each other. Specifically, we apply graph convolutional networks (GCNs) to combine multi-aspect information of entities, including topological connections, relations, and attributes of entities, to learn entity embeddings. To exploit the literal descriptions of entities expressed in different languages, we propose two uses of a pretrained multilingual BERT model to bridge cross-lingual gaps. We further propose two strategies to integrate GCN-based and BERT-based modules to boost performance. Extensive experiments on two benchmark datasets demonstrate that our method significantly outperforms existing systems.

A core problem of information retrieval (IR) is relevance matching, which is to rank documents by relevance to a user’s query. On the other hand, many NLP problems, such as question answering and paraphrase identification, can be considered variants of semantic matching, which is to measure the semantic distance between two pieces of short texts. While at a high level both relevance and semantic matching require modeling textual similarity, many existing techniques for one cannot be easily adapted to the other. To bridge this gap, we propose a novel model, HCAN (Hybrid Co-Attention Network), that comprises (1) a hybrid encoder module that includes ConvNet-based and LSTM-based encoders, (2) a relevance matching module that measures soft term matches with importance weighting at multiple granularities, and (3) a semantic matching module with co-attention mechanisms that capture context-aware semantic relatedness. Evaluations on multiple IR and NLP benchmarks demonstrate state-of-the-art effectiveness compared to approaches that do not exploit pretraining on external data. Extensive ablation studies suggest that relevance and semantic matching signals are complementary across many problem settings, regardless of the choice of underlying encoders.

Memory neurons of long short-term memory (LSTM) networks encode and process information in powerful yet mysterious ways. While there has been work to analyze their behavior in carrying low-level information such as linguistic properties, how they directly contribute to label prediction remains unclear. We find inspiration from biologists and study the affinity between individual neurons and labels, propose a novel metric to quantify the sensitivity of neurons to each label, and conduct experiments to show the validity of our proposed metric. We discover that some neurons are trained to specialize on a subset of labels, and while dropping an arbitrary neuron has little effect on the overall accuracy of the model, dropping label-specialized neurons predictably and significantly degrades prediction accuracy on the associated label. We further examine the consistency of neuron-label affinity across different models. These observations provide insight into the inner mechanisms of LSTMs.

We present Birch, a system that applies BERT to document retrieval via integration with the open-source Anserini information retrieval toolkit to demonstrate end-to-end search over large document collections. Birch implements simple ranking models that achieve state-of-the-art effectiveness on standard TREC newswire and social media test collections. This demonstration focuses on technical challenges in the integration of NLP and IR capabilities, along with the design rationale behind our approach to tightly-coupled integration between Python (to support neural networks) and the Java Virtual Machine (to support document retrieval using the open-source Lucene search library). We demonstrate integration of Birch with an existing search interface as well as interactive notebooks that highlight its capabilities in an easy-to-understand manner.

Used for simple commands recognition on devices from smart speakers to mobile phones, keyword spotting systems are everywhere. Ubiquitous as well are web applications, which have grown in popularity and complexity over the last decade. However, despite their obvious advantages in natural language interaction, voice-enabled web applications are still few and far between. We attempt to bridge this gap with Honkling, a novel, JavaScript-based keyword spotting system. Purely client-side and cross-device compatible, Honkling can be deployed directly on user devices. Our in-browser implementation enables seamless personalization, which can greatly improve model quality; in the presence of underrepresented, non-American user accents, we can achieve up to an absolute 10% increase in accuracy in the personalized model with only a few examples.

Knowledge distillation can effectively transfer knowledge from BERT, a deep language representation model, to traditional, shallow word embedding-based neural networks, helping them approach or exceed the quality of other heavyweight language representation models. As shown in previous work, critical to this distillation procedure is the construction of an unlabeled transfer dataset, which enables effective knowledge transfer. To create transfer set examples, we propose to sample from pretrained language models fine-tuned on task-specific text. Unlike previous techniques, this directly captures the purpose of the transfer set. We hypothesize that this principled, general approach outperforms rule-based techniques. On four datasets in sentiment classification, sentence similarity, and linguistic acceptability, we show that our approach improves upon previous methods. We outperform OpenAI GPT, a deep pretrained transformer, on three of the datasets, while using a single-layer bidirectional LSTM that runs at least ten times faster.

We present a scalable, open-source platform that “distills” a potentially large text collection into a knowledge graph. Our platform takes documents stored in Apache Solr and scales out the Stanford CoreNLP toolkit via Apache Spark integration to extract mentions and relations that are then ingested into the Neo4j graph database. The raw knowledge graph is then enriched with facts extracted from an external knowledge graph. The complete product can be manipulated by various applications using Neo4j’s native Cypher query language: We present a subgraph-matching approach to align extracted relations with external facts and show that fact verification, locating textual support for asserted facts, detecting inconsistent and missing facts, and extracting distantly-supervised training data can all be performed within the same framework.

We examine the problem of question answering over knowledge graphs, focusing on simple questions that can be answered by the lookup of a single fact. Adopting a straightforward decomposition of the problem into entity detection, entity linking, relation prediction, and evidence combination, we explore simple yet strong baselines. On the popular SimpleQuestions dataset, we find that basic LSTMs and GRUs plus a few heuristics yield accuracies that approach the state of the art, and techniques that do not use neural networks also perform reasonably well. These results show that gains from sophisticated deep learning techniques proposed in the literature are quite modest and that some previous models exhibit unnecessary complexity.

We demonstrate the serverless deployment of neural networks for model inferencing in NLP applications using Amazon’s Lambda service for feedforward evaluation and DynamoDB for storing word embeddings. Our architecture realizes a pay-per-request pricing model, requiring zero ongoing costs for maintaining server instances. All virtual machine management is handled behind the scenes by the cloud provider without any direct developer intervention. We describe a number of techniques that allow efficient use of serverless resources, and evaluations confirm that our design is both scalable and inexpensive.

We demonstrate a JavaScript implementation of a convolutional neural network that performs feedforward inference completely in the browser. Such a deployment means that models can run completely on the client, on a wide range of devices, without making backend server requests. This design is useful for applications with stringent latency requirements or low connectivity. Our evaluations show the feasibility of JavaScript as a deployment target. Furthermore, an in-browser implementation enables seamless integration with the JavaScript ecosystem for information visualization, providing opportunities to visually inspect neural networks and better understand their inner workings.

Question answering over knowledge graphs is an important problem of interest both commercially and academically. There is substantial interest in the class of natural language questions that can be answered via the lookup of a single fact, driven by the availability of the popular SimpleQuestions dataset. The problem with this dataset, however, is that answer triples are provided from Freebase, which has been defunct for several years. As a result, it is difficult to build “real-world” question answering systems that are operationally deployable. Furthermore, a defunct knowledge graph means that much of the infrastructure for querying, browsing, and manipulating triples no longer exists. To address this problem, we present SimpleDBpediaQA, a new benchmark dataset for simple question answering over knowledge graphs that was created by mapping SimpleQuestions entities and predicates from Freebase to DBpedia. Although this mapping is conceptually straightforward, there are a number of nuances that make the task non-trivial, owing to the different conceptual organizations of the two knowledge graphs. To lay the foundation for future research using this dataset, we leverage recent work to provide simple yet strong baselines with and without neural networks.

Mining biomedical text offers an opportunity to automatically discover important facts and infer associations among them. As new scientific findings appear across a large collection of biomedical publications, our aim is to tap into this literature to automate biomedical knowledge extraction and identify important insights from them. Towards that goal, we develop a system with novel deep neural networks to extract insights on biomedical literature. Evaluation shows our system is able to provide insights with competitive accuracy of human acceptance and its relation extraction component outperforms previous work.

Grammars for machine translation can be materialized on demand by finding source phrases in an indexed parallel corpus and extracting their translations. This approach is limited in practical applications by the computational expense of online lookup and extraction. For phrase-based models, recent work has shown that on-demand grammar extraction can be greatly accelerated by parallelization on general purpose graphics processing units (GPUs), but these algorithms do not work for hierarchical models, which require matching patterns that contain gaps. We address this limitation by presenting a novel GPU algorithm for on-demand hierarchical grammar extraction that is at least an order of magnitude faster than a comparable CPU algorithm when processing large batches of sentences. In terms of end-to-end translation, with decoding on the CPU, we increase throughput by roughly two thirds on a standard MT evaluation dataset. The GPU necessary to achieve these improvements increases the cost of a server by about a third. We believe that GPU-based extraction of hierarchical grammars is an attractive proposition, particularly for MT applications that demand high throughput.