Recently proposed long-form question answering (QA) systems, supported by large language models (LLMs), have shown promising capabilities. Yet, attributing and verifying their generated abstractive answers can be difficult, and automatically evaluating their accuracy remains an ongoing challenge.In this work, we introduce a new QA task for answering multi-answer questions by summarizing multiple diverse sources in a semi-extractive fashion. Specifically, Semi-extractive Multi-source QA (SEMQA) requires models to output a comprehensive answer, while mixing factual quoted spans—copied verbatim from given input sources—and non-factual free-text connectors that glue these spans together into a single cohesive passage. This setting bridges the gap between the outputs of well-grounded but constrained extractive QA systems and more fluent but harder to attribute fully abstractive answers. Particularly, it enables a new mode for language models that leverages their advanced language generation capabilities, while also producing fine in-line attributions by-design that are easy to verify, interpret, and evaluate. To study this task, we create the first dataset of this kind, QuoteSum, with human-written semi-extractive answers to natural and generated questions, and define text-based evaluation metrics. Experimenting with several LLMs in various settings, we find this task to be surprisingly challenging, demonstrating the importance of QuoteSum for developing and studying such consolidation capabilities.
Ranking documents using Large Language Models (LLMs) by directly feeding the query and candidate documents into the prompt is an interesting and practical problem. However, researchers have found it difficult to outperform fine-tuned baseline rankers on benchmark datasets.We analyze pointwise and listwise ranking prompts used by existing methods and argue that off-the-shelf LLMs do not fully understand these challenging ranking formulations. In this paper, we propose to significantly reduce the burden on LLMs by using a new technique called Pairwise Ranking Prompting (PRP).Our results are the first in the literature to achieve state-of-the-art ranking performance on standard benchmarks using moderate-sized open-sourced LLMs. On TREC-DL 2019&2020, PRP based on the Flan-UL2 model with 20B parameters performs favorably with the previous best approach in the literature, which is based on the blackbox commercial GPT-4 that has 50x (estimated) model size, while outperforming other LLM-based solutions, such as InstructGPT which has 175B parameters, by over 10% for all ranking metrics. By using the same prompt template on seven BEIR tasks, PRP outperforms supervised baselines and outperforms the blackbox commercial ChatGPT solution by 4.2% and pointwise LLM-based solutions by more than 10% on average NDCG@10.Furthermore, we propose several variants of PRP to improve efficiency and show that it is possible to achieve competitive results even with linear complexity.
We develop and evaluate multilingual scientific documents similarity measurement models in this work. Such models can be used to find related papers in different languages, which can help multilingual researchers find and explore papers more efficiently. We propose the first multilingual scientific documents dataset, Open-access Multilingual Scientific Documents (OpenMSD), which has 74M papers in 103 languages and 778M citation pairs. With OpenMSD, we develop multilingual SDSM models by adjusting and extending the state-of-the-art methods designed for English SDSM tasks. We find that: (i)Some highly successful methods in English SDSM yield significantly worse performance in multilingual SDSM. (ii)Our best model, which enriches the non-English papers with English summaries, outperforms strong baselines by 7% (in mean average precision) on multilingual SDSM tasks, without compromising the performance on English SDSM tasks.
There have been a lot of interest in the scaling properties of Transformer models. However, not much has been done on the front of investigating the effect of scaling properties of different inductive biases and model architectures. Do model architectures scale differently? If so, how does inductive bias affect scaling behaviour? How does this influence upstream (pretraining) and downstream (transfer)? This paper conducts a systematic study of scaling behaviour of ten diverse model architectures such as Transformers, Switch Transformers, Universal Transformers, Dynamic convolutions, Performers, and recently proposed MLP-Mixers. Via extensive experiments, we show that (1) architecture is an indeed an important consideration when performing scaling and (2) the best performing model can fluctuate at different scales. We believe that the findings outlined in this work has significant implications to how model architectures are currently evaluated in the community.
Recent studies show that large language models (LLMs) can be instructed to effectively perform zero-shot passage re-ranking, in which the results of a first stage retrieval method, such as BM25, are rated and reordered to improve relevance. In this work, we improve LLM-based re-ranking by algorithmically selecting few-shot demonstrations to include in the prompt. Our analysis investigates the conditions where demonstrations are most helpful, and shows that adding even one demonstration is significantly beneficial. We propose a novel demonstration selection strategy based on difficulty rather than the commonly used semantic similarity. Furthermore, we find that demonstrations helpful for ranking are also effective at question generation. We hope our work will spur more principled research into question generation and passage ranking.
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.
Scaling language models improves performance but comes with significant computational costs. This paper proposes UL2R, a method that substantially improves existing language models and their scaling curves with a relatively tiny amount of extra compute. The key idea is to continue training a state-of-the-art large language model on a few more steps with UL2’s mixture-of-denoiser objective. We show that, with almost negligible extra computational costs and no new sources of data, we are able to substantially improve the scaling properties of large language models on downstream metrics. In this paper, we continue training a baseline language model, PaLM, with ULR2, introducing a new set of models at 8B, 62B, and 540B scale which we call U-PaLM. Impressively, at 540B scale, we show an approximately 2x computational savings rate where U-PaLM achieves the same performance as the final PaLM 540B model at around half its computational budget (i.e., saving ~4.4 million TPUv4 hours). We further show that this improved scaling curve leads to “emergent abilities” on challenging BIG-Bench tasks—for instance, U-PaLM does much better on some tasks or demonstrates better quality at much smaller scale (62B as opposed to 540B). Overall, we show that U-PaLM outperforms PaLM on many few-shot setups, including reasoning tasks with chain-of-thought (e.g., GSM8K), multilingual tasks (MGSM, TydiQA), MMLU and challenging BIG-Bench tasks.
Differentiable Search Indices (DSIs) encode a corpus of documents in the parameters of a model and use the same model to map queries directly to relevant document identifiers. Despite the solid performance of DSI models, successfully deploying them in scenarios where document corpora change with time is an open problem. In this work, we introduce DSI++, a continual learning challenge for DSI with the goal of continuously indexing new documents while being able to answer queries related to both previously and newly indexed documents. Across different model scales and document identifier representations, we show that continual indexing of new documents leads to considerable forgetting of previously indexed documents. We also hypothesize and verify that the model experiences forgetting events during training, leading to unstable learning. To mitigate these issues, we investigate two approaches. The first focuses on modifying the training dynamics. Flatter minima implicitly alleviates forgetting, so we explicitly optimize for flatter loss basins and show that the model stably memorizes more documents (+12%). Next, we introduce a parametric memory to generate pseudo-queries for documents and supplement them during incremental indexing to prevent forgetting for the retrieval task. Extensive experiments on a novel continual indexing benchmark based on Natural Questions demonstrate that our proposed solution mitigates the forgetting in DSI++ by a significant margin and improves the average Hits@10 by +21.1% over competitive baselines.
Transformer encoders contextualize token representations by attending to all other tokens at each layer, leading to quadratic increase in compute effort with the input length. In practice, however, the input text of many NLP tasks can be seen as a sequence of related segments (e.g., the sequence of sentences within a passage, or the hypothesis and premise in NLI). While attending across these segments is highly beneficial for many tasks, we hypothesize that this interaction can be delayed until later encoding stages. To this end, we introduce Layer-Adjustable Interactions in Transformers (LAIT). Within LAIT, segmented inputs are first encoded independently, and then jointly. This partial two-tower architecture bridges the gap between a Dual Encoder’s ability to pre-compute representations for segments and a fully self-attentive Transformer’s capacity to model cross-segment attention. The LAIT framework effectively leverages existing pretrained Transformers and converts them into the hybrid of the two aforementioned architectures, allowing for easy and intuitive control over the performance-efficiency tradeoff. Experimenting on a wide range of NLP tasks, we find LAIT able to reduce 30-50% of the attention FLOPs on many tasks, while preserving high accuracy; in some practical settings, LAIT could reduce actual latency by orders of magnitude.
With the devastating outbreak of COVID-19, vaccines are one of the crucial lines of defense against mass infection in this global pandemic. Given the protection they provide, vaccines are becoming mandatory in certain social and professional settings. This paper presents a classification model for detecting COVID-19 vaccination related search queries, a machine learning model that is used to generate search insights for COVID-19 vaccinations. The proposed method combines and leverages advancements from modern state-of-the-art (SOTA) natural language understanding (NLU) techniques such as pretrained Transformers with traditional dense features. We propose a novel approach of considering dense features as memory tokens that the model can attend to. We show that this new modeling approach enables a significant improvement to the Vaccine Search Insights (VSI) task, improving a strong well-established gradient-boosting baseline by relative +15% improvement in F1 score and +14% in precision.
State-of-the-art neural models typically encode document-query pairs using cross-attention for re-ranking. To this end, models generally utilize an encoder-only (like BERT) paradigm or an encoder-decoder (like T5) approach. These paradigms, however, are not without flaws, i.e., running the model on all query-document pairs at inference-time incurs a significant computational cost. This paper proposes a new training and inference paradigm for re-ranking. We propose to finetune a pretrained encoder-decoder model using in the form of document to query generation. Subsequently, we show that this encoder-decoder architecture can be decomposed into a decoder-only language model during inference. This results in significant inference time speedups since the decoder-only architecture only needs to learn to interpret static encoder embeddings during inference. Our experiments show that this new paradigm achieves results that are comparable to the more expensive cross-attention ranking approaches while being up to 6.8X faster. We believe this work paves the way for more efficient neural rankers that leverage large pretrained models.
Natural Language Inference (NLI) has been extensively studied by the NLP community as a framework for estimating the semantic relation between sentence pairs. While early work identified certain biases in NLI models, recent advancements in modeling and datasets demonstrated promising performance.In this work, we further explore the direct zero-shot applicability of NLI models to real applications, beyond the sentence-pair setting they were trained on. First, we analyze the robustness of these models to longer and out-of-domain inputs. Then, we develop new aggregation methods to allow operating over full documents, reaching state-of-the-art performance on the ContractNLI dataset. Interestingly, we find NLI scores to provide strong retrieval signals, leading to more relevant evidence extractions compared to common similarity-based methods. Finally, we go further and investigate whole document clusters to identify both discrepancies and consensus among sources. In a test case, we find real inconsistencies between Wikipedia pages in different languages about the same topic.
In the era of pre-trained language models, Transformers are the de facto choice of model architectures. While recent research has shown promise in entirely convolutional, or CNN, architectures, they have not been explored using the pre-train-fine-tune paradigm. In the context of language models, are convolutional models competitive to Transformers when pre-trained? This paper investigates this research question and presents several interesting findings. Across an extensive set of experiments on 8 datasets/tasks, we find that CNN-based pre-trained models are competitive and outperform their Transformer counterpart in certain scenarios, albeit with caveats. Overall, the findings outlined in this paper suggest that conflating pre-training and architectural advances is misguided and that both advances should be considered independently. We believe our research paves the way for a healthy amount of optimism in alternative architectures.
There are two major classes of natural language grammars — the dependency grammar that models one-to-one correspondences between words and the constituency grammar that models the assembly of one or several corresponded words. While previous unsupervised parsing methods mostly focus on only inducing one class of grammars, we introduce a novel model, StructFormer, that can induce dependency and constituency structure at the same time. To achieve this, we propose a new parsing framework that can jointly generate a constituency tree and dependency graph. Then we integrate the induced dependency relations into the transformer, in a differentiable manner, through a novel dependency-constrained self-attention mechanism. Experimental results show that our model can achieve strong results on unsupervised constituency parsing, unsupervised dependency parsing, and masked language modeling at the same time.
Recent advances in neural text generation modeling have resulted in a number of societal concerns related to how such approaches might be used in malicious ways. It is therefore desirable to develop a deeper understanding of the fundamental properties of such models. The study of artifacts that emerge in machine generated text as a result of modeling choices is a nascent research area. To this end, the extent and degree to which these artifacts surface in generated text is still unclear. In the spirit of better understanding generative text models and their artifacts, we propose the new task of distinguishing which of several variants of a given model generated some piece of text. Specifically, we conduct an extensive suite of diagnostic tests to observe whether modeling choices (e.g., sampling methods, top-k probabilities, model architectures, etc.) leave detectable artifacts in the text they generate. Our key finding, which is backed by a rigorous set of experiments, is that such artifacts are present and that different modeling choices can be inferred by looking at generated text alone. This suggests that neural text generators may actually be more sensitive to various modeling choices than previously thought.