Teaching large language models (LLMs) to generate text with attribution to evidence sources can reduce hallucinations, improve verifiability in question answering systems (QA), and increase reliability of retrieval augmented LLMs. Despite gaining increasing popularity for usage in QA systems and search engines, current LLMs struggle with attribution for long-form responses which require reasoning over multiple evidence sources. To address this, in this paper we aim to improve the attribution capability of LLMs for long-form answer generation to multiple sources, with multiple citations per sentence. However, data for training multi-source attributable QA systems is difficult and expensive to annotate, and therefore scarce. To overcome this challenge, we transform existing QA datasets for this task (MultiAttr), and empirically demonstrate, on a wide range of attribution benchmark datasets, that fine-tuning on MultiAttr provides significant improvements over training only on the target QA domain. Lastly, to fill a gap in existing benchmarks, we present a multi-source attribution dataset containing multi-paragraph answers, PolitiICite, based on PolitiFact articles that discuss events closely related to implementation statuses of election promises.
Users of AI-based virtual assistants and search systems encounter challenges in articulating their intents while seeking information on unfamiliar topics, possibly due to complexity of the user’s intent or the lack of meta-information on the topic. We posit that an iterative suggested question-answering (SQA) conversation can improve the trade-off between the satisfaction of the user’s intent while keeping the information exchange natural and cognitive load of the interaction minimal on the users. In this paper, we evaluate a novel setting ProMISe by means of a sequence of interactions between a user, having a predefined information-seeking intent, and an agent that generates a set of SQA pairs at each step to aid the user to get closer to their intent. We simulate this two-player setting to create a multi-turn conversational dataset of SQAs and user choices (1025 dialogues comprising 4453 turns and 17812 SQAs) using human-feedback, chain-of-thought prompting and web-retrieval augmented large language models. We evaluate the quality of the SQs in the dataset on attributes such as diversity, specificity, grounding, etc, and benchmark the performance of different language models for the task of replicating user behavior.
In this paper, we investigate which questions are challenging for retrieval-based Question Answering (QA). We (i) propose retrieval complexity (RC), a novel metric conditioned on the completeness of retrieved documents, which measures the difficulty of answering questions, and (ii) propose an unsupervised pipeline to measure RC given an arbitrary retrieval system.Our proposed pipeline measures RC more accurately than alternative estimators, including LLMs, on six challenging QA benchmarks. Further investigation reveals that RC scores strongly correlate with both QA performance and expert judgment across five of the six studied benchmarks, indicating that RC is an effective measure of question difficulty.Subsequent categorization of high-RC questions shows that they span a broad set of question shapes, including multi-hop, compositional, and temporal QA, indicating that RC scores can categorize a new subset of complex questions. Our system can also have a major impact on retrieval-based systems by helping to identify more challenging questions on existing datasets.
Recent studies show that sentence-level extractive QA, i.e., based on Answer Sentence Selection (AS2), is outperformed by Generation-based QA (GenQA) models, which generate answers using the top-k answer sentences ranked by AS2 models (a la retrieval-augmented generation style). In this paper, we propose a novel training paradigm for GenQA using supervision from automatic QA evaluation models (GAVA). Specifically, we propose three strategies to transfer knowledge from these QA evaluation models to a GenQA model: (i) augmenting training data with answers generated by the GenQA model and labelled by GAVA (either statically, before training, or (ii) dynamically, at every training epoch); and (iii) using the GAVA score for weighting the generator loss during the learning of the GenQA model. We evaluate our proposed methods on two academic and one industrial dataset, obtaining a significant improvement in answering accuracy over the previous state of the art.
Answer Sentence Selection (AS2) is a core component for building an accurate Question Answering pipeline. AS2 models rank a set of candidate sentences based on how likely they answer a given question. The state of the art in AS2 exploits pre-trained transformers by transferring them on large annotated datasets, while using local contextual information around the candidate sentence. In this paper, we propose three pre-training objectives designed to mimic the downstream fine-tuning task of contextual AS2. This allows for specializing LMs when fine-tuning for contextual AS2. Our experiments on three public and two large-scale industrial datasets show that our pre-training approaches (applied to RoBERTa and ELECTRA) can improve baseline contextual AS2 accuracy by up to 8% on some datasets.
Recent studies show that Question Answering (QA) based on Answer Sentence Selection (AS2) can be improved by generating an improved answer from the top-k ranked answer sentences (termed GenQA). This allows for synthesizing the information from multiple candidates into a concise, natural-sounding answer. However, creating large-scale supervised training data for GenQA models is very challenging. In this paper, we propose to train a GenQA model by transferring knowledge from a trained AS2 model, to overcome the aforementioned issue. First, we use an AS2 model to produce a ranking over answer candidates for a set of questions. Then, we use the top ranked candidate as the generation target, and the next k top ranked candidates as context for training a GenQA model. We also propose to use the AS2 model prediction scores for loss weighting and score-conditioned input/output shaping, to aid the knowledge transfer. Our evaluation on three public and one large industrial datasets demonstrates the superiority of our approach over the AS2 baseline, and GenQA trained using supervised data.
An important task for designing QA systems is answer sentence selection (AS2): selecting the sentence containing (or constituting) the answer to a question from a set of retrieved relevant documents. In this paper, we propose three novel sentence-level transformer pre-training objectives that incorporate paragraph-level semantics within and across documents, to improve the performance of transformers for AS2, and mitigate the requirement of large labeled datasets. Specifically, the model is tasked to predict whether: (i) two sentences are extracted from the same paragraph, (ii) a given sentence is extracted from a given paragraph, and (iii) two paragraphs are extracted from the same document. Our experiments on three public and one industrial AS2 datasets demonstrate the empirical superiority of our pre-trained transformers over baseline models such as RoBERTa and ELECTRA for AS2.
Inference tasks such as answer sentence selection (AS2) or fact verification are typically solved by fine-tuning transformer-based models as individual sentence-pair classifiers. Recent studies show that these tasks benefit from modeling dependencies across multiple candidate sentences jointly. In this paper, we first show that popular pre-trained transformers perform poorly when used for fine-tuning on multi-candidate inference tasks. We then propose a new pre-training objective that models the paragraph-level semantics across multiple input sentences. Our evaluation on three AS2 and one fact verification datasets demonstrates the superiority of our pre-training technique over the traditional ones for transformers used as joint models for multi-candidate inference tasks, as well as when used as cross-encoders for sentence-pair formulations of these tasks.
In this paper we propose a novel approach towards improving the efficiency of Question Answering (QA) systems by filtering out questions that will not be answered by them. This is based on an interesting new finding: the answer confidence scores of state-of-the-art QA systems can be approximated well by models solely using the input question text. This enables preemptive filtering of questions that are not answered by the system due to their answer confidence scores being lower than the system threshold. Specifically, we learn Transformer-based question models by distilling Transformer-based answering models. Our experiments on three popular QA datasets and one industrial QA benchmark demonstrate the ability of our question models to approximate the Precision/Recall curves of the target QA system well. These question models, when used as filters, can effectively trade off lower computation cost of QA systems for lower Recall, e.g., reducing computation by ~60%, while only losing ~3-4% of Recall.
Fine-tuning (FT) pre-trained sentence embedding models on small datasets has been shown to have limitations. In this paper we show that concatenating the embeddings from the pre-trained model with those from a simple sentence embedding model trained only on the target data, can improve over the performance of FT for few-sample tasks. To this end, a linear classifier is trained on the combined embeddings, either by freezing the embedding model weights or training the classifier and embedding models end-to-end. We perform evaluation on seven small datasets from NLP tasks and show that our approach with end-to-end training outperforms FT with negligible computational overhead. Further, we also show that sophisticated combination techniques like CCA and KCCA do not work as well in practice as concatenation. We provide theoretical analysis to explain this empirical observation.
Modern text classification models are susceptible to adversarial examples, perturbed versions of the original text indiscernible by humans which get misclassified by the model. Recent works in NLP use rule-based synonym replacement strategies to generate adversarial examples. These strategies can lead to out-of-context and unnaturally complex token replacements, which are easily identifiable by humans. We present BAE, a black box attack for generating adversarial examples using contextual perturbations from a BERT masked language model. BAE replaces and inserts tokens in the original text by masking a portion of the text and leveraging the BERT-MLM to generate alternatives for the masked tokens. Through automatic and human evaluations, we show that BAE performs a stronger attack, in addition to generating adversarial examples with improved grammaticality and semantic coherence as compared to prior work.
In this paper we show that a simple beam approximation of the joint distribution between attention and output is an easy, accurate, and efficient attention mechanism for sequence to sequence learning. The method combines the advantage of sharp focus in hard attention and the implementation ease of soft attention. On five translation tasks we show effortless and consistent gains in BLEU compared to existing attention mechanisms.