Instruction fine-tuning has recently emerged as a promising approach for improving the zero-shot capabilities of Large Language Models (LLMs) on new tasks. This technique has shown particular strength in improving the performance of modestly sized LLMs, sometimes inducing performance competitive with much larger model variants. In this paper we ask two questions: (1) How sensitive are instruction-tuned models to the particular phrasings of instructions, and, (2) How can we make them more robust to such natural language variation? To answer the former, we collect a set of 319 instructions manually written by NLP practitioners for over 80 unique tasks included in widely used benchmarks, and we evaluate the variance and average performance of these instructions as compared to instruction phrasings observed during instruction fine-tuning. We find that using novel (unobserved) but appropriate instruction phrasings consistently degrades model performance, sometimes substantially so. Further, such natural instructions yield a wide variance in downstream performance, despite their semantic equivalence. Put another way, instruction-tuned models are not especially robust to instruction re-phrasings. We propose a simple method to mitigate this issue by introducing ``soft prompt'' embedding parameters and optimizing these to maximize the similarity between representations of semantically equivalent instructions. We show that this method consistently improves the robustness of instruction-tuned models.
An abundance of datasets exist for training and evaluating models on the task of summary generation.However, these datasets are often derived heuristically, and lack sufficient annotations to support research into all aspects of summarization, such as evidence extraction and controllable summarization. We introduce a benchmark comprising 8 tasks that require multi-dimensional understanding of summarization, e.g., surfacing evidence for a summary, assessing its correctness, and gauging its relevance to different topics. We compare various methods on this benchmark and discover that on multiple tasks, moderately-sized fine-tuned models consistently outperform much larger few-shot prompted language models. For factuality related tasks, we also evaluate existing heuristics to create training data and find that training on them performs worse than training on $20\times$ less human-labeled data. Our benchmark consists of data from 6 different domains, allowing us to study cross-domain performance of trained models. We find that for some tasks, the amount of training data matters more than the domain where it comes from, while for other tasks training specifically on data from the target domain, even if limited, is more beneficial. Our work fulfills the need for a well-annotated summarization benchmark with diverse tasks, and provides useful insights about the impact of the quality, size and domain of training data.
Evaluating multi-document summarization (MDS) quality is difficult. This is especially true in the case of MDS for biomedical literature reviews, where models must synthesize contradicting evidence reported across different documents. Prior work has shown that rather than performing the task, models may exploit shortcuts that are difficult to detect using standard n-gram similarity metrics such as ROUGE. Better automated evaluation metrics are needed, but few resources exist to assess metrics when they are proposed. Therefore, we introduce a dataset of human-assessed summary quality facets and pairwise preferences to encourage and support the development of better automated evaluation methods for literature review MDS. We take advantage of community submissions to the Multi-document Summarization for Literature Review (MSLR) shared task to compile a diverse and representative sample of generated summaries. We analyze how automated summarization evaluation metrics correlate with lexical features of generated summaries, to other automated metrics including several we propose in this work, and to aspects of human-assessed summary quality. We find that not only do automated metrics fail to capture aspects of quality as assessed by humans, in many cases the system rankings produced by these metrics are anti-correlated with rankings according to human annotators.
Automated text simplification aims to produce simple versions of complex texts. This task is especially useful in the medical domain, where the latest medical findings are typically communicated via complex and technical articles. This creates barriers for laypeople seeking access to up-to-date medical findings, consequently impeding progress on health literacy. Most existing work on medical text simplification has focused on monolingual settings, with the result that such evidence would be available only in just one language (most often, English). This work addresses this limitation via multilingual simplification, i.e., directly simplifying complex texts into simplified texts in multiple languages. We introduce MultiCochrane, the first sentence-aligned multilingual text simplification dataset for the medical domain in four languages: English, Spanish, French, and Farsi. We evaluate fine-tuned and zero-shot models across these languages, with extensive human assessments and analyses. Although models can now generate viable simplified texts, we identify outstanding challenges that this dataset might be used to address.
Medical systematic reviews are crucial for informing clinical decision making and healthcare policy. But producing such reviews is onerous and time-consuming. Thus, high-quality evidence synopses are not available for many questions and may be outdated even when they are available. Large language models (LLMs) are now capable of generating long-form texts, suggesting the tantalizing possibility of automatically generating literature reviews on demand. However, LLMs sometimes generate inaccurate (and potentially misleading) texts by hallucinating or omitting important information. In the healthcare context, this may render LLMs unusable at best and dangerous at worst. Most discussion surrounding the benefits and risks of LLMs have been divorced from specific applications. In this work, we seek to qualitatively characterize the potential utility and risks of LLMs for assisting in production of medical evidence reviews. We conducted 16 semi-structured interviews with international experts in systematic reviews, grounding discussion in the context of generating evidence reviews. Domain experts indicated that LLMs could aid writing reviews, as a tool for drafting or creating plain language summaries, generating templates or suggestions, distilling information, crosschecking, and synthesizing or interpreting text inputs. But they also identified issues with model outputs and expressed concerns about potential downstream harms of confidently composed but inaccurate LLM outputs which might mislead. Other anticipated potential downstream harms included lessened accountability and proliferation of automatically generated reviews that might be of low quality. Informed by this qualitative analysis, we identify criteria for rigorous evaluation of biomedical LLMs aligned with domain expert views.
Large language models, particularly GPT-3, are able to produce high quality summaries of general domain news articles in few- and zero-shot settings. However, it is unclear if such models are similarly capable in more specialized, high-stakes domains such as biomedicine. In this paper, we enlist domain experts (individuals with medical training) to evaluate summaries of biomedical articles generated by GPT-3, given zero supervision. We consider both single- and multi-document settings. In the former, GPT-3 is tasked with generating regular and plain-language summaries of articles describing randomized controlled trials; in the latter, we assess the degree to which GPT-3 is able to \emph{synthesize} evidence reported across a collection of articles. We design an annotation scheme for evaluating model outputs, with an emphasis on assessing the factual accuracy of generated summaries. We find that while GPT-3 is able to summarize and simplify single biomedical articles faithfully, it struggles to provide accurate aggregations of findings over multiple documents. We release all data and annotations used in this work.
Relation extraction (RE) is the core NLP task of inferring semantic relationships between entities from text. Standard supervised RE techniques entail training modules to tag tokens comprising entity spans and then predict the relationship between them. Recent work has instead treated the problem as a \emph{sequence-to-sequence} task, linearizing relations between entities as target strings to be generated conditioned on the input. Here we push the limits of this approach, using larger language models (GPT-3 and Flan-T5 large) than considered in prior work and evaluating their performance on standard RE tasks under varying levels of supervision. We address issues inherent to evaluating generative approaches to RE by doing human evaluations, in lieu of relying on exact matching. Under this refined evaluation, we find that: (1) Few-shot prompting with GPT-3 achieves near SOTA performance, i.e., roughly equivalent to existing fully supervised models; (2) Flan-T5 is not as capable in the few-shot setting, but supervising and fine-tuning it with Chain-of-Thought (CoT) style explanations (generated via GPT-3) yields SOTA results. We release this model as a new baseline for RE tasks.
Results from Randomized Controlled Trials (RCTs) establish the comparative effectiveness of interventions, and are in turn critical inputs for evidence-based care. However, results from RCTs are presented in (often unstructured) natural language articles describing the design, execution, and outcomes of trials; clinicians must manually extract findings pertaining to interventions and outcomes of interest from such articles. This onerous manual process has motivated work on (semi-)automating extraction of structured evidence from trial reports. In this work we propose and evaluate a text-to-text model built on instruction-tuned Large Language Models (LLMs) to jointly extract Interventions, Outcomes, and Comparators (ICO elements) from clinical abstracts, and infer the associated results reported. Manual (expert) and automated evaluations indicate that framing evidence extraction as a conditional generation task and fine-tuning LLMs for this purpose realizes considerable ($\sim$20 point absolute F1 score) gains over the previous SOTA. We perform ablations and error analyses to assess aspects that contribute to model performance, and to highlight potential directions for further improvements. We apply our model to a collection of published RCTs through mid-2022, and release a searchable database of structured findings (anonymously for now): bit.ly/joint-relations-extraction-mlhc
We present TrialsSummarizer, a system that aims to automatically summarize evidence presented in the set of randomized controlled trials most relevant to a given query. Building on prior work, the system retrieves trial publications matching a query specifying a combination of condition, intervention(s), and outcome(s), and ranks these according to sample size and estimated study quality. The top-k such studies are passed through a neural multi-document summarization system, yielding a synopsis of these trials. We consider two architectures: A standard sequence-to-sequence model based on BART and a multi-headed architecture intended to provide greater transparency to end-users. Both models produce fluent and relevant summaries of evidence retrieved for queries, but their tendency to introduce unsupported statements render them inappropriate for use in this domain at present. The proposed architecture may help users verify outputs allowing users to trace generated tokens back to inputs.