VariErr NLI: Separating Annotation Error from Human Label Variation

Human label variation arises when annotators assign different labels to the same item for valid reasons, while annotation errors occur when labels are assigned for invalid reasons. These two issues are prevalent in NLP benchmarks, yet existing research has studied them in isolation. To the best of our knowledge, there exists no prior work that focuses on teasing apart error from signal, especially in cases where signal is beyond black-and-white. To fill this gap, we introduce a systematic methodology and a new dataset, VariErr (variation versus error), focusing on the NLI task in English. We propose a 2-round annotation scheme with annotators explaining each label and subsequently judging the validity of label-explanation pairs. \name{} contains 7,574 validity judgments on 1,933 explanations for 500 re-annotated NLI items. We assess the effectiveness of various automatic error detection (AED) methods and GPTs in uncovering errors versus human label variation. We find that state-of-the-art AED methods significantly underperform compared to GPTs and humans. While GPT-4 is the best system, it still falls short of human performance. Our methodology is applicable beyond NLI, offering fertile ground for future research on error versus plausible variation, which in turn can yield better and more trustworthy NLP systems.

"My Answer is C": First-Token Probabilities Do Not Match Text Answers in Instruction-Tuned Language Models

The open-ended nature of language generation makes the evaluation of autoregressive large language models (LLMs) challenging. One common evaluation approach uses multiple-choice questions (MCQ) to limit the response space. The model is then evaluated by ranking the candidate answers by the log probability of the first token prediction. However, first-tokens may not consistently reflect the final response output, due to model's diverse response styles such as starting with "Sure" or refusing to answer. Consequently, MCQ evaluation is not indicative of model behaviour when interacting with users. But by how much? We evaluate how aligned first-token evaluation is with the text output along several dimensions, namely final option choice, refusal rate, choice distribution and robustness under prompt perturbation. Our results show that the two approaches are severely misaligned on all dimensions, reaching mismatch rates over 60%. Models heavily fine-tuned on conversational or safety data are especially impacted. Crucially, models remain misaligned even when we increasingly constrain prompts, i.e., force them to start with an option letter or example template. Our findings i) underscore the importance of inspecting the text output, too and ii) caution against relying solely on first-token evaluation.

HunFlair2 in a cross-corpus evaluation of biomedical named entity recognition and normalization tools

With the exponential growth of the life science literature, biomedical text mining (BTM) has become an essential technology for accelerating the extraction of insights from publications. Identifying named entities (e.g., diseases, drugs, or genes) in texts and their linkage to reference knowledge bases are crucial steps in BTM pipelines to enable information aggregation from different documents. However, tools for these two steps are rarely applied in the same context in which they were developed. Instead, they are applied in the wild, i.e., on application-dependent text collections different from those used for the tools' training, varying, e.g., in focus, genre, style, and text type. This raises the question of whether the reported performance of BTM tools can be trusted for downstream applications. Here, we report on the results of a carefully designed cross-corpus benchmark for named entity extraction, where tools were applied systematically to corpora not used during their training. Based on a survey of 28 published systems, we selected five for an in-depth analysis on three publicly available corpora encompassing four different entity types. Comparison between tools results in a mixed picture and shows that, in a cross-corpus setting, the performance is significantly lower than the one reported in an in-corpus setting. HunFlair2 showed the best performance on average, being closely followed by PubTator. Our results indicate that users of BTM tools should expect diminishing performances when applying them in the wild compared to original publications and show that further research is necessary to make BTM tools more robust.

Donkii: Can Annotation Error Detection Methods Find Errors in Instruction-Tuning Datasets?

Instruction-tuning has become an integral part of training pipelines for Large Language Models (LLMs) and has been shown to yield strong performance gains. In an orthogonal line of research, Annotation Error Detection (AED) has emerged as a tool for detecting quality issues of gold-standard labels. But so far, the application of AED methods is limited to discriminative settings. It is an open question how well AED methods generalize to generative settings which are becoming widespread via generative LLMs. In this work, we present a first and new benchmark for AED on instruction-tuning data: Donkii. It encompasses three instruction-tuning datasets enriched with annotations by experts and semi-automatic methods. We find that all three datasets contain clear-cut errors that sometimes directly propagate into instruction-tuned LLMs. We propose four AED baselines for the generative setting and evaluate them comprehensively on the newly introduced dataset. Our results demonstrate that choosing the right AED method and model size is indeed crucial, thereby deriving practical recommendations. To gain insights, we provide a first case-study to examine how the quality of the instruction-tuning datasets influences downstream performance.

BELB: a Biomedical Entity Linking Benchmark

Biomedical entity linking (BEL) is the task of grounding entity mentions to a knowledge base. It plays a vital role in information extraction pipelines for the life sciences literature. We review recent work in the field and find that, as the task is absent from existing benchmarks for biomedical text mining, different studies adopt different experimental setups making comparisons based on published numbers problematic. Furthermore, neural systems are tested primarily on instances linked to the broad coverage knowledge base UMLS, leaving their performance to more specialized ones, e.g. genes or variants, understudied. We therefore developed BELB, a Biomedical Entity Linking Benchmark, providing access in a unified format to 11 corpora linked to 7 knowledge bases and spanning six entity types: gene, disease, chemical, species, cell line and variant. BELB greatly reduces preprocessing overhead in testing BEL systems on multiple corpora offering a standardized testbed for reproducible experiments. Using BELB we perform an extensive evaluation of six rule-based entity-specific systems and three recent neural approaches leveraging pre-trained language models. Our results reveal a mixed picture showing that neural approaches fail to perform consistently across entity types, highlighting the need of further studies towards entity-agnostic models.