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 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, first-token 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 as well and ii) caution against relying solely on first-token evaluation.

The data-centric revolution in AI has revealed the importance of high-quality training data for developing successful AI models. However, annotations are sensitive to annotator characteristics, training materials, and to the design and wording of the data collection instrument. This paper explores the impact of observation order on annotations. We find that annotators’ judgments change based on the order in which they see observations. We use ideas from social psychology to motivate hypotheses about why this order effect occurs. We believe that insights from social science can help AI researchers improve data and model quality.

Although the NLP community has adopted central differential privacy as a go-to framework for privacy-preserving model training or data sharing, the choice and interpretation of the key parameter, privacy budget 𝜀 that governs the strength of privacy protection, remains largely arbitrary. We argue that determining the 𝜀 value should not be solely in the hands of researchers or system developers, but must also take into account the actual people who share their potentially sensitive data. In other words: Would you share your instant messages for 𝜀 of 10? We address this research gap by designing, implementing, and conducting a behavioral experiment (311 lay participants) to study the behavior of people in uncertain decision-making situations with respect to privacy-threatening situations. Framing the risk perception in terms of two realistic NLP scenarios and using a vignette behavioral study help us determine what 𝜀 thresholds would lead lay people to be willing to share sensitive textual data – to our knowledge, the first study of its kind.

We present a research agenda focused on efficiently extracting, assuring quality, and consolidating textual company sustainability information to address urgent climate change decision-making needs. Starting from the goal to create integrated FAIR (Findable, Accessible, Interoperable, Reusable) climate-related data, we identify research needs pertaining to the technical aspects of information extraction as well as to the design of the integrated sustainability datasets that we seek to compile. Regarding extraction, we leverage technological advancements, particularly in large language models (LLMs) and Retrieval-Augmented Generation (RAG) pipelines, to unlock the underutilized potential of unstructured textual information contained in corporate sustainability reports. In applying these techniques, we review key challenges, which include the retrieval and extraction of CO₂ emission values from PDF documents, especially from unstructured tables and graphs therein, and the validation of automatically extracted data through comparisons with human-annotated values. We also review how existing use cases and practices in climate risk analytics relate to choices of what textual information should be extracted and how it could be linked to existing structured data.

Prompt-based methods have been successfully applied to multilingual pretrained language models for zero-shot cross-lingual understanding. However, most previous studies primarily focused on sentence-level classification tasks, and only a few considered token-level labeling tasks such as Named Entity Recognition (NER) and Part-of-Speech (POS) tagging. In this paper, we propose Token-Level Prompt Decomposition (ToPro), which facilitates the prompt-based method for token-level sequence labeling tasks. The ToPro method decomposes an input sentence into single tokens and applies one prompt template to each token. Our experiments on multilingual NER and POS tagging datasets demonstrate that ToPro-based fine-tuning outperforms Vanilla fine-tuning and Prompt-Tuning in zero-shot cross-lingual transfer, especially for languages that are typologically different from the source language English. Our method also attains state-of-the-art performance when employed with the mT5 model. Besides, our exploratory study in multilingual large language models shows that ToPro performs much better than the current in-context learning method. Overall, the performance improvements show that ToPro could potentially serve as a novel and simple benchmarking method for sequence labeling tasks.

When training data are collected from human annotators, the design of the annotation instrument, the instructions given to annotators, the characteristics of the annotators, and their interactions can impact training data. This study demonstrates that design choices made when creating an annotation instrument also impact the models trained on the resulting annotations. We introduce the term annotation sensitivity to refer to the impact of annotation data collection methods on the annotations themselves and on downstream model performance and predictions. We collect annotations of hate speech and offensive language in five experimental conditions of an annotation instrument, randomly assigning annotators to conditions. We then fine-tune BERT models on each of the five resulting datasets and evaluate model performance on a holdout portion of each condition. We find considerable differences between the conditions for 1) the share of hate speech/offensive language annotations, 2) model performance, 3) model predictions, and 4) model learning curves. Our results emphasize the crucial role played by the annotation instrument which has received little attention in the machine learning literature. We call for additional research into how and why the instrument impacts the annotations to inform the development of best practices in instrument design.