Daily Papers

LLM-based evaluators have rapidly become a standard tool for machine translation evaluation, yet the design choices that shape their behavior remain poorly understood. We present a systematic study of how instruction design, few-shot demonstrations, backbone model choice, and multi-step workflows affect span-level and ranking-based MT evaluation. We collect new expert RATE annotations on WMT24 and WMT25 English-Russian translations, with transfer validation on MQM- and ESA-annotated benchmarks across five additional language pairs. Our experiments reveal four robust findings: instruction structure shapes evaluator behavior, high-quality demonstrations matter more than quantity, frontier backbones exhibit distinct evaluator profiles (recall-, precision-, or ranking-oriented), and fusing outputs across diverse backbones outperforms single-evaluator multi-call setups. Although evaluator design exposes a large combinatorial space, several discovered design principles transfer consistently across annotation protocols and language pairs. We synthesize these findings into ORBIT, a family of LLM-based MT evaluators. Our strongest configurations achieve expert-level performance in error span detection and in Accuracy and Fluency ranking at roughly 20x lower cost than expert human annotation.

This paper provides a detailed analysis of the NeuroPiano dataset, which comprise 104 audio recordings of student piano performances accompanied with 2255 textual feedback and ratings given by professional pianists. We offer a statistical overview of the dataset, focusing on the standardization of annotations and inter-annotator agreement across 12 evaluative questions concerning performance quality. We also explore the predictive relationship between audio features and teacher ratings via machine learning, as well as annotations provided for text analysis of the responses.

Human image datasets used to develop and evaluate technology should represent the diversity of human phenotypes, including skin tone. Datasets that include skin tone information frequently rely on manual skin tone ratings based on the Fitzpatrick Skin Type (FST) or the Monk Skin Tone (MST) scales in lieu of the actual measured skin tone of the image dataset subjects. However, perceived skin tone is subject to known biases and skin tone appearance in digital images can vary substantially depending on the capture camera and environment, confounding manual ratings. Surprisingly, the relationship between skin-tone ratings and measured skin tone has not been explored. To close this research gap, we measured the relationship between skin tone ratings from existing scales (FST, MST) and skin tone values measured by a calibrated colorimeter. We also propose and assess a novel Colorimetric Skin Tone (CST) scale developed based on prior colorimetric measurements. Using experiments requiring humans to rate their own skin tone and the skin tone of subjects in images, we show that the new CST scale is more sensitive, consistent, and colorimetrically accurate. While skin tone ratings appeared to correct for some color variation across images, they introduced biases related to race and other factors. These biases must be considered before using manual skin-tone ratings in technology evaluations or for engineering decisions.

We present GaRAGe, a large RAG benchmark with human-curated long-form answers and annotations of each grounding passage, allowing a fine-grained evaluation of whether LLMs can identify relevant grounding when generating RAG answers. Our benchmark contains 2366 questions of diverse complexity, dynamism, and topics, and includes over 35K annotated passages retrieved from both private document sets and the Web, to reflect real-world RAG use cases. This makes it an ideal test bed to evaluate an LLM's ability to identify only the relevant information necessary to compose a response, or provide a deflective response when there is insufficient information. Evaluations of multiple state-of-the-art LLMs on GaRAGe show that the models tend to over-summarise rather than (a) ground their answers strictly on the annotated relevant passages (reaching at most a Relevance-Aware Factuality Score of 60%), or (b) deflect when no relevant grounding is available (reaching at most 31% true positive rate in deflections). The F1 in attribution to relevant sources is at most 58.9%, and we show that performance is particularly reduced when answering time-sensitive questions and when having to draw knowledge from sparser private grounding sources.

Achieving a reliable LiDAR-based object detector in autonomous driving is paramount, but its success hinges on obtaining large amounts of precise 3D annotations. Active learning (AL) seeks to mitigate the annotation burden through algorithms that use fewer labels and can attain performance comparable to fully supervised learning. Although AL has shown promise, current approaches prioritize the selection of unlabeled point clouds with high uncertainty and/or diversity, leading to the selection of more instances for labeling and reduced computational efficiency. In this paper, we resort to a novel kernel coding rate maximization (KECOR) strategy which aims to identify the most informative point clouds to acquire labels through the lens of information theory. Greedy search is applied to seek desired point clouds that can maximize the minimal number of bits required to encode the latent features. To determine the uniqueness and informativeness of the selected samples from the model perspective, we construct a proxy network of the 3D detector head and compute the outer product of Jacobians from all proxy layers to form the empirical neural tangent kernel (NTK) matrix. To accommodate both one-stage (i.e., SECOND) and two-stage detectors (i.e., PVRCNN), we further incorporate the classification entropy maximization and well trade-off between detection performance and the total number of bounding boxes selected for annotation. Extensive experiments conducted on two 3D benchmarks and a 2D detection dataset evidence the superiority and versatility of the proposed approach. Our results show that approximately 44% box-level annotation costs and 26% computational time are reduced compared to the state-of-the-art AL method, without compromising detection performance.

Some users of social media are spreading racist, sexist, and otherwise hateful content. For the purpose of training a hate speech detection system, the reliability of the annotations is crucial, but there is no universally agreed-upon definition. We collected potentially hateful messages and asked two groups of internet users to determine whether they were hate speech or not, whether they should be banned or not and to rate their degree of offensiveness. One of the groups was shown a definition prior to completing the survey. We aimed to assess whether hate speech can be annotated reliably, and the extent to which existing definitions are in accordance with subjective ratings. Our results indicate that showing users a definition caused them to partially align their own opinion with the definition but did not improve reliability, which was very low overall. We conclude that the presence of hate speech should perhaps not be considered a binary yes-or-no decision, and raters need more detailed instructions for the annotation.

Warning: this paper contains content that may be offensive or upsetting. Most hate speech datasets neglect the cultural diversity within a single language, resulting in a critical shortcoming in hate speech detection. To address this, we introduce CREHate, a CRoss-cultural English Hate speech dataset. To construct CREHate, we follow a two-step procedure: 1) cultural post collection and 2) cross-cultural annotation. We sample posts from the SBIC dataset, which predominantly represents North America, and collect posts from four geographically diverse English-speaking countries (Australia, United Kingdom, Singapore, and South Africa) using culturally hateful keywords we retrieve from our survey. Annotations are collected from the four countries plus the United States to establish representative labels for each country. Our analysis highlights statistically significant disparities across countries in hate speech annotations. Only 56.2% of the posts in CREHate achieve consensus among all countries, with the highest pairwise label difference rate of 26%. Qualitative analysis shows that label disagreement occurs mostly due to different interpretations of sarcasm and the personal bias of annotators on divisive topics. Lastly, we evaluate large language models (LLMs) under a zero-shot setting and show that current LLMs tend to show higher accuracies on Anglosphere country labels in CREHate. Our dataset and codes are available at: https://github.com/nlee0212/CREHate

The COVID-19 pandemic has been severely impacting global society since December 2019. Massive research has been undertaken to understand the characteristics of the virus and design vaccines and drugs. The related findings have been reported in biomedical literature at a rate of about 10,000 articles on COVID-19 per month. Such rapid growth significantly challenges manual curation and interpretation. For instance, LitCovid is a literature database of COVID-19-related articles in PubMed, which has accumulated more than 200,000 articles with millions of accesses each month by users worldwide. One primary curation task is to assign up to eight topics (e.g., Diagnosis and Treatment) to the articles in LitCovid. Despite the continuing advances in biomedical text mining methods, few have been dedicated to topic annotations in COVID-19 literature. To close the gap, we organized the BioCreative LitCovid track to call for a community effort to tackle automated topic annotation for COVID-19 literature. The BioCreative LitCovid dataset, consisting of over 30,000 articles with manually reviewed topics, was created for training and testing. It is one of the largest multilabel classification datasets in biomedical scientific literature. 19 teams worldwide participated and made 80 submissions in total. Most teams used hybrid systems based on transformers. The highest performing submissions achieved 0.8875, 0.9181, and 0.9394 for macro F1-score, micro F1-score, and instance-based F1-score, respectively. The level of participation and results demonstrate a successful track and help close the gap between dataset curation and method development. The dataset is publicly available via https://ftp.ncbi.nlm.nih.gov/pub/lu/LitCovid/biocreative/ for benchmarking and further development.

Everything a language model sees is tokens. The serving stack knows what each span is -- user input, tool output, instructions -- but the model must keep track of that itself, and it can lose track or be confused: text can be written to read like anything. Prompt injection is a natural exploit of this phenomenon. By scrambling the model's understanding of span identity, an attacker can induce unwanted and potentially dangerous actions. Adding a non-textual channel to the model's input -- a way to communicate span identity beyond text -- mitigates this class of attack. We thus introduce a general steering technique called Semantic Overlays: small learned adapters applied at chosen prefill positions to a frozen model's residual stream. Laying an overlay over a span creates an out-of-band annotation channel that cannot be replicated by tokens. Unlike steering vectors, Semantic Overlays are trained, adaptable, and selectively applied. An overlay can encode complex semantics that reshape how the model perceives the marked span: asked to copy a code snippet under an overlay asserting that it is in a different programming language than it is, the model rewrites the snippet, faithfully, in the asserted language. Overlays are also composable, allow for transparent reading of underlying content, and can carry complex payloads -- including imperatives that the model will follow. An overlay which marks a span as "non-executable" defends against the broad class of prompt injections that add instructions in untrusted context. We report strong results on prompt injection benchmarks: SEP separation rises from 24.3% to 96.5% with utility unchanged (our scoring rule; we also correct a defect in the published grader), TensorTrust attack success rate falls from 34.8% to 6.6%, and all four PIArena attack families drop to 0% compliance, all while marked spans stay readable (92.5% exact copy rate).

Robotic manipulation often requires acting on information that is no longer visible, yet Vision-Language-Action policies are usually evaluated when the current observation largely determines the next action. Existing robotic memory benchmarks expose this gap, but they still rely mainly on final task success and therefore conflate forgetting with manipulation failure. We present MEMOBench, a benchmark for process level memory evaluation in robotic manipulation. MEMOBench includes 30 history dependent tasks, 1{,}500 expert demonstrations, and 4{,}200 executable checkpoint instances from 84 templates. Each checkpoint pairs coarse to fine language with a simulator predicate and labels one memory operation: Storage, Update, or Compression. These annotations define Memory Storage Rate, Memory Update Rate, and Memory Compression Rate, which measure memory fidelity alongside task success. Across standard and memory augmented VLA policies, the strongest memory module baseline reaches only 31.9\% average success rate, and high storage often coexists with weak update and compression. Checkpoint language also supervises semantic, contrastive, and framewise memory alignment objectives, yielding modest gains across different memory operations. MEMOBench provides a diagnostic evaluation suite and training supervision for memory grounded robotic policies. The project page is available at https://github.com/Collab-Gen/MEMOBench.

TACRED (Zhang et al., 2017) is one of the largest, most widely used crowdsourced datasets in Relation Extraction (RE). But, even with recent advances in unsupervised pre-training and knowledge enhanced neural RE, models still show a high error rate. In this paper, we investigate the questions: Have we reached a performance ceiling or is there still room for improvement? And how do crowd annotations, dataset, and models contribute to this error rate? To answer these questions, we first validate the most challenging 5K examples in the development and test sets using trained annotators. We find that label errors account for 8% absolute F1 test error, and that more than 50% of the examples need to be relabeled. On the relabeled test set the average F1 score of a large baseline model set improves from 62.1 to 70.1. After validation, we analyze misclassifications on the challenging instances, categorize them into linguistically motivated error groups, and verify the resulting error hypotheses on three state-of-the-art RE models. We show that two groups of ambiguous relations are responsible for most of the remaining errors and that models may adopt shallow heuristics on the dataset when entities are not masked.

Reasoning in multimodal large language models (MLLMs) has shown strong promise in medical imaging. However, this reasoning is usually free-form text judged only by its final answer, making it hard to interpret and verify, especially in 3D radiology, where a diagnosis should be traceable to evidence in the scan. Existing chest CT question-answering datasets compound this by reducing expert radiology reports to answer-only pairs, dropping the reasoning that links findings to conclusions and omitting the patient history clinicians rely on. As a result, reasoning-capable 3D chest CT MLLMs remain out of reach, as neither the structured supervision needed to train them nor the protocol needed to verify their reasoning yet exists. We introduce CORTEX (Clinically Organized Reasoning and sTructured EXplanation), a structured reasoning benchmark for 3D chest CT. For each question, CORTEX restores the missing reasoning as a four-stage diagnostic trace mirroring a radiologist's workflow: task understanding, visual observation, diagnostic reasoning, and answer synthesis. We generate these traces using frontier large language models with broad medical and general-domain knowledge, then filter and verify them with a stage-level evaluation protocol combining automated rubric scoring with expert radiologist review. Crucially, both the reasoning structure and evaluation rubrics are designed in close collaboration with clinicians. Built on CT-RATE, a large, publicly available chest CT dataset without reasoning annotations, CORTEX comprises 76,177 validated reasoning traces across open-ended VQA, closed-ended VQA, and report generation, providing both the structured supervision and the stage-level evaluation protocol needed to build and evaluate trustworthy reasoning models for 3D chest CT. Our dataset and evaluation code will be made publicly available upon acceptance.

Human feedback has become the de facto standard for evaluating the performance of Large Language Models, and is increasingly being used as a training objective. However, it is not clear which properties of a generated output this single `preference' score captures. We hypothesise that preference scores are subjective and open to undesirable biases. We critically analyse the use of human feedback for both training and evaluation, to verify whether it fully captures a range of crucial error criteria. We find that while preference scores have fairly good coverage, they under-represent important aspects like factuality. We further hypothesise that both preference scores and error annotation may be affected by confounders, and leverage instruction-tuned models to generate outputs that vary along two possible confounding dimensions: assertiveness and complexity. We find that the assertiveness of an output skews the perceived rate of factuality errors, indicating that human annotations are not a fully reliable evaluation metric or training objective. Finally, we offer preliminary evidence that using human feedback as a training objective disproportionately increases the assertiveness of model outputs. We encourage future work to carefully consider whether preference scores are well aligned with the desired objective.

Research on AI-generated text detection has presented a number of approaches to discern human from AI prose, some of which achieving high in-distribution performance. However, real-world applicability has stalled because their outputs are misaligned with the needs of users, such as professors, who are presented with a numeric score that has no attached explanation. We tackle this issue with a novel architecture, TELL, that bakes explainability from the ground-up. While our system still offers a numerical score like other detectors for comparability, TELL takes a fundamentally different approach where we aim to show the user the "tells" by which the model believes a text is AI or human-written, to empower the user to decide who wrote a text using their own judgment and understanding of the context of the writing and its alleged author. We train TELL on a custom SFT dataset of domain-specific authorship annotations, and further refine the system using GRPO with curriculum learning to improve performance. We achieve competitive performance with state-of-the-art detectors (AUROC 0.927) while natively providing annotations that explain the basis for the detector's decision. We further evaluate the quality of our explanations using a dataset of human annotations and report a high (mean 72.3%) win-rate on annotation concreteness, falsifiability, coherence, plausibility and grounding, allowing users to critically think and decide for themselves. Our work thus reframes the problem of AI-generated text detection in a human-centric perspective and paves the way for a new family of detectors that focus on native explainability.

Enhancing the generalization capability of robotic learning to enable robots to operate effectively in diverse, unseen scenes is a fundamental and challenging problem. Existing approaches often depend on pretraining with large-scale data collection, which is labor-intensive and time-consuming, or on semantic data augmentation techniques that necessitate an impractical assumption of flawless upstream object detection in real-world scenarios. In this work, we propose RoboAug, a novel generative data augmentation framework that significantly minimizes the reliance on large-scale pretraining and the perfect visual recognition assumption by requiring only the bounding box annotation of a single image during training. Leveraging this minimal information, RoboAug employs pre-trained generative models for precise semantic data augmentation and integrates a plug-and-play region-contrastive loss to help models focus on task-relevant regions, thereby improving generalization and boosting task success rates. We conduct extensive real-world experiments on three robots, namely UR-5e, AgileX, and Tien Kung 2.0, spanning over 35k rollouts. Empirical results demonstrate that RoboAug significantly outperforms state-of-the-art data augmentation baselines. Specifically, when evaluating generalization capabilities in unseen scenes featuring diverse combinations of backgrounds, distractors, and lighting conditions, our method achieves substantial gains over the baseline without augmentation. The success rates increase from 0.09 to 0.47 on UR-5e, from 0.16 to 0.60 on AgileX, and from 0.19 to 0.67 on Tien Kung 2.0. These results highlight the superior generalization and effectiveness of RoboAug in real-world manipulation tasks. Our project is available at https://x-roboaug.github.io/.

Training of autonomous driving systems requires extensive datasets with precise annotations to attain robust performance. Human annotations suffer from imperfections, and multiple iterations are often needed to produce high-quality datasets. However, manually reviewing large datasets is laborious and expensive. In this paper, we introduce AutoVDC (Automated Vision Data Cleaning) framework and investigate the utilization of Vision-Language Models (VLMs) to automatically identify erroneous annotations in vision datasets, thereby enabling users to eliminate these errors and enhance data quality. We validate our approach using the KITTI and nuImages datasets, which contain object detection benchmarks for autonomous driving. To test the effectiveness of AutoVDC, we create dataset variants with intentionally injected erroneous annotations and observe the error detection rate of our approach. Additionally, we compare the detection rates using different VLMs and explore the impact of VLM fine-tuning on our pipeline. The results demonstrate our method's high performance in error detection and data cleaning experiments, indicating its potential to significantly improve the reliability and accuracy of large-scale production datasets in autonomous driving.

We present Distributional RewArds for Generative OptimizatioN (DRAGON), a versatile framework for fine-tuning media generation models towards a desired outcome. Compared with traditional reinforcement learning with human feedback (RLHF) or pairwise preference approaches such as direct preference optimization (DPO), DRAGON is more flexible. It can optimize reward functions that evaluate either individual examples or distributions of them, making it compatible with a broad spectrum of instance-wise, instance-to-distribution, and distribution-to-distribution rewards. Leveraging this versatility, we construct novel reward functions by selecting an encoder and a set of reference examples to create an exemplar distribution. When cross-modality encoders such as CLAP are used, the reference examples may be of a different modality (e.g., text versus audio). Then, DRAGON gathers online and on-policy generations, scores them to construct a positive demonstration set and a negative set, and leverages the contrast between the two sets to maximize the reward. For evaluation, we fine-tune an audio-domain text-to-music diffusion model with 20 different reward functions, including a custom music aesthetics model, CLAP score, Vendi diversity, and Frechet audio distance (FAD). We further compare instance-wise (per-song) and full-dataset FAD settings while ablating multiple FAD encoders and reference sets. Over all 20 target rewards, DRAGON achieves an 81.45% average win rate. Moreover, reward functions based on exemplar sets indeed enhance generations and are comparable to model-based rewards. With an appropriate exemplar set, DRAGON achieves a 60.95% human-voted music quality win rate without training on human preference annotations. As such, DRAGON exhibits a new approach to designing and optimizing reward functions for improving human-perceived quality. Sound examples at https://ml-dragon.github.io/web.

When developing new large language models (LLMs), a key step is evaluating their final performance, often by computing the win-rate against a reference model based on external feedback. Human feedback is the gold standard, particularly for capturing nuanced qualities like coherence, readability, and alignment with human expectations. However, human evaluations are costly -- even for large tech companies -- and when conducted with active users, they may negatively impact user experience. A promising alternative is synthetic feedback, where evaluations are conducted by other large language models, including reward models. While this eliminates the need for costly human annotations, it introduces biases that may distort the evaluation process. In this work, we propose a statistically principled framework that integrates human and synthetic feedback to reduce reliance on human annotations while maintaining unbiased win-rate calculations. Our experiments demonstrate a reduction in human annotations by up to 12.2% with an off-the-shelf synthetic evaluator and up to 24.8% with a finetuned variant. Apart from being generalizable, scalable, and free of hyper-parameter tuning, our method offers predictable annotation savings, which can be estimated based on data-dependent characteristics.

Multi-turn interactions between large language models (LLMs) and users naturally include implicit feedback signals. If an LLM responds in an unexpected way to an instruction, the user is likely to signal it by rephrasing the request, expressing frustration, or pivoting to an alternative task. Such signals are task-independent and occupy a relatively constrained subspace of language, allowing the LLM to identify them even if it fails on the actual task. This creates an avenue for continually learning from interactions without additional annotations. We introduce ReSpect, a method to learn from such signals in past interactions via retrospection. We deploy ReSpect in a new multimodal interaction scenario, where humans instruct an LLM to solve an abstract reasoning task with a combinatorial solution space. Through thousands of interactions with humans, we show how ReSpect gradually improves task completion rate from 31% to 82%, all without any external annotation.

In this study, we investigated the ability of the large language model (LLM) to enhance healthcare data interoperability. We leveraged the LLM to convert clinical texts into their corresponding FHIR resources. Our experiments, conducted on 3,671 snippets of clinical text, demonstrated that the LLM not only streamlines the multi-step natural language processing and human calibration processes but also achieves an exceptional accuracy rate of over 90% in exact matches when compared to human annotations.

We introduce a corpus of 7,032 sentences rated by human annotators for formality, informativeness, and implicature on a 1-7 scale. The corpus was annotated using Amazon Mechanical Turk. Reliability in the obtained judgments was examined by comparing mean ratings across two MTurk experiments, and correlation with pilot annotations (on sentence formality) conducted in a more controlled setting. Despite the subjectivity and inherent difficulty of the annotation task, correlations between mean ratings were quite encouraging, especially on formality and informativeness. We further explored correlation between the three linguistic variables, genre-wise variation of ratings and correlations within genres, compatibility with automatic stylistic scoring, and sentential make-up of a document in terms of style. To date, our corpus is the largest sentence-level annotated corpus released for formality, informativeness, and implicature.

This paper presents a new challenging information extraction task in the domain of materials science. We develop an annotation scheme for marking information on experiments related to solid oxide fuel cells in scientific publications, such as involved materials and measurement conditions. With this paper, we publish our annotation guidelines, as well as our SOFC-Exp corpus consisting of 45 open-access scholarly articles annotated by domain experts. A corpus and an inter-annotator agreement study demonstrate the complexity of the suggested named entity recognition and slot filling tasks as well as high annotation quality. We also present strong neural-network based models for a variety of tasks that can be addressed on the basis of our new data set. On all tasks, using BERT embeddings leads to large performance gains, but with increasing task complexity, adding a recurrent neural network on top seems beneficial. Our models will serve as competitive baselines in future work, and analysis of their performance highlights difficult cases when modeling the data and suggests promising research directions.

This paper introduces a novel, entity-aware metric, termed as Radiological Report (Text) Evaluation (RaTEScore), to assess the quality of medical reports generated by AI models. RaTEScore emphasizes crucial medical entities such as diagnostic outcomes and anatomical details, and is robust against complex medical synonyms and sensitive to negation expressions. Technically, we developed a comprehensive medical NER dataset, RaTE-NER, and trained an NER model specifically for this purpose. This model enables the decomposition of complex radiological reports into constituent medical entities. The metric itself is derived by comparing the similarity of entity embeddings, obtained from a language model, based on their types and relevance to clinical significance. Our evaluations demonstrate that RaTEScore aligns more closely with human preference than existing metrics, validated both on established public benchmarks and our newly proposed RaTE-Eval benchmark.

Annotation quality is difficult to sustain when campaigns span weeks or months with small annotator pools. We present a Setswana sentiment dataset of 3,565 tweets annotated by three native-speaker annotators across eight batches and examine why inter-annotator agreement (IAA) declines over time. Despite an aggregate Randolph's free-marginal Kappa of κ= 0.76, "excellent," per-batch κ falls by more than 32 points across the annotation task. Through six targeted analyses, we find that (i) label confusion concentrates on the negative/neutral boundary, (ii) two annotators show run-length drift consistent with autopilot labeling, and (iii) the dominant predictor of κ is temporal simultaneity: tweets labeled within one minute achieve κ= 0.98, while those labeled more than a day apart reach only κ= 0.65. Annotation speed and tweet-level linguistic features show no meaningful association with κ. We benchmark three open multilingual encoders and proprietary models (GPT-5 and Gemini) on three-class sentiment classification; fine-tuning yields gains of 29 to 43 macro-F1 points over pretrained baselines, with GPT-5 few-shot leading overall (62.2 macro-F1). We release the dataset, per-annotation timestamps, and analysis code to support reproducible quality auditing for future African language NLP resources.

The rise of large language models (LLMs) has brought a critical need for high-quality human-labeled data, particularly for processes like human feedback and evaluation. A common practice is to label data via consensus annotation over crowdworker judgments. However, annotators' judgments for subjective tasks can differ in many ways: they may have different qualitative judgments about an example, and they may map those to a labeling scheme in different ways. We show that these nuances can be captured by natural language explanations, and propose a method to rescale ordinal annotations and explanations using LLMs. Specifically, we feed annotators' Likert ratings and corresponding explanations into an LLM and prompt it to produce a numeric score anchored in a scoring rubric. These scores should reflect the annotators' underlying assessments of the example. The rubric can be designed or modified after annotation, and include distinctions that may not have been known when the original error taxonomy was devised. We explore our technique in the context of rating system outputs for a document-grounded question answering task, where LLMs achieve near-human performance. Our method rescales the raw judgments without impacting agreement and brings the scores closer to human judgments grounded in the same scoring rubric.

Filtering and annotating textual data are routine tasks in many areas, like social media or news analytics. Automating these tasks allows to scale the analyses wrt. speed and breadth of content covered and decreases the manual effort required. Due to technical advancements in Natural Language Processing, specifically the success of large foundation models, a new tool for automating such annotation processes by using a text-to-text interface given written guidelines without providing training samples has become available. In this work, we assess these advancements in-the-wild by empirically testing them in an annotation task on German Twitter data about social and political European crises. We compare the prompt-based results with our human annotation and preceding classification approaches, including Naive Bayes and a BERT-based fine-tuning/domain adaptation pipeline. Our results show that the prompt-based approach - despite being limited by local computation resources during the model selection - is comparable with the fine-tuned BERT but without any annotated training data. Our findings emphasize the ongoing paradigm shift in the NLP landscape, i.e., the unification of downstream tasks and elimination of the need for pre-labeled training data.

This paper explores the use of open generative Large Language Models (LLMs) for annotation tasks in the social sciences. The study highlights the challenges associated with proprietary models, such as limited reproducibility and privacy concerns, and advocates for the adoption of open (source) models that can be operated on independent devices. Two examples of annotation tasks, sentiment analysis in tweets and identification of leisure activities in childhood aspirational essays are provided. The study evaluates the performance of different prompting strategies and models (neural-chat-7b-v3-2, Starling-LM-7B-alpha, openchat_3.5, zephyr-7b-alpha and zephyr-7b-beta). The results indicate the need for careful validation and tailored prompt engineering. The study highlights the advantages of open models for data privacy and reproducibility.

Retrieval models are often evaluated on partially-annotated datasets. Each query is mapped to a few relevant texts and the remaining corpus is assumed to be irrelevant. As a result, models that successfully retrieve false negatives are punished in evaluation. Unfortunately, completely annotating all texts for every query is not resource efficient. In this work, we show that using partially-annotated datasets in evaluation can paint a distorted picture. We curate D-MERIT, a passage retrieval evaluation set from Wikipedia, aspiring to contain all relevant passages for each query. Queries describe a group (e.g., ``journals about linguistics'') and relevant passages are evidence that entities belong to the group (e.g., a passage indicating that Language is a journal about linguistics). We show that evaluating on a dataset containing annotations for only a subset of the relevant passages might result in misleading ranking of the retrieval systems and that as more relevant texts are included in the evaluation set, the rankings converge. We propose our dataset as a resource for evaluation and our study as a recommendation for balance between resource-efficiency and reliable evaluation when annotating evaluation sets for text retrieval.

As text generated by large language models proliferates, it becomes vital to understand how humans engage with such text, and whether or not they are able to detect when the text they are reading did not originate with a human writer. Prior work on human detection of generated text focuses on the case where an entire passage is either human-written or machine-generated. In this paper, we study a more realistic setting where text begins as human-written and transitions to being generated by state-of-the-art neural language models. We show that, while annotators often struggle at this task, there is substantial variance in annotator skill and that given proper incentives, annotators can improve at this task over time. Furthermore, we conduct a detailed comparison study and analyze how a variety of variables (model size, decoding strategy, fine-tuning, prompt genre, etc.) affect human detection performance. Finally, we collect error annotations from our participants and use them to show that certain textual genres influence models to make different types of errors and that certain sentence-level features correlate highly with annotator selection. We release the RoFT dataset: a collection of over 21,000 human annotations paired with error classifications to encourage future work in human detection and evaluation of generated text.

Researchers have proposed numerous text-to-SQL techniques to streamline data analytics and accelerate the development of data-driven applications. To compare these techniques and select the best one for deployment, the community depends on public benchmarks and their leaderboards. Since these benchmarks heavily rely on human annotations during question construction and answer evaluation, the validity of the annotations is crucial. In this paper, we conduct an empirical study that (i) benchmarks annotation error rates for two widely used text-to-SQL benchmarks, BIRD and Spider 2.0-Snow, and (ii) corrects a subset of the BIRD development (Dev) set to measure the impact of annotation errors on text-to-SQL agent performance and leaderboard rankings. Through expert analysis, we show that BIRD Mini-Dev and Spider 2.0-Snow have error rates of 52.8% and 62.8%, respectively. We re-evaluate all 16 open-source agents from the BIRD leaderboard on both the original and the corrected BIRD Dev subsets. We show that performance changes range from -7% to 31% (in relative terms) and rank changes range from -9 to +9 positions. We further assess whether these impacts generalize to the full BIRD Dev set. We find that the rankings of agents on the uncorrected subset correlate strongly with those on the full Dev set (Spearman's r_s=0.85, p=3.26e-5), whereas they correlate weakly with those on the corrected subset (Spearman's r_s=0.32, p=0.23). These findings show that annotation errors can significantly distort reported performance and rankings, potentially misguiding research directions or deployment choices. Our code and data are available at https://github.com/uiuc-kang-lab/text_to_sql_benchmarks.

The NLP community has long advocated for the construction of multi-annotator datasets to better capture the nuances of language interpretation, subjectivity, and ambiguity. This paper conducts a retrospective study to show how performance scores can vary when a dataset expands from a single annotation per instance to multiple annotations. We propose a novel multi-annotator simulation process to generate datasets with varying annotation budgets. We show that similar datasets with the same annotation budget can lead to varying performance gains. Our findings challenge the popular belief that models trained on multi-annotation examples always lead to better performance than models trained on single or few-annotation examples.

The increased use of large language models (LLMs) across a variety of real-world applications calls for mechanisms to verify the factual accuracy of their outputs. In this work, we present a holistic end-to-end solution for annotating the factuality of LLM-generated responses, which encompasses a multi-stage annotation scheme designed to yield detailed labels concerning the verifiability and factual inconsistencies found in LLM outputs. We design and build an annotation tool to speed up the labelling procedure and ease the workload of raters. It allows flexible incorporation of automatic results in any stage, e.g. automatically-retrieved evidence. We further construct an open-domain document-level factuality benchmark in three-level granularity: claim, sentence and document. Preliminary experiments show that FacTool, FactScore and Perplexity.ai are struggling to identify false claims with the best F1=0.53. Annotation tool, benchmark and code are available at https://github.com/yuxiaw/Factcheck-GPT.

In this paper we revisit automatic metrics for paraphrase evaluation and obtain two findings that disobey conventional wisdom: (1) Reference-free metrics achieve better performance than their reference-based counterparts. (2) Most commonly used metrics do not align well with human annotation. Underlying reasons behind the above findings are explored through additional experiments and in-depth analyses. Based on the experiments and analyses, we propose ParaScore, a new evaluation metric for paraphrase generation. It possesses the merits of reference-based and reference-free metrics and explicitly models lexical divergence. Experimental results demonstrate that ParaScore significantly outperforms existing metrics.

Human evaluation is increasingly critical for assessing large language models, capturing linguistic nuances, and reflecting user preferences more accurately than traditional automated metrics. However, the resource-intensive nature of this type of annotation process poses significant challenges. The key question driving our work: "is it feasible to minimize human-in-the-loop feedback by prioritizing data instances which most effectively distinguish between models?" We evaluate several metric-based methods and find that these metrics enhance the efficiency of human evaluations by minimizing the number of required annotations, thus saving time and cost, while ensuring a robust performance evaluation. We show that our method is effective across widely used model families, reducing instances of indecisive (or "tie") outcomes by up to 54% compared to a random sample when focusing on the top-20 percentile of prioritized instances. This potential reduction in required human effort positions our approach as a valuable strategy in future large language model evaluations.

Recently, open-domain question answering systems have begun to rely heavily on annotated datasets to train neural passage retrievers. However, manually annotating such datasets is both difficult and time-consuming, which limits their availability for less popular languages. In this work, we experiment with several methods for automatically collecting weakly labeled datasets and show how they affect the performance of the neural passage retrieval models. As a result of our work, we publish the MAUPQA dataset, consisting of nearly 400,000 question-passage pairs for Polish, as well as the HerBERT-QA neural retriever.

Evaluating text revision in scientific writing remains a challenge, as traditional metrics such as ROUGE and BERTScore primarily focus on similarity rather than capturing meaningful improvements. In this work, we analyse and identify the limitations of these metrics and explore alternative evaluation methods that better align with human judgments. We first conduct a manual annotation study to assess the quality of different revisions. Then, we investigate reference-free evaluation metrics from related NLP domains. Additionally, we examine LLM-as-a-judge approaches, analysing their ability to assess revisions with and without a gold reference. Our results show that LLMs effectively assess instruction-following but struggle with correctness, while domain-specific metrics provide complementary insights. We find that a hybrid approach combining LLM-as-a-judge evaluation and task-specific metrics offers the most reliable assessment of revision quality.

Automatic evaluation of retrieval augmented generation (RAG) systems relies on fine-grained dimensions like faithfulness and relevance, as judged by expert human annotators. Meta-evaluation benchmarks support the development of automatic evaluators that correlate well with human judgement. However, existing benchmarks predominantly focus on English or use translated data, which fails to capture cultural nuances. A native approach provides a better representation of the end user experience. In this work, we develop a Multilingual End-to-end Meta-Evaluation RAG benchmark (MEMERAG). Our benchmark builds on the popular MIRACL dataset, using native-language questions and generating responses with diverse large language models (LLMs), which are then assessed by expert annotators for faithfulness and relevance. We describe our annotation process and show that it achieves high inter-annotator agreement. We then analyse the performance of the answer-generating LLMs across languages as per the human evaluators. Finally we apply the dataset to our main use-case which is to benchmark multilingual automatic evaluators (LLM-as-a-judge). We show that our benchmark can reliably identify improvements offered by advanced prompting techniques and LLMs. Our dataset is available at https://github.com/amazon-science/MEMERAG

Multimodal Review Helpfulness Prediction (MRHP) is an essential task in recommender systems, particularly in E-commerce platforms. Determining the helpfulness of user-generated reviews enhances user experience and improves consumer decision-making. However, existing datasets focus predominantly on English and Indonesian, resulting in a lack of linguistic diversity, especially for low-resource languages such as Vietnamese. In this paper, we introduce ViMRHP (Vietnamese Multimodal Review Helpfulness Prediction), a large-scale benchmark dataset for MRHP task in Vietnamese. This dataset covers four domains, including 2K products with 46K reviews. Meanwhile, a large-scale dataset requires considerable time and cost. To optimize the annotation process, we leverage AI to assist annotators in constructing the ViMRHP dataset. With AI assistance, annotation time is reduced (90 to 120 seconds per task down to 20 to 40 seconds per task) while maintaining data quality and lowering overall costs by approximately 65%. However, AI-generated annotations still have limitations in complex annotation tasks, which we further examine through a detailed performance analysis. In our experiment on ViMRHP, we evaluate baseline models on human-verified and AI-generated annotations to assess their quality differences. The ViMRHP dataset is publicly available at https://github.com/trng28/ViMRHP

Recently proposed systems for open-domain question answering (OpenQA) require large amounts of training data to achieve state-of-the-art performance. However, data annotation is known to be time-consuming and therefore expensive to acquire. As a result, the appropriate datasets are available only for a handful of languages (mainly English and Chinese). In this work, we introduce and publicly release PolQA, the first Polish dataset for OpenQA. It consists of 7,000 questions, 87,525 manually labeled evidence passages, and a corpus of over 7,097,322 candidate passages. Each question is classified according to its formulation, type, as well as entity type of the answer. This resource allows us to evaluate the impact of different annotation choices on the performance of the QA system and propose an efficient annotation strategy that increases the passage retrieval performance by 10.55 p.p. while reducing the annotation cost by 82%.

Large language models (LLMs) have shown impressive results while requiring little or no direct supervision. Further, there is mounting evidence that LLMs may have potential in information-seeking scenarios. We believe the ability of an LLM to attribute the text that it generates is likely to be crucial in this setting. We formulate and study Attributed QA as a key first step in the development of attributed LLMs. We propose a reproducible evaluation framework for the task and benchmark a broad set of architectures. We take human annotations as a gold standard and show that a correlated automatic metric is suitable for development. Our experimental work gives concrete answers to two key questions (How to measure attribution?, and How well do current state-of-the-art methods perform on attribution?), and give some hints as to how to address a third (How to build LLMs with attribution?).

This study examines the performance of open-source Large Language Models (LLMs) in text annotation tasks and compares it with proprietary models like ChatGPT and human-based services such as MTurk. While prior research demonstrated the high performance of ChatGPT across numerous NLP tasks, open-source LLMs like HugginChat and FLAN are gaining attention for their cost-effectiveness, transparency, reproducibility, and superior data protection. We assess these models using both zero-shot and few-shot approaches and different temperature parameters across a range of text annotation tasks. Our findings show that while ChatGPT achieves the best performance in most tasks, open-source LLMs not only outperform MTurk but also demonstrate competitive potential against ChatGPT in specific tasks.

Evaluating the quality of search, ranking and RAG systems traditionally requires a significant number of human relevance annotations. In recent times, several deployed systems have explored the usage of Large Language Models (LLMs) as automated judges for this task while their inherent biases prevent direct use for metric estimation. We present a statistical framework extending Prediction-Powered Inference (PPI) that combines minimal human annotations with LLM judgments to produce reliable estimates of metrics which require sub-instance annotations. Our method requires as few as 100 human-annotated queries and 10,000 unlabeled examples, reducing annotation requirements significantly compared to traditional approaches. We formulate our proposed framework (PRECISE) for inference of relevance uplift for an LLM-based query reformulation application, extending PPI to sub-instance annotations at the query-document level. By reformulating the metric-integration space, we reduced the computational complexity from O(2^|C|) to O(2^K), where |C| represents corpus size (in order of millions). Detailed experiments across prominent retrieval datasets demonstrate that our method reduces the variance of estimates for the business-critical Precision@K metric, while effectively correcting for LLM bias in low-resource settings.

Precise geocoding and time normalization for text requires that location and time phrases be identified. Many state-of-the-art geoparsers and temporal parsers suffer from low recall. Categories commonly missed by parsers are: nouns used in a non- spatiotemporal sense, adjectival and adverbial phrases, prepositional phrases, and numerical phrases. We collected and annotated data set by querying commercial web searches API with such spatiotemporal expressions as were missed by state-of-the- art parsers. Due to the high cost of sentence annotation, active learning was used to label training data, and a new strategy was designed to better select training examples to reduce labeling cost. For the learning algorithm, we applied an average perceptron trained Featurized Hidden Markov Model (FHMM). Five FHMM instances were used to create an ensemble, with the output phrase selected by voting. Our ensemble model was tested on a range of sequential labeling tasks, and has shown competitive performance. Our contributions include (1) an new dataset annotated with named entities and expanded spatiotemporal expressions; (2) a comparison of inference algorithms for ensemble models showing the superior accuracy of Belief Propagation over Viterbi Decoding; (3) a new example re-weighting method for active ensemble learning that 'memorizes' the latest examples trained; (4) a spatiotemporal parser that jointly recognizes expanded spatiotemporal expressions as well as named entities.

We present a corpus of 5,000 richly annotated abstracts of medical articles describing clinical randomized controlled trials. Annotations include demarcations of text spans that describe the Patient population enrolled, the Interventions studied and to what they were Compared, and the Outcomes measured (the `PICO' elements). These spans are further annotated at a more granular level, e.g., individual interventions within them are marked and mapped onto a structured medical vocabulary. We acquired annotations from a diverse set of workers with varying levels of expertise and cost. We describe our data collection process and the corpus itself in detail. We then outline a set of challenging NLP tasks that would aid searching of the medical literature and the practice of evidence-based medicine.

Is it possible to build a general and automatic natural language generation (NLG) evaluation metric? Existing learned metrics either perform unsatisfactorily or are restricted to tasks where large human rating data is already available. We introduce SESCORE, a model-based metric that is highly correlated with human judgements without requiring human annotation, by utilizing a novel, iterative error synthesis and severity scoring pipeline. This pipeline applies a series of plausible errors to raw text and assigns severity labels by simulating human judgements with entailment. We evaluate SESCORE against existing metrics by comparing how their scores correlate with human ratings. SESCORE outperforms all prior unsupervised metrics on multiple diverse NLG tasks including machine translation, image captioning, and WebNLG text generation. For WMT 20/21 En-De and Zh-En, SESCORE improve the average Kendall correlation with human judgement from 0.154 to 0.195. SESCORE even achieves comparable performance to the best supervised metric COMET, despite receiving no human-annotated training data.

The task of annotating data into concise summaries poses a significant challenge across various domains, frequently requiring the allocation of significant time and specialized knowledge by human experts. Despite existing efforts to use large language models for annotation tasks, significant problems such as limited applicability to unlabeled data, the absence of self-supervised methods, and the lack of focus on complex structured data still persist. In this work, we propose a GPT self-supervision annotation method, which embodies a generating-recovering paradigm that leverages the one-shot learning capabilities of the Generative Pretrained Transformer (GPT). The proposed approach comprises a one-shot tuning phase followed by a generation phase. In the one-shot tuning phase, we sample a data from the support set as part of the prompt for GPT to generate a textual summary, which is then used to recover the original data. The alignment score between the recovered and original data serves as a self-supervision navigator to refine the process. In the generation stage, the optimally selected one-shot sample serves as a template in the prompt and is applied to generating summaries from challenging datasets. The annotation performance is evaluated by tuning several human feedback reward networks and by calculating alignment scores between original and recovered data at both sentence and structure levels. Our self-supervised annotation method consistently achieves competitive scores, convincingly demonstrating its robust strength in various data-to-summary annotation tasks.

In this paper we introduce a new publicly available dataset for verification against textual sources, FEVER: Fact Extraction and VERification. It consists of 185,445 claims generated by altering sentences extracted from Wikipedia and subsequently verified without knowledge of the sentence they were derived from. The claims are classified as Supported, Refuted or NotEnoughInfo by annotators achieving 0.6841 in Fleiss kappa. For the first two classes, the annotators also recorded the sentence(s) forming the necessary evidence for their judgment. To characterize the challenge of the dataset presented, we develop a pipeline approach and compare it to suitably designed oracles. The best accuracy we achieve on labeling a claim accompanied by the correct evidence is 31.87%, while if we ignore the evidence we achieve 50.91%. Thus we believe that FEVER is a challenging testbed that will help stimulate progress on claim verification against textual sources.

Learning from human feedback has enabled the alignment of language models (LMs) with human preferences. However, directly collecting human preferences can be expensive, time-consuming, and can have high variance. An appealing alternative is to distill preferences from LMs as a source of synthetic annotations as they are more consistent, cheaper, and scale better than human annotation; however, they are also prone to biases and errors. In this work, we introduce a routing framework that combines inputs from humans and LMs to achieve better annotation quality, while reducing the total cost of human annotation. The crux of our approach is to identify preference instances that will benefit from human annotations. We formulate this as an optimization problem: given a preference dataset and an evaluation metric, we train a performance prediction model to predict a reward model's performance on an arbitrary combination of human and LM annotations and employ a routing strategy that selects a combination that maximizes predicted performance. We train the performance prediction model on MultiPref, a new preference dataset with 10K instances paired with human and LM labels. We show that the selected hybrid mixture of LM and direct human preferences using our routing framework achieves better reward model performance compared to using either one exclusively. We simulate selective human preference collection on three other datasets and show that our method generalizes well to all three. We analyze features from the routing model to identify characteristics of instances that can benefit from human feedback, e.g., prompts with a moderate safety concern or moderate intent complexity. We release the dataset, annotation platform, and source code used in this study to foster more efficient and accurate preference collection in the future.

High-quality data annotation is an essential but laborious and costly aspect of developing machine learning-based software. We explore the inherent tradeoff between annotation accuracy and cost by detecting and removing minority reports -- instances where annotators provide incorrect responses -- that indicate unnecessary redundancy in task assignments. We propose an approach to prune potentially redundant annotation task assignments before they are executed by estimating the likelihood of an annotator disagreeing with the majority vote for a given task. Our approach is informed by an empirical analysis over computer vision datasets annotated by a professional data annotation platform, which reveals that the likelihood of a minority report event is dependent primarily on image ambiguity, worker variability, and worker fatigue. Simulations over these datasets show that we can reduce the number of annotations required by over 60% with a small compromise in label quality, saving approximately 6.6 days-equivalent of labor. Our approach provides annotation service platforms with a method to balance cost and dataset quality. Machine learning practitioners can tailor annotation accuracy levels according to specific application needs, thereby optimizing budget allocation while maintaining the data quality necessary for critical settings like autonomous driving technology.

Extracting who says what to whom is a crucial part in analyzing human communication in today's abundance of data such as online news articles. Yet, the lack of annotated data for this task in German news articles severely limits the quality and usability of possible systems. To remedy this, we present a new, freely available, creative-commons-licensed dataset for quotation attribution in German news articles based on WIKINEWS. The dataset provides curated, high-quality annotations across 1000 documents (250,000 tokens) in a fine-grained annotation schema enabling various downstream uses for the dataset. The annotations not only specify who said what but also how, in which context, to whom and define the type of quotation. We specify our annotation schema, describe the creation of the dataset and provide a quantitative analysis. Further, we describe suitable evaluation metrics, apply two existing systems for quotation attribution, discuss their results to evaluate the utility of our dataset and outline use cases of our dataset in downstream tasks.

Modern AI systems depend on annotated data, but annotation is rarely treated as the act of measurement that it is. Instead, annotation quality is commonly reduced to agreement: if multiple annotators assign the same annotation to a data instance, the annotations are taken to be high-quality. Yet agreement does not establish whether annotations validly capture the underlying concept they are meant to represent. In this paper, we argue that data annotation should be understood as a measurement problem. Like other forms of measurement, annotation requires defining a concept, operationalizing it through an instrument, applying that instrument, and evaluating the reliability and validity of the resulting measurements. Drawing on a literature review of annotation quality research (N=132) and semi-structured interviews with annotation team members (N=10), we develop a framework for diagnosing and correcting annotation issues. First, we map key decision points across annotation processes - including task design, annotator management, quality assessment, quality improvement, and adjudication - that shape annotation outcomes. Second, we identify five distinct sources of annotation issues: error, ambiguity, impossibility, subjectivity, and annotator identity. Annotation problems that appear similar at the level of outcomes often require different process-level interventions based on their sources. Finally, we translate measurement theory into practical guidance for annotation teams, showing how assessments of reliability and validity can move beyond agreement alone. By reframing annotation as measurement, we offer a conceptual foundation for improving the quality of annotated data used in AI research and practice.

Learning specific hands-on skills such as cooking, car maintenance, and home repairs increasingly happens via instructional videos. The user experience with such videos is known to be improved by meta-information such as time-stamped annotations for the main steps involved. Generating such annotations automatically is challenging, and we describe here two relevant contributions. First, we construct and release a new dense video captioning dataset, Video Timeline Tags (ViTT), featuring a variety of instructional videos together with time-stamped annotations. Second, we explore several multimodal sequence-to-sequence pretraining strategies that leverage large unsupervised datasets of videos and caption-like texts. We pretrain and subsequently finetune dense video captioning models using both YouCook2 and ViTT. We show that such models generalize well and are robust over a wide variety of instructional videos.

Textual domain is a crucial property within the Natural Language Processing (NLP) community due to its effects on downstream model performance. The concept itself is, however, loosely defined and, in practice, refers to any non-typological property, such as genre, topic, medium or style of a document. We investigate the core notion of domains via human proficiency in identifying related intrinsic textual properties, specifically the concepts of genre (communicative purpose) and topic (subject matter). We publish our annotations in *TGeGUM*: A collection of 9.1k sentences from the GUM dataset (Zeldes, 2017) with single sentence and larger context (i.e., prose) annotations for one of 11 genres (source type), and its topic/subtopic as per the Dewey Decimal library classification system (Dewey, 1979), consisting of 10/100 hierarchical topics of increased granularity. Each instance is annotated by three annotators, for a total of 32.7k annotations, allowing us to examine the level of human disagreement and the relative difficulty of each annotation task. With a Fleiss' kappa of at most 0.53 on the sentence level and 0.66 at the prose level, it is evident that despite the ubiquity of domains in NLP, there is little human consensus on how to define them. By training classifiers to perform the same task, we find that this uncertainty also extends to NLP models.

Data annotation is an important but time-consuming and costly procedure. To sort a text into two classes, the very first thing we need is a good annotation guideline, establishing what is required to qualify for each class. In the literature, the difficulties associated with an appropriate data annotation has been underestimated. In this paper, we present a novel approach to automatically construct an annotated sentiment corpus for Algerian dialect (a Maghrebi Arabic dialect). The construction of this corpus is based on an Algerian sentiment lexicon that is also constructed automatically. The presented work deals with the two widely used scripts on Arabic social media: Arabic and Arabizi. The proposed approach automatically constructs a sentiment corpus containing 8000 messages (where 4000 are dedicated to Arabic and 4000 to Arabizi). The achieved F1-score is up to 72% and 78% for an Arabic and Arabizi test sets, respectively. Ongoing work is aimed at integrating transliteration process for Arabizi messages to further improve the obtained results.

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.

This article presents the creation of an Estonian-language dataset for document-level subjectivity, analyzes the resulting annotations, and reports an initial experiment of automatic subjectivity analysis using a large language model (LLM). The dataset comprises of 1,000 documents-300 journalistic articles and 700 randomly selected web texts-each rated for subjectivity on a continuous scale from 0 (fully objective) to 100 (fully subjective) by four annotators. As the inter-annotator correlations were moderate, with some texts receiving scores at the opposite ends of the scale, a subset of texts with the most divergent scores was re-annotated, with the inter-annotator correlation improving. In addition to human annotations, the dataset includes scores generated by GPT-5 as an experiment on annotation automation. These scores were similar to human annotators, however several differences emerged, suggesting that while LLM based automatic subjectivity scoring is feasible, it is not an interchangeable alternative to human annotation, and its suitability depends on the intended application.

Political scientists are rapidly adopting large language models (LLMs) for text annotation, yet the sensitivity of annotation results to implementation choices remains poorly understood. Most evaluations test a single model or configuration; how model choice, model size, learning approach, and prompt style interact, and whether popular "best practices" survive controlled comparison, are largely unexplored. We present a controlled evaluation of these pipeline choices, testing six open-weight models across four political science annotation tasks under identical quantisation, hardware, and prompt-template conditions. Our central finding is methodological: interaction effects dominate main effects, so seemingly reasonable pipeline choices can become consequential researcher degrees of freedom. No single model, prompt style, or learning approach is uniformly superior, and the best-performing model varies across tasks. Two corollaries follow. First, model size is an unreliable guide both to cost and to performance: cross-family efficiency differences are so large that some larger models are less resource-intensive than much smaller alternatives, while within model families mid-range variants often match or exceed larger counterparts. Second, widely recommended prompt engineering techniques yield inconsistent and sometimes negative effects on annotation performance. We use these benchmark results to develop a validation-first framework - with a principled ordering of pipeline decisions, guidance on prompt freezing and held-out evaluation, reporting standards, and open-source tools - to help researchers navigate this decision space transparently.

Many recent approaches to natural language tasks are built on the remarkable abilities of large language models. Large language models can perform in-context learning, where they learn a new task from a few task demonstrations, without any parameter updates. This work examines the implications of in-context learning for the creation of datasets for new natural language tasks. Departing from recent in-context learning methods, we formulate an annotation-efficient, two-step framework: selective annotation that chooses a pool of examples to annotate from unlabeled data in advance, followed by prompt retrieval that retrieves task examples from the annotated pool at test time. Based on this framework, we propose an unsupervised, graph-based selective annotation method, voke-k, to select diverse, representative examples to annotate. Extensive experiments on 10 datasets (covering classification, commonsense reasoning, dialogue, and text/code generation) demonstrate that our selective annotation method improves the task performance by a large margin. On average, vote-k achieves a 12.9%/11.4% relative gain under an annotation budget of 18/100, as compared to randomly selecting examples to annotate. Compared to state-of-the-art supervised finetuning approaches, it yields similar performance with 10-100x less annotation cost across 10 tasks. We further analyze the effectiveness of our framework in various scenarios: language models with varying sizes, alternative selective annotation methods, and cases where there is a test data domain shift. We hope that our studies will serve as a basis for data annotations as large language models are increasingly applied to new tasks. Our code is available at https://github.com/HKUNLP/icl-selective-annotation.

Privacy redaction must remove personal information while preserving relationships expressed in text. We develop a multilingual named-entity tagger with fine-grained distinctions supporting varied redaction policies and methods for cheaply learning additional distinctions. We fine-tune a multilingual encoder with an affine span-tagging head on frontier-model annotations in 35 languages, replay mapped human gold with coverage-aware masking so unannotated types are not treated as negatives, and repair subword boundaries with a learned +/-1-character adjustment. On 1,283 human-gold test segments in seven languages, best measured redaction F1 is 88.8, against 69.1 for published GLiNER2 with 11 unrepresentable types excluded from its task (68.8 without that exemption), 67.8 for GLiNER2 adapted to the new training data, 57.3 for Microsoft Presidio and 35.8 for the best published OpenAI Privacy Filter fine-tune. Adding about 50,000 annotated training sentences and increasing human-gold replay improves exact typed-span F1 from 74.5 to 76.3 on Ont3, our 31-type frontier-annotated NER evaluation of 1,201 development segments. Mapped-gold replay alone raises human-gold F1 by ten points without loss on frontier-annotated text; boundary adjustment adds 1.7 exact typed-span F1 points on Ont3. Local LLMs fitting on a single 96-GB GPU underperformed as prompted annotators and frozen encoders, with encoding 30-95 times slower than XLM-R inference and prompted annotation roughly 180-1,100 times slower in the evaluated configurations. The encoder architecture delivers 4.9 times GLiNER2's CPU throughput. We release code, prompts and training recipes, with data-acquisition scripts and source links.

Retrieval models typically rely on costly human-labeled query-document relevance annotations for training and evaluation. To reduce this cost and leverage the potential of Large Language Models (LLMs) in relevance judgments, we aim to explore whether LLM-generated annotations can effectively replace human annotations in training retrieval models. Retrieval usually emphasizes relevance, which indicates "topic-relatedness" of a document to a query, while in RAG, the value of a document (or utility) depends on how it contributes to answer generation. Recognizing this mismatch, some researchers use LLM performance on downstream tasks with documents as labels, but this approach requires manual answers for specific tasks, leading to high costs and limited generalization. In another line of work, prompting LLMs to select useful documents as RAG references eliminates the need for human annotation and is not task-specific. If we leverage LLMs' utility judgments to annotate retrieval data, we may retain cross-task generalization without human annotation in large-scale corpora. Therefore, we investigate utility-focused annotation via LLMs for large-scale retriever training data across both in-domain and out-of-domain settings on the retrieval and RAG tasks. To reduce the impact of low-quality positives labeled by LLMs, we design a novel loss function, i.e., Disj-InfoNCE. Our experiments reveal that: (1) Retrievers trained on utility-focused annotations significantly outperform those trained on human annotations in the out-of-domain setting on both tasks, demonstrating superior generalization capabilities. (2) LLM annotation does not replace human annotation in the in-domain setting. However, incorporating just 20% human-annotated data enables retrievers trained with utility-focused annotations to match the performance of models trained entirely with human annotations.

Generating unbiased summaries in real-world settings such as political perspective summarization remains a crucial application of Large Language Models (LLMs). Yet, existing evaluation frameworks rely on traditional metrics for measuring key attributes such as coverage and faithfulness without verifying their applicability, and efforts to develop improved summarizers are still nascent. We address these gaps by (1) identifying reliable metrics for measuring perspective summary quality, and (2) investigating the efficacy of LLM-based methods beyond zero-shot inference. Namely, we build a test set for benchmarking metric reliability using human annotations and show that traditional metrics underperform compared to language model-based metrics, which prove to be strong evaluators. Using these metrics, we show that reranking-based methods yield strong results, and preference tuning with synthetically generated and reranking-labeled data further boosts performance. Our findings aim to contribute to the reliable evaluation and development of perspective summarization methods.

We present a large-scale dataset for the task of rewriting an ill-formed natural language question to a well-formed one. Our multi-domain question rewriting MQR dataset is constructed from human contributed Stack Exchange question edit histories. The dataset contains 427,719 question pairs which come from 303 domains. We provide human annotations for a subset of the dataset as a quality estimate. When moving from ill-formed to well-formed questions, the question quality improves by an average of 45 points across three aspects. We train sequence-to-sequence neural models on the constructed dataset and obtain an improvement of 13.2% in BLEU-4 over baseline methods built from other data resources. We release the MQR dataset to encourage research on the problem of question rewriting.

Skill extraction and recommendation systems have been studied from recruiter, applicant, and education perspectives. While AI applications in job advertisements have received broad attention, deficiencies in the instructed skills side remain a challenge. In this work, we address the scarcity of publicly available datasets by releasing both manually annotated and synthetic datasets of skills from the European Skills, Competences, Qualifications and Occupations (ESCO) taxonomy and university course pairs and publishing corresponding annotation guidelines. Specifically, we match graduate-level university courses with skills from the Systems Analysts and Management and Organization Analyst ESCO occupation groups at two granularities: course title with a skill, and course sentence with a skill. We train language models on this dataset to serve as a baseline for retrieval and recommendation systems for course-to-skill and skill-to-course matching. We evaluate the models on a portion of the annotated data. Our BERT model achieves 87% F1-score, showing that course and skill matching is a feasible task.

To compliment the existing set of datasets, we present a small dataset entitled vocadito, consisting of 40 short excerpts of monophonic singing, sung in 7 different languages by singers with varying of levels of training, and recorded on a variety of devices. We provide several types of annotations, including f_0, lyrics, and two different note annotations. All annotations were created by musicians. We provide an analysis of the differences between the two note annotations, and see that the agreement level is low, which has implications for evaluating vocal note estimation algorithms. We also analyze the relation between the f_0 and note annotations, and show that quantizing f_0 values in frequency does not provide a reasonable note estimate, reinforcing the difficulty of the note estimation task for singing voice. Finally, we provide baseline results from recent algorithms on vocadito for note and f_0 transcription. Vocadito is made freely available for public use.

Human annotation is the empirical foundation of much NLP research, from dataset construction to model evaluation, but papers often leave unclear who produced the annotations and how the annotation process was controlled. We provide the first large-scale, task-level audit of human annotation reporting across major NLP venues, asking which annotation details are documented, which are missing, and how reporting varies across time, topic, venue, and intended use of human judgment. We introduce a unified taxonomy of annotation-reporting practices and validate an LLM-assisted extraction pipeline against Annotated-gold, a human-adjudicated gold standard of 41 papers and 72 annotation tasks, where the best model reaches human-comparable agreement with adjudicated labels, with Krippendorff's alpha of 0.606 versus 0.585 for human-human agreement. Using this pipeline, we construct Annotated-llm, a dataset covering ACL-venue papers from 2018-2025, with 2,667 extracted annotation tasks from 1,603 papers, and find that papers frequently report operational details such as recruitment strategies, annotator expertise, and annotation volume, but often omit details needed to assess annotation validity, including training, language proficiency, compensation, socio-demographics, adjudication, and agreement values, especially in model-evaluation studies. Our results show that annotation reporting in NLP has improved over time but remains uneven, and they establish a scalable framework and bare-minimum reporting recommendations for making human annotation more reliable, reproducible, and interpretable.

As Large Language Models (LLMs) and Retrieval Augmentation Generation (RAG) techniques have evolved, query rewriting has been widely incorporated into the RAG system for downstream tasks like open-domain QA. Many works have attempted to utilize small models with reinforcement learning rather than costly LLMs to improve query rewriting. However, current methods require annotations (e.g., labeled relevant documents or downstream answers) or predesigned rewards for feedback, which lack generalization, and fail to utilize signals tailored for query rewriting. In this paper, we propose ours, a framework for training query rewriting models free of annotations. By leveraging a publicly available reranker, ours~provides feedback aligned well with the rewriting objectives. Experimental results demonstrate that ours~can obtain better performance than baselines.

Human evaluation is the foundation upon which the evaluation of both summarization systems and automatic metrics rests. However, existing human evaluation studies for summarization either exhibit a low inter-annotator agreement or have insufficient scale, and an in-depth analysis of human evaluation is lacking. Therefore, we address the shortcomings of existing summarization evaluation along the following axes: (1) We propose a modified summarization salience protocol, Atomic Content Units (ACUs), which is based on fine-grained semantic units and allows for a high inter-annotator agreement. (2) We curate the Robust Summarization Evaluation (RoSE) benchmark, a large human evaluation dataset consisting of 22,000 summary-level annotations over 28 top-performing systems on three datasets. (3) We conduct a comparative study of four human evaluation protocols, underscoring potential confounding factors in evaluation setups. (4) We evaluate 50 automatic metrics and their variants using the collected human annotations across evaluation protocols and demonstrate how our benchmark leads to more statistically stable and significant results. The metrics we benchmarked include recent methods based on large language models (LLMs), GPTScore and G-Eval. Furthermore, our findings have important implications for evaluating LLMs, as we show that LLMs adjusted by human feedback (e.g., GPT-3.5) may overfit unconstrained human evaluation, which is affected by the annotators' prior, input-agnostic preferences, calling for more robust, targeted evaluation methods.

Foundation Models (FMs) have demonstrated unprecedented capabilities including zero-shot learning, high fidelity data synthesis, and out of domain generalization. However, as we show in this paper, FMs still have poor out-of-the-box performance on expert tasks (e.g. retrieval of car manuals technical illustrations from language queries), data for which is either unseen or belonging to a long-tail part of the data distribution of the huge datasets used for FM pre-training. This underlines the necessity to explicitly evaluate and finetune FMs on such expert tasks, arguably ones that appear the most in practical real-world applications. In this paper, we propose a first of its kind FETA benchmark built around the task of teaching FMs to understand technical documentation, via learning to match their graphical illustrations to corresponding language descriptions. Our FETA benchmark focuses on text-to-image and image-to-text retrieval in public car manuals and sales catalogue brochures. FETA is equipped with a procedure for completely automatic annotation extraction (code would be released upon acceptance), allowing easy extension of FETA to more documentation types and application domains in the future. Our automatic annotation leads to an automated performance metric shown to be consistent with metrics computed on human-curated annotations (also released). We provide multiple baselines and analysis of popular FMs on FETA leading to several interesting findings that we believe would be very valuable to the FM community, paving the way towards real-world application of FMs for practical expert tasks currently 'overlooked' by standard benchmarks focusing on common objects.

Lectures are a learning experience for both students and teachers. Students learn from teachers about the subject material, while teachers learn from students about how to refine their instruction. However, online student feedback is unstructured and abundant, making it challenging for teachers to learn and improve. We take a step towards tackling this challenge. First, we contribute a dataset for studying this problem: SIGHT is a large dataset of 288 math lecture transcripts and 15,784 comments collected from the Massachusetts Institute of Technology OpenCourseWare (MIT OCW) YouTube channel. Second, we develop a rubric for categorizing feedback types using qualitative analysis. Qualitative analysis methods are powerful in uncovering domain-specific insights, however they are costly to apply to large data sources. To overcome this challenge, we propose a set of best practices for using large language models (LLMs) to cheaply classify the comments at scale. We observe a striking correlation between the model's and humans' annotation: Categories with consistent human annotations (>0.9 inter-rater reliability, IRR) also display higher human-model agreement (>0.7), while categories with less consistent human annotations (0.7-0.8 IRR) correspondingly demonstrate lower human-model agreement (0.3-0.5). These techniques uncover useful student feedback from thousands of comments, costing around 0.002$ per comment. We conclude by discussing exciting future directions on using online student feedback and improving automated annotation techniques for qualitative research.

Character Error Rate (CER), the standard metric for OCR and vision-language models (VLMs), charges a typo as much as a wrong name. We introduce semCER, which keeps CER's alignment and denominator and charges each grapheme edit a severity from an LLM judge, with no word segmentation. Normalized CER keeps the score of a whole transcription, where it tracks readers as well as semCER or better, and semCER breaks that score down into named edits. The charges track readers' ratings edit by edit in English, Vietnamese, and Chinese and put the edits readers rate most severe near the top of a sorted list. On pairs drawn where CER and semCER disagreed, readers side with semCER more often than with normalized CER in English and Chinese, and equally often in Vietnamese. Across 16 languages, the judge charges a wrong diacritic or tone mark almost nothing in Spanish but nearly a wrong letter's severity in Telugu. Where we can count real words, a general VLM's wrong letter forms one far above chance while a classical pipeline's stays near it. We release the 61 million judged edits, the aggregated reader ratings, and the distilled scorer at https://github.com/nguyenx2t/semcer.

Column type annotation is the task of annotating the columns of a relational table with the semantic type of the values contained in each column. Column type annotation is an important pre-processing step for data search and data integration in the context of data lakes. State-of-the-art column type annotation methods either rely on matching table columns to properties of a knowledge graph or fine-tune pre-trained language models such as BERT for column type annotation. In this work, we take a different approach and explore using ChatGPT for column type annotation. We evaluate different prompt designs in zero- and few-shot settings and experiment with providing task definitions and detailed instructions to the model. We further implement a two-step table annotation pipeline which first determines the class of the entities described in the table and depending on this class asks ChatGPT to annotate columns using only the relevant subset of the overall vocabulary. Using instructions as well as the two-step pipeline, ChatGPT reaches F1 scores of over 85% in zero- and one-shot setups. To reach a similar F1 score a RoBERTa model needs to be fine-tuned with 356 examples. This comparison shows that ChatGPT is able deliver competitive results for the column type annotation task given no or only a minimal amount of task-specific demonstrations.

Understanding the sentiment of a comment from a video or an image is an essential task in many applications. Sentiment analysis of a text can be useful for various decision-making processes. One such application is to analyse the popular sentiments of videos on social media based on viewer comments. However, comments from social media do not follow strict rules of grammar, and they contain mixing of more than one language, often written in non-native scripts. Non-availability of annotated code-mixed data for a low-resourced language like Tamil also adds difficulty to this problem. To overcome this, we created a gold standard Tamil-English code-switched, sentiment-annotated corpus containing 15,744 comment posts from YouTube. In this paper, we describe the process of creating the corpus and assigning polarities. We present inter-annotator agreement and show the results of sentiment analysis trained on this corpus as a benchmark.

This position paper argues that the machine learning community should prioritize early-stage quality assurance in annotation pipelines over the prevailing practice of late-stage validation. Data quality bottlenecks increasingly limit foundation model improvement, yet quality assurance research focuses almost exclusively on validation methods rather than validation timing. When validation occurs, not merely what methods are employed, fundamentally determines both error rates and annotation costs. This temporal neglect is puzzling given the well-established "shift-left" principle from software engineering, where empirical studies demonstrate 4--100x cost multipliers for defects detected in later stages (Boehm, 1981; Shull et al., 2002). Annotation pipelines exhibit analogous dynamics: errors caught before annotation begins cost a fraction of those discovered after review cycles complete. We propose a taxonomy of three QA trigger points, namely pre-annotation (T0), post-annotation (T1), and post-review (T2), that decompose annotation workflows into discrete validation opportunities. A parametric error-propagation model formalizes when timing affects final error rates versus only economics, making timing a measurable design variable rather than a configuration afterthought. A survey of 47 recent papers reveals that only 4% report when validation occurs, a striking gap given timing's demonstrated impact in adjacent fields. Without explicit attention to QA timing, the community risks optimizing validation methods while ignoring the structural variable that may matter most. Acting on this position requires three steps: researchers should report QA timing configurations alongside validation methods; annotation platforms should expose timing as a first-class parameter; and the community should run controlled experiments that measure stage-specific detection rates directly.

Visual captioning benchmarks have become outdated with the emergence of modern multimodal large language models (MLLMs), as the brief ground-truth sentences and traditional metrics fail to assess detailed captions effectively. While recent benchmarks attempt to address this by focusing on keyword extraction or object-centric evaluation, they remain limited to vague-view or object-view analyses and incomplete visual element coverage. In this paper, we introduce CAPability, a comprehensive multi-view benchmark for evaluating visual captioning across 12 dimensions spanning six critical views. We curate nearly 11K human-annotated images and videos with visual element annotations to evaluate the generated captions. CAPability stably assesses both the correctness and thoroughness of captions using F1-score. By converting annotations to QA pairs, we further introduce a heuristic metric, know but cannot tell (KT), indicating a significant performance gap between QA and caption capabilities. Our work provides the first holistic analysis of MLLMs' captioning abilities, as we identify their strengths and weaknesses across various dimensions, guiding future research to enhance specific aspects of capabilities.

We present the first sizeable corpus of Thai speech emotion recognition, THAI-SER, containing 41 hours and 36 minutes (27,854 utterances) from 100 recordings made in different recording environments: Zoom and two studio setups. The recordings contain both scripted and improvised sessions, acted by 200 professional actors (112 females and 88 males, aged 18 to 55) and were directed by professional directors. There are five primary emotions: neutral, angry, happy, sad, and frustrated, assigned to the actors when recording utterances. The utterances are annotated with an emotional category using crowdsourcing. To control the annotation process's quality, we also design an extensive filtering and quality control scheme to ensure that the majority agreement score remains above 0.71. We evaluate our annotated corpus using two metrics: inter-annotator reliability and human recognition accuracy. Inter-annotator reliability score was calculated using Krippendorff's alpha, where our corpus, after filtering, achieved an alpha score of 0.692, higher than a recommendation of 0.667. For human recognition accuracy, our corpus scored up to 0.772 post-filtering. We also provide the results of the model trained on the corpus evaluated on both in-corpus and cross-corpus setups. The corpus is publicly available under a Creative Commons BY-SA 4.0, as well as our codes for the experiments.

We present a novel corpus for subjectivity detection at the sentence level. We develop new annotation guidelines for the task, which are not limited to language-specific cues, and apply them to produce a new corpus in English. The corpus consists of 411 subjective and 638 objective sentences extracted from ongoing coverage of political affairs from online news outlets. This new resource paves the way for the development of models for subjectivity detection in English and across other languages, without relying on language-specific tools like lexicons or machine translation. We evaluate state-of-the-art multilingual transformer-based models on the task, both in mono- and cross-lingual settings, the latter with a similar existing corpus in Italian language. We observe that enriching our corpus with resources in other languages improves the results on the task.

Anthropogenic ecological crisis constitutes a significant challenge that all within the academy must urgently face, including the Natural Language Processing (NLP) community. While recent years have seen increasing work revolving around climate-centric discourse, crucial environmental and ecological topics outside of climate change remain largely unaddressed, despite their prominent importance. Mainstream NLP tasks, such as sentiment analysis, dominate the scene, but there remains an untouched space in the literature involving the analysis of environmental impacts of certain events and practices. To address this gap, this paper presents EcoVerse, an annotated English Twitter dataset of 3,023 tweets spanning a wide spectrum of environmental topics. We propose a three-level annotation scheme designed for Eco-Relevance Classification, Stance Detection, and introducing an original approach for Environmental Impact Analysis. We detail the data collection, filtering, and labeling process that led to the creation of the dataset. Remarkable Inter-Annotator Agreement indicates that the annotation scheme produces consistent annotations of high quality. Subsequent classification experiments using BERT-based models, including ClimateBERT, are presented. These yield encouraging results, while also indicating room for a model specifically tailored for environmental texts. The dataset is made freely available to stimulate further research.

Machine learning models are commonly used to detect toxicity in online conversations. These models are trained on datasets annotated by human raters. We explore how raters' self-described identities impact how they annotate toxicity in online comments. We first define the concept of specialized rater pools: rater pools formed based on raters' self-described identities, rather than at random. We formed three such rater pools for this study--specialized rater pools of raters from the U.S. who identify as African American, LGBTQ, and those who identify as neither. Each of these rater pools annotated the same set of comments, which contains many references to these identity groups. We found that rater identity is a statistically significant factor in how raters will annotate toxicity for identity-related annotations. Using preliminary content analysis, we examined the comments with the most disagreement between rater pools and found nuanced differences in the toxicity annotations. Next, we trained models on the annotations from each of the different rater pools, and compared the scores of these models on comments from several test sets. Finally, we discuss how using raters that self-identify with the subjects of comments can create more inclusive machine learning models, and provide more nuanced ratings than those by random raters.

Many NLP applications require manual data annotations for a variety of tasks, notably to train classifiers or evaluate the performance of unsupervised models. Depending on the size and degree of complexity, the tasks may be conducted by crowd-workers on platforms such as MTurk as well as trained annotators, such as research assistants. Using a sample of 2,382 tweets, we demonstrate that ChatGPT outperforms crowd-workers for several annotation tasks, including relevance, stance, topics, and frames detection. Specifically, the zero-shot accuracy of ChatGPT exceeds that of crowd-workers for four out of five tasks, while ChatGPT's intercoder agreement exceeds that of both crowd-workers and trained annotators for all tasks. Moreover, the per-annotation cost of ChatGPT is less than $0.003 -- about twenty times cheaper than MTurk. These results show the potential of large language models to drastically increase the efficiency of text classification.

In this paper, we introduce a new Czech subjectivity dataset of 10k manually annotated subjective and objective sentences from movie reviews and descriptions. Our prime motivation is to provide a reliable dataset that can be used with the existing English dataset as a benchmark to test the ability of pre-trained multilingual models to transfer knowledge between Czech and English and vice versa. Two annotators annotated the dataset reaching 0.83 of the Cohen's appa inter-annotator agreement. To the best of our knowledge, this is the first subjectivity dataset for the Czech language. We also created an additional dataset that consists of 200k automatically labeled sentences. Both datasets are freely available for research purposes. Furthermore, we fine-tune five pre-trained BERT-like models to set a monolingual baseline for the new dataset and we achieve 93.56% of accuracy. We fine-tune models on the existing English dataset for which we obtained results that are on par with the current state-of-the-art results. Finally, we perform zero-shot cross-lingual subjectivity classification between Czech and English to verify the usability of our dataset as the cross-lingual benchmark. We compare and discuss the cross-lingual and monolingual results and the ability of multilingual models to transfer knowledge between languages.

Evaluating free-text explanations is a multifaceted, subjective, and labor-intensive task. Large language models (LLMs) present an appealing alternative due to their potential for consistency, scalability, and cost-efficiency. In this work, we present ACORN, a new dataset of 3,500 free-text explanations and aspect-wise quality ratings, and use it to gain insights into how LLMs evaluate explanations. We observed that replacing one of the human ratings sometimes maintained, but more often lowered the inter-annotator agreement across different settings and quality aspects, suggesting that their judgments are not always consistent with human raters. We further quantified this difference by comparing the correlation between LLM-generated ratings with majority-voted human ratings across different quality aspects. With the best system, Spearman's rank correlation ranged between 0.53 to 0.95, averaging 0.72 across aspects, indicating moderately high but imperfect alignment. Finally, we considered the alternative of using an LLM as an additional rater when human raters are scarce, and measured the correlation between majority-voted labels with a limited human pool and LLMs as an additional rater, compared to the original gold labels. While GPT-4 improved the outcome when there were only two human raters, in all other observed cases, LLMs were neutral to detrimental when there were three or more human raters. We publicly release the dataset to support future improvements in LLM-in-the-loop evaluation here: https://github.com/a-brassard/ACORN.

Query rewriting is a crucial technique for passage retrieval in open-domain conversational question answering (CQA). It decontexualizes conversational queries into self-contained questions suitable for off-the-shelf retrievers. Existing methods attempt to incorporate retriever's preference during the training of rewriting models. However, these approaches typically rely on extensive annotations such as in-domain rewrites and/or relevant passage labels, limiting the models' generalization and adaptation capabilities. In this paper, we introduce AdaQR (Adaptive Query Rewriting), a framework for training query rewriting models with limited rewrite annotations from seed datasets and completely no passage label. Our approach begins by fine-tuning compact large language models using only ~10% of rewrite annotations from the seed dataset training split. The models are then utilized to generate rewrite candidates for each query instance. A novel approach is then proposed to assess retriever's preference for these candidates by the probability of answers conditioned on the conversational query by marginalizing the Top-K passages. This serves as the reward for optimizing the rewriter further using Direct Preference Optimization (DPO), a process free of rewrite and retrieval annotations. Experimental results on four open-domain CQA datasets demonstrate that AdaQR not only enhances the in-domain capabilities of the rewriter with limited annotation requirement, but also adapts effectively to out-of-domain datasets.

Training learnable metrics using modern language models has recently emerged as a promising method for the automatic evaluation of machine translation. However, existing human evaluation datasets for text simplification have limited annotations that are based on unitary or outdated models, making them unsuitable for this approach. To address these issues, we introduce the SimpEval corpus that contains: SimpEval_past, comprising 12K human ratings on 2.4K simplifications of 24 past systems, and SimpEval_2022, a challenging simplification benchmark consisting of over 1K human ratings of 360 simplifications including GPT-3.5 generated text. Training on SimpEval, we present LENS, a Learnable Evaluation Metric for Text Simplification. Extensive empirical results show that LENS correlates much better with human judgment than existing metrics, paving the way for future progress in the evaluation of text simplification. We also introduce Rank and Rate, a human evaluation framework that rates simplifications from several models in a list-wise manner using an interactive interface, which ensures both consistency and accuracy in the evaluation process and is used to create the SimpEval datasets.

We describe a dataset developed for Named Entity Recognition in German federal court decisions. It consists of approx. 67,000 sentences with over 2 million tokens. The resource contains 54,000 manually annotated entities, mapped to 19 fine-grained semantic classes: person, judge, lawyer, country, city, street, landscape, organization, company, institution, court, brand, law, ordinance, European legal norm, regulation, contract, court decision, and legal literature. The legal documents were, furthermore, automatically annotated with more than 35,000 TimeML-based time expressions. The dataset, which is available under a CC-BY 4.0 license in the CoNNL-2002 format, was developed for training an NER service for German legal documents in the EU project Lynx.

Though pre-training vision-language models have demonstrated significant benefits in boosting video-text retrieval performance from large-scale web videos, fine-tuning still plays a critical role with manually annotated clips with start and end times, which requires considerable human effort. To address this issue, we explore an alternative cheaper source of annotations, single timestamps, for video-text retrieval. We initialise clips from timestamps in a heuristic way to warm up a retrieval model. Then a video clip editing method is proposed to refine the initial rough boundaries to improve retrieval performance. A student-teacher network is introduced for video clip editing. The teacher model is employed to edit the clips in the training set whereas the student model trains on the edited clips. The teacher weights are updated from the student's after the student's performance increases. Our method is model agnostic and applicable to any retrieval models. We conduct experiments based on three state-of-the-art retrieval models, COOT, VideoCLIP and CLIP4Clip. Experiments conducted on three video retrieval datasets, YouCook2, DiDeMo and ActivityNet-Captions show that our edited clips consistently improve retrieval performance over initial clips across all the three retrieval models.

We introduce Biomed-Enriched, a biomedical text dataset constructed from PubMed via a two-stage annotation process. In the first stage, a large language model annotates 400K paragraphs from PubMed scientific articles, assigning scores for their type (review, study, clinical case, other), domain (clinical, biomedical, other), and educational quality. The educational quality score (rated 1 to 5) estimates how useful a paragraph is for college-level learning. These annotations are then used to fine-tune a small language model, which propagates the labels across the full PMC-OA corpus. The resulting metadata allows us to extract refined subsets, including 2M clinical case paragraphs with over 450K high-quality ones from articles with commercial-use licenses, and to construct several variants via quality filtering and domain upsampling. Clinical text is typically difficult to access due to privacy constraints, as hospital records cannot be publicly shared. Hence, our dataset provides an alternative large-scale, openly available collection of clinical cases from PubMed, making it a valuable resource for biomedical and clinical NLP. Preliminary continual-pretraining experiments with OLMo2 suggest these curated subsets enable targeted improvements, with clinical upsampling boosting performance by ~5% on MMLU ProfMed and educational quality filtering improving MedQA and MedMCQA by ~1%. Combinations of these techniques led to faster convergence, reaching same performance with a third of training tokens, indicating potential for more efficient and effective biomedical pretraining strategies.

Multimodal Information Retrieval has made significant progress in recent years, leveraging the increasingly strong multimodal abilities of deep pre-trained models to represent information across modalities. Music Information Retrieval (MIR), in particular, has considerably increased in quality, with neural representations of music even making its way into everyday life products. However, there is a lack of high-quality benchmarks for evaluating music retrieval performance. To address this issue, we introduce IncompeBench, a carefully annotated benchmark comprising 1,574 permissively licensed, high-quality music snippets, 500 diverse queries, and over 125,000 individual relevance judgements. These annotations were created through the use of a multi-stage pipeline, resulting in high agreement between human annotators and the generated data. The resulting datasets are publicly available at https://huggingface.co/datasets/mixedbread-ai/incompebench-strict and https://huggingface.co/datasets/mixedbread-ai/incompebench-lenient with the prompts available at https://github.com/mixedbread-ai/incompebench-programs.

With PRECISE, we extended Prediction-Powered Inference to produce bias-corrected estimates of ranking evaluation metrics by combining a small human-labeled set with a large LLM-judged set. PPI is provably unbiased regardless of the LLM judge's error profile. We make it applicable to hierarchical metrics like Precision@K, where annotations are per-document but the metric is per-query, by reducing the output-space computation from O(2^|C|) to O(2^K). On the ESCI benchmark, augmenting 30 human annotations with Claude 3 Sonnet judgments reduces the standard error of Precision@4 estimates from 4.45 to 3.50 (a 21% relative reduction). In a production system, our framework correctly identified the best of three system variants from 100 human labels and 2 hours of domain-expert annotation; A/B testing confirmed this ranking with +407 bps in daily sales.

We introduce IndiaFinBench, to our knowledge the first publicly available evaluation benchmark for assessing large language model (LLM) performance on Indian financial regulatory text. Existing financial NLP benchmarks draw exclusively from Western financial corpora (SEC filings, US earnings reports, and English-language financial news), leaving a significant gap in coverage of non-Western regulatory frameworks. IndiaFinBench addresses this gap with 406 expert-annotated question-answer pairs drawn from 192 documents sourced from the Securities and Exchange Board of India (SEBI) and the Reserve Bank of India (RBI), spanning four task types: regulatory interpretation (174 items), numerical reasoning (92 items), contradiction detection (62 items), and temporal reasoning (78 items). Annotation quality is validated through a model-based secondary pass (kappa=0.918 on contradiction detection) and a 60-item human inter-annotator agreement evaluation (kappa=0.611; 76.7% overall agreement). We evaluate twelve models under zero-shot conditions, with accuracy ranging from 70.4% (Gemma 4 E4B) to 89.7% (Gemini 2.5 Flash). All models substantially outperform a non-specialist human baseline of 60.0%. Numerical reasoning is the most discriminative task, with a 35.9 percentage-point spread across models. Bootstrap significance testing (10,000 resamples) reveals three statistically distinct performance tiers. The dataset, evaluation code, and all model outputs are available at https://github.com/rajveerpall/IndiaFinBench

MeetEval is an open-source toolkit to evaluate all kinds of meeting transcription systems. It provides a unified interface for the computation of commonly used Word Error Rates (WERs), specifically cpWER, ORC-WER and MIMO-WER along other WER definitions. We extend the cpWER computation by a temporal constraint to ensure that only words are identified as correct when the temporal alignment is plausible. This leads to a better quality of the matching of the hypothesis string to the reference string that more closely resembles the actual transcription quality, and a system is penalized if it provides poor time annotations. Since word-level timing information is often not available, we present a way to approximate exact word-level timings from segment-level timings (e.g., a sentence) and show that the approximation leads to a similar WER as a matching with exact word-level annotations. At the same time, the time constraint leads to a speedup of the matching algorithm, which outweighs the additional overhead caused by processing the time stamps.

This paper introduces the Decomposed Requirements Following Ratio (DRFR), a new metric for evaluating Large Language Models' (LLMs) ability to follow instructions. Addressing a gap in current methodologies, DRFR breaks down complex instructions into simpler criteria, facilitating a detailed analysis of LLMs' compliance with various aspects of tasks. Alongside this metric, we present InFoBench, a benchmark comprising 500 diverse instructions and 2,250 decomposed questions across multiple constraint categories. Our experiments compare DRFR with traditional scoring methods and explore annotation sources, including human experts, crowd-sourced workers, and GPT-4. The findings demonstrate DRFR's higher reliability and the effectiveness of using GPT-4 as a cost-efficient annotator. The evaluation of several advanced LLMs using this framework reveals their strengths and areas needing improvement, particularly in complex instruction-following. This study contributes a novel metric and benchmark, offering insights for future LLM development and evaluation.

Automatic speech recognition replaces typing only when correction costs less than manual entry, a threshold determined by error types, not counts: fixing a misrecognized domain term costs far more than inserting a comma. Word error rate (WER) fails on two fronts: it collapses distinct error categories into a single scalar, and it structurally penalizes agglutinative languages where valid sandhi merges inflate scores. We introduce SCRIBE, a diagnostic framework that provides categorical error decomposition into lexical, punctuation, numeral, and domain-entity rates through sandhi-tolerant alignment with domain vocabulary injection. Human validation confirms SCRIBE aligns with expert judgment where WER does not. We release SCRIBE, an LLM curation pipeline, benchmarks, and open-weight rich transcription models for Hindi, Malayalam, and Kannada.

As Large Language Model (LLM) alignment evolves from simple completions to complex, highly sophisticated generation, Reward Models are increasingly shifting toward rubric-guided evaluation to mitigate surface-level biases. However, the community lacks a unified benchmark to assess this evaluation paradigm, as existing benchmarks lack both the discriminative complexity and the ground-truth rubric annotations required for rigorous analysis. To bridge this gap, we introduce RubricBench, a curated benchmark with 1,147 pairwise comparisons specifically designed to assess the reliability of rubric-based evaluation. Our construction employs a multi-dimensional filtration pipeline to target hard samples featuring nuanced input complexity and misleading surface bias, augmenting each with expert-annotated, atomic rubrics derived strictly from instructions. Comprehensive experiments reveal a substantial capability gap between human-annotated and model-generated rubrics, indicating that even state-of-the-art models struggle to autonomously specify valid evaluation criteria, lagging considerably behind human-guided performance.

Open community-driven platforms like Chatbot Arena that collect user preference data from site visitors have gained a reputation as one of the most trustworthy publicly available benchmarks for LLM performance. While now standard, it is tricky to implement effective guardrails to collect high-quality annotations from humans. In this paper, we demonstrate that three sources of bad annotations, both malicious and otherwise, can corrupt the reliability of open leaderboard rankings. In particular, we show that only 10\% of poor quality votes by apathetic (site visitors not appropriately incentivized to give correct votes) or adversarial (bad actors seeking to inflate the ranking of a target model) annotators can change the rankings of models by up to 5 places on the leaderboard. Finally, we discuss open challenges in ensuring high-quality human annotations.

News headline generation is a crucial task in increasing productivity for both the readers and producers of news. This task can easily be aided by automated News headline-generation models. However, the presence of irrelevant headlines in scraped news articles results in sub-optimal performance of generation models. We propose that relevance-based headline classification can greatly aid the task of generating relevant headlines. Relevance-based headline classification involves categorizing news headlines based on their relevance to the corresponding news articles. While this task is well-established in English, it remains under-explored in low-resource languages like Telugu due to a lack of annotated data. To address this gap, we present TeClass, the first-ever human-annotated Telugu news headline classification dataset, containing 78,534 annotations across 26,178 article-headline pairs. We experiment with various baseline models and provide a comprehensive analysis of their results. We further demonstrate the impact of this work by fine-tuning various headline generation models using TeClass dataset. The headlines generated by the models fine-tuned on highly relevant article-headline pairs, showed about a 5 point increment in the ROUGE-L scores. To encourage future research, the annotated dataset as well as the annotation guidelines will be made publicly available.

Large Language Models (LLMs) are increasingly used for zero-shot annotation and LLM-as-a-judge tasks, yet their reliability hinges on how model-internalized priors interact with user-provided instructions. We investigate three dimensions of this interaction: (1) how an LLM's familiarity with data and task definitions affects performance, (2) the extent to which additional information in prompts can correct zero-shot errors ("decision stickiness"), and (3) model susceptibility to misaligned task definitions. Through experiments on toxicity detection across diverse datasets (spanning social media, gaming, news, and forums) using both dense and mixture-of-experts models, we find that nearly two-thirds of zero-shot errors are resistant to correction, with an overall rescue rate (fraction of initial errors corrected by prompting) of only 34.8%. High-confidence errors prove especially resistant to correction. When given misaligned definitions, LLMs follow them while maintaining confidence levels unchanged from the aligned condition. Crucially, we introduce Definition-Specific Familiarity (DSF), which measures alignment between a model's internal concept and the task definition. After controlling for dataset-level confounds, DSF shows a positive association with model performance (partial r = +0.41), while three distinct memorization metrics (ROUGE-L, BERTScore, and embedding cosine similarity) all fail to show a positive association. These findings show the limitations of prompt-based correction in annotation tasks, highlighting the importance of definition alignment over text-level memorization.

Traditional automatic speech recognition (ASR) models output lower-cased words without punctuation marks, which reduces readability and necessitates a subsequent text processing model to convert ASR transcripts into a proper format. Simultaneously, the development of end-to-end ASR models capable of predicting punctuation and capitalization presents several challenges, primarily due to limited data availability and shortcomings in the existing evaluation methods, such as inadequate assessment of punctuation prediction. In this paper, we introduce a LibriSpeech-PC benchmark designed to assess the punctuation and capitalization prediction capabilities of end-to-end ASR models. The benchmark includes a LibriSpeech-PC dataset with restored punctuation and capitalization, a novel evaluation metric called Punctuation Error Rate (PER) that focuses on punctuation marks, and initial baseline models. All code, data, and models are publicly available.

Detecting subjectivity in news sentences is crucial for identifying media bias, enhancing credibility, and combating misinformation by flagging opinion-based content. It provides insights into public sentiment, empowers readers to make informed decisions, and encourages critical thinking. While research has developed methods and systems for this purpose, most efforts have focused on English and other high-resourced languages. In this study, we present the first large dataset for subjectivity detection in Arabic, consisting of ~3.6K manually annotated sentences, and GPT-4o based explanation. In addition, we included instructions (both in English and Arabic) to facilitate LLM based fine-tuning. We provide an in-depth analysis of the dataset, annotation process, and extensive benchmark results, including PLMs and LLMs. Our analysis of the annotation process highlights that annotators were strongly influenced by their political, cultural, and religious backgrounds, especially at the beginning of the annotation process. The experimental results suggest that LLMs with in-context learning provide better performance. We aim to release the dataset and resources for the community.

Certain forms of linguistic annotation, like part of speech and semantic tagging, can be automated with high accuracy. However, manual annotation is still necessary for complex pragmatic and discursive features that lack a direct mapping to lexical forms. This manual process is time-consuming and error-prone, limiting the scalability of function-to-form approaches in corpus linguistics. To address this, our study explores automating pragma-discursive corpus annotation using large language models (LLMs). We compare ChatGPT, the Bing chatbot, and a human coder in annotating apology components in English based on the local grammar framework. We find that the Bing chatbot outperformed ChatGPT, with accuracy approaching that of a human coder. These results suggest that AI can be successfully deployed to aid pragma-discursive corpus annotation, making the process more efficient and scalable. Keywords: linguistic annotation, function-to-form approaches, large language models, local grammar analysis, Bing chatbot, ChatGPT

High annotation costs from hiring or crowdsourcing complicate the creation of large, high-quality datasets needed for training reliable text classifiers. Recent research suggests using Large Language Models (LLMs) to automate the annotation process, reducing these costs while maintaining data quality. LLMs have shown promising results in annotating downstream tasks like hate speech detection and political framing. Building on the success in these areas, this study investigates whether LLMs are viable for annotating the complex task of media bias detection and whether a downstream media bias classifier can be trained on such data. We create annolexical, the first large-scale dataset for media bias classification with over 48000 synthetically annotated examples. Our classifier, fine-tuned on this dataset, surpasses all of the annotator LLMs by 5-9 percent in Matthews Correlation Coefficient (MCC) and performs close to or outperforms the model trained on human-labeled data when evaluated on two media bias benchmark datasets (BABE and BASIL). This study demonstrates how our approach significantly reduces the cost of dataset creation in the media bias domain and, by extension, the development of classifiers, while our subsequent behavioral stress-testing reveals some of its current limitations and trade-offs.

Modern automatic speech recognition (ASR) systems have achieved superhuman Word Error Rate (WER) on many common corpora despite lacking adequate performance on speech in the wild. Beyond that, there is a lack of real-world, accented corpora to properly benchmark academic and commercial models. To ensure this type of speech is represented in ASR benchmarking, we present Earnings-22, a 125 file, 119 hour corpus of English-language earnings calls gathered from global companies. We run a comparison across 4 commercial models showing the variation in performance when taking country of origin into consideration. Looking at hypothesis transcriptions, we explore errors common to all ASR systems tested. By examining Individual Word Error Rate (IWER), we find that key speech features impact model performance more for certain accents than others. Earnings-22 provides a free-to-use benchmark of real-world, accented audio to bridge academic and industrial research.

Expression of sentiment in parliamentary debates is deemed to be significantly different from that on social media or in product reviews. This paper adds to an emerging body of research on parliamentary debates with a dataset of sentences annotated for detection sentiment polarity in political discourse. We sample the sentences for annotation from the proceedings of three Southeast European parliaments: Croatia, Bosnia-Herzegovina, and Serbia. A six-level schema is applied to the data with the aim of training a classification model for the detection of sentiment in parliamentary proceedings. Krippendorff's alpha measuring the inter-annotator agreement ranges from 0.6 for the six-level annotation schema to 0.75 for the three-level schema and 0.83 for the two-level schema. Our initial experiments on the dataset show that transformer models perform significantly better than those using a simpler architecture. Furthermore, regardless of the similarity of the three languages, we observe differences in performance across different languages. Performing parliament-specific training and evaluation shows that the main reason for the differing performance between parliaments seems to be the different complexity of the automatic classification task, which is not observable in annotator performance. Language distance does not seem to play any role neither in annotator nor in automatic classification performance. We release the dataset and the best-performing model under permissive licences.

Central banks around the world play a crucial role in maintaining economic stability. Deciphering policy implications in their communications is essential, especially as misinterpretations can disproportionately impact vulnerable populations. To address this, we introduce the World Central Banks (WCB) dataset, the most comprehensive monetary policy corpus to date, comprising over 380k sentences from 25 central banks across diverse geographic regions, spanning 28 years of historical data. After uniformly sampling 1k sentences per bank (25k total) across all available years, we annotate and review each sentence using dual annotators, disagreement resolutions, and secondary expert reviews. We define three tasks: Stance Detection, Temporal Classification, and Uncertainty Estimation, with each sentence annotated for all three. We benchmark seven Pretrained Language Models (PLMs) and nine Large Language Models (LLMs) (Zero-Shot, Few-Shot, and with annotation guide) on these tasks, running 15,075 benchmarking experiments. We find that a model trained on aggregated data across banks significantly surpasses a model trained on an individual bank's data, confirming the principle "the whole is greater than the sum of its parts." Additionally, rigorous human evaluations, error analyses, and predictive tasks validate our framework's economic utility. Our artifacts are accessible through the HuggingFace and GitHub under the CC-BY-NC-SA 4.0 license.

Recent work has demonstrated that pre-training in-domain language models can boost performance when adapting to a new domain. However, the costs associated with pre-training raise an important question: given a fixed budget, what steps should an NLP practitioner take to maximize performance? In this paper, we view domain adaptation with a constrained budget as a consumer choice problem, where the goal is to select an optimal combination of data annotation and pre-training. We measure annotation costs of three procedural text datasets, along with the pre-training costs of several in-domain language models. The utility of different combinations of pre-training and data annotation are evaluated under varying budget constraints to assess which combination strategy works best. We find that for small budgets, spending all funds on annotation leads to the best performance; once the budget becomes large enough, however, a combination of data annotation and in-domain pre-training yields better performance. Our experiments suggest task-specific data annotation should be part of an economical strategy when adapting an NLP model to a new domain.

Citations in scientific papers not only help us trace the intellectual lineage but also are a useful indicator of the scientific significance of the work. Citation intents prove beneficial as they specify the role of the citation in a given context. In this paper, we present CitePrompt, a framework which uses the hitherto unexplored approach of prompt-based learning for citation intent classification. We argue that with the proper choice of the pretrained language model, the prompt template, and the prompt verbalizer, we can not only get results that are better than or comparable to those obtained with the state-of-the-art methods but also do it with much less exterior information about the scientific document. We report state-of-the-art results on the ACL-ARC dataset, and also show significant improvement on the SciCite dataset over all baseline models except one. As suitably large labelled datasets for citation intent classification can be quite hard to find, in a first, we propose the conversion of this task to the few-shot and zero-shot settings. For the ACL-ARC dataset, we report a 53.86% F1 score for the zero-shot setting, which improves to 63.61% and 66.99% for the 5-shot and 10-shot settings, respectively.

Language models that are sensitive to external context can more effectively capture the speaking patterns of individuals with specific characteristics or in particular environments. However, obtaining and leveraging such annotations can be challenging. In this work, we show how to leverage rich character and film annotations to personalise language models in a scalable manner. Our best model can reduce perplexity by up to 6.5% compared to a parameter-matched language model. Our approach performs on par with speaker-specific fine-tuning when the fine-tuning data (i.e. past dialogue) for individual speakers is available. On top of that, it also generalises well to a scenario with no such data, relying on combinations of demographic characteristics expressed via metadata. Our findings are consistent across two corpora, one of which is also a contribution of this paper: Cornell-rich contains rich manual annotations for 863 speaking characters from the Cornell Movie Dialog Corpus, including features such as characteristic quotes and character descriptions, along with six automatically extracted metadata features for over 95% of the featured films. Finally, we also present a cost-benefit analysis highlighting which annotations are most cost-effective in reducing perplexity.

Off-the-shelf large language models (LLMs) are increasingly used to automate text annotation, yet their effectiveness remains underexplored for underrepresented languages and specialized domains where the class definition requires subtle expert understanding. We investigate LLM-based annotation for a novel legal NLP task: identifying the presence and sentiment of credibility assessments in asylum decision texts. We introduce RAB-Cred, a Danish text classification dataset featuring high-quality, expert annotations and valuable metadata such as annotator confidence and asylum case outcome. We benchmark 21 open-weight models and 30 system-user prompt combinations for this task, and systematically evaluate the effect of model and prompt choice for zero-shot and few-shot classification. We zoom in on the errors made by top-performing models and prompts, investigating error consistency across LLMs, inter-class confusion, correlation with human confidence and sample-wise difficulty and severity of LLM mistakes. Our results confirm the potential of LLMs for cost-effective labeling of asylum decisions, but highlight the imperfect and inconsistent nature of LLM annotators, and the need to look beyond the predictions of a single, arbitrarily chosen model. The RAB-Cred dataset and code are available at https://github.com/glhr/RAB-Cred

Disagreement in annotation is a common phenomenon in the development of NLP datasets and serves as a valuable source of insight. While majority voting remains the dominant strategy for aggregating labels, recent work has explored modeling individual annotators to preserve their perspectives. However, modeling each annotator is resource-intensive and remains underexplored across various NLP tasks. We propose an agreement-based clustering technique to model the disagreement between the annotators. We conduct comprehensive experiments in 40 datasets in 18 typologically diverse languages, covering three subjective NLP tasks: sentiment analysis, emotion classification, and hate speech detection. We evaluate four aggregation approaches: majority vote, ensemble, multi-label, and multitask. The results demonstrate that agreement-based clustering can leverage the full spectrum of annotator perspectives and significantly enhance classification performance in subjective NLP tasks compared to majority voting and individual annotator modeling. Regarding the aggregation approach, the multi-label and multitask approaches are better for modeling clustered annotators than an ensemble and model majority vote.

Finance reporting is a natural proving ground for large language models, and the very-long-context capabilities of recent models across all sizes make rigorous evaluation in this domain an increasingly pressing need. Yet most public financial resources reduce the task to plain-text SEC 10-K filings paired with a handful of question-answer items. We release LEDGER (Long-context Evaluation of Documents for Grounded Extraction and Retrieval), a corpus of 4,999 digitized corporate annual reports - full documents with figures, tables, and narrative, not just regulatory filings. Each report is labeled with 31 consolidated financial KPIs to be extracted and linked to the market's reaction at the earnings date. From this data we derive three evaluation benchmarks spanning the difficulty spectrum: a pure page-level KPI retrieval task with TREC-style relevance judgments over 118,048 questions in natural language, a conversational "needle-in-a-haystack" single-value lookup, and a full KPI extraction task, both from long, numerically dense reports. We additionally provide human OCR-quality annotations with inter-annotator agreement and the complete extraction, validation, and scoring toolchain. We further demonstrate the dataset's research utility with a case study linking CEO-letter rhetoric to post-publication market impact.

Aspect sentiment quadruple prediction (ASQP) facilitates a detailed understanding of opinions expressed in a text by identifying the opinion term, aspect term, aspect category and sentiment polarity for each opinion. However, annotating a full set of training examples to fine-tune models for ASQP is a resource-intensive process. In this study, we explore the capabilities of large language models (LLMs) for zero- and few-shot learning on the ASQP task across five diverse datasets. We report F1 scores slightly below those obtained with state-of-the-art fine-tuned models but exceeding previously reported zero- and few-shot performance. In the 40-shot setting on the Rest16 restaurant domain dataset, LLMs achieved an F1 score of 52.46, compared to 60.39 by the best-performing fine-tuned method MVP. Additionally, we report the performance of LLMs in target aspect sentiment detection (TASD), where the F1 scores were also close to fine-tuned models, achieving 66.03 on Rest16 in the 40-shot setting, compared to 72.76 with MVP. While human annotators remain essential for achieving optimal performance, LLMs can reduce the need for extensive manual annotation in ASQP tasks.

Today, most large-scale conversational AI agents (e.g. Alexa, Siri, or Google Assistant) are built using manually annotated data to train the different components of the system. Typically, the accuracy of the ML models in these components are improved by manually transcribing and annotating data. As the scope of these systems increase to cover more scenarios and domains, manual annotation to improve the accuracy of these components becomes prohibitively costly and time consuming. In this paper, we propose a system that leverages user-system interaction feedback signals to automate learning without any manual annotation. Users here tend to modify a previous query in hopes of fixing an error in the previous turn to get the right results. These reformulations, which are often preceded by defective experiences caused by errors in ASR, NLU, ER or the application. In some cases, users may not properly formulate their requests (e.g. providing partial title of a song), but gleaning across a wider pool of users and sessions reveals the underlying recurrent patterns. Our proposed self-learning system automatically detects the errors, generate reformulations and deploys fixes to the runtime system to correct different types of errors occurring in different components of the system. In particular, we propose leveraging an absorbing Markov Chain model as a collaborative filtering mechanism in a novel attempt to mine these patterns. We show that our approach is highly scalable, and able to learn reformulations that reduce Alexa-user errors by pooling anonymized data across millions of customers. The proposed self-learning system achieves a win/loss ratio of 11.8 and effectively reduces the defect rate by more than 30% on utterance level reformulations in our production A/B tests. To the best of our knowledge, this is the first self-learning large-scale conversational AI system in production.

We present an overview of the FIGNEWS shared task, organized as part of the ArabicNLP 2024 conference co-located with ACL 2024. The shared task addresses bias and propaganda annotation in multilingual news posts. We focus on the early days of the Israel War on Gaza as a case study. The task aims to foster collaboration in developing annotation guidelines for subjective tasks by creating frameworks for analyzing diverse narratives highlighting potential bias and propaganda. In a spirit of fostering and encouraging diversity, we address the problem from a multilingual perspective, namely within five languages: English, French, Arabic, Hebrew, and Hindi. A total of 17 teams participated in two annotation subtasks: bias (16 teams) and propaganda (6 teams). The teams competed in four evaluation tracks: guidelines development, annotation quality, annotation quantity, and consistency. Collectively, the teams produced 129,800 data points. Key findings and implications for the field are discussed.

Moral sentiments expressed in natural language significantly influence both online and offline environments, shaping behavioral styles and interaction patterns, including social media selfpresentation, cyberbullying, adherence to social norms, and ethical decision-making. To effectively measure moral sentiments in natural language processing texts, it is crucial to utilize large, annotated datasets that provide nuanced understanding for accurate analysis and modeltraining. However, existing corpora, while valuable, often face linguistic limitations. To address this gap in the Chinese language domain,we introduce the Moral Foundation Weibo Corpus. This corpus consists of 25,671 Chinese comments on Weibo, encompassing six diverse topic areas. Each comment is manually annotated by at least three systematically trained annotators based on ten moral categories derived from a grounded theory of morality. To assess annotator reliability, we present the kappa testresults, a gold standard for measuring consistency. Additionally, we apply several the latest large language models to supplement the manual annotations, conducting analytical experiments to compare their performance and report baseline results for moral sentiment classification.

As the NLP community increasingly addresses challenges associated with multilingualism, robust annotation tools are essential to handle multilingual datasets efficiently. In this paper, we introduce a code-mixed multilingual text annotation framework, COMMENTATOR, specifically designed for annotating code-mixed text. The tool demonstrates its effectiveness in token-level and sentence-level language annotation tasks for Hinglish text. We perform robust qualitative human-based evaluations to showcase COMMENTATOR led to 5x faster annotations than the best baseline. Our code is publicly available at https://github.com/lingo-iitgn/commentator. The demonstration video is available at https://bit.ly/commentator_video.

Traditional Information Retrieval (IR) metrics, such as nDCG, MAP, and MRR, assume that human users sequentially examine documents with diminishing attention to lower ranks. This assumption breaks down in Retrieval Augmented Generation (RAG) systems, where search results are consumed by Large Language Models (LLMs), which, unlike humans, process all retrieved documents as a whole rather than sequentially. Additionally, traditional IR metrics do not account for related but irrelevant documents that actively degrade generation quality, rather than merely being ignored. Due to these two major misalignments, namely human vs. machine position discount and human relevance vs. machine utility, classical IR metrics do not accurately predict RAG performance. We introduce a utility-based annotation schema that quantifies both the positive contribution of relevant passages and the negative impact of distracting ones. Building on this foundation, we propose UDCG (Utility and Distraction-aware Cumulative Gain), a metric using an LLM-oriented positional discount to directly optimize the correlation with the end-to-end answer accuracy. Experiments on five datasets and six LLMs demonstrate that UDCG improves correlation by up to 36% compared to traditional metrics. Our work provides a critical step toward aligning IR evaluation with LLM consumers and enables more reliable assessment of RAG components

Human cognition is deeply intertwined with a sense of time, known as Chronoception. This sense allows us to judge how long facts remain valid and when knowledge becomes outdated. Despite progress in vision, language, and motor control, AI still struggles to reason about temporal validity. We introduce Chronocept, the first benchmark to model temporal validity as a continuous probability distribution over time. Using skew-normal curves fitted along semantically decomposed temporal axes, Chronocept captures nuanced patterns of emergence, decay, and peak relevance. It includes two datasets: Benchmark I (atomic facts) and Benchmark II (multi-sentence passages). Annotations show strong inter-annotator agreement (84% and 89%). Our baselines predict curve parameters - location, scale, and skewness - enabling interpretable, generalizable learning and outperforming classification-based approaches. Chronocept fills a foundational gap in AI's temporal reasoning, supporting applications in knowledge grounding, fact-checking, retrieval-augmented generation (RAG), and proactive agents. Code and data are publicly available.

Knowledge-intensive language tasks require NLP systems to both provide the correct answer and retrieve supporting evidence for it in a given corpus. Autoregressive language models are emerging as the de-facto standard for generating answers, with newer and more powerful systems emerging at an astonishing pace. In this paper we argue that all this (and future) progress can be directly applied to the retrieval problem with minimal intervention to the models' architecture. Previous work has explored ways to partition the search space into hierarchical structures and retrieve documents by autoregressively generating their unique identifier. In this work we propose an alternative that doesn't force any structure in the search space: using all ngrams in a passage as its possible identifiers. This setup allows us to use an autoregressive model to generate and score distinctive ngrams, that are then mapped to full passages through an efficient data structure. Empirically, we show this not only outperforms prior autoregressive approaches but also leads to an average improvement of at least 10 points over more established retrieval solutions for passage-level retrieval on the KILT benchmark, establishing new state-of-the-art downstream performance on some datasets, while using a considerably lighter memory footprint than competing systems. Code and pre-trained models at https://github.com/facebookresearch/SEAL.

With the massive surge in ML models on platforms like Hugging Face, users often lose track and struggle to choose the best model for their downstream tasks, frequently relying on model popularity indicated by download counts, likes, or recency. We investigate whether this popularity aligns with actual model performance and how the comprehensiveness of model documentation correlates with both popularity and performance. In our study, we evaluated a comprehensive set of 500 Sentiment Analysis models on Hugging Face. This evaluation involved massive annotation efforts, with human annotators completing nearly 80,000 annotations, alongside extensive model training and evaluation. Our findings reveal that model popularity does not necessarily correlate with performance. Additionally, we identify critical inconsistencies in model card reporting: approximately 80\% of the models analyzed lack detailed information about the model, training, and evaluation processes. Furthermore, about 88\% of model authors overstate their models' performance in the model cards. Based on our findings, we provide a checklist of guidelines for users to choose good models for downstream tasks.

LLM judges are used to reduce the need for costly human labor in evaluating open-ended text generation. However, the reliability of these judges depends critically on their alignment with human raters -- a property that itself depends on costly human annotations. In this work, we develop a method (Metric Match) for estimating correlation-based reliability metrics of LLM judges from limited annotations. Metric Match selects a subset of samples for human annotation such that the subset matches the population reliability metric with respect to acquired synthetic labels. We empirically show that Metric Match achieves a win-rate of 0.838 against random subset selection across four different correlation metrics and 15 datasets, with an 18.7% decrease in average estimation error and reduces annotation needs by 32.5%. We provide a cost model and highlight a medical case study where our method saves $1,041.67 compared to random selection for expert annotation. Further, we shift our task from reliability estimation to reliability classification of whether a given judge is above a deployment threshold, outperforming random selection with Metric Match. All project code is publicly available, and we additionally provide an installable package for ease of use.

Aligning large language models (LLMs) with human values and intents critically involves the use of human or AI feedback. While dense feedback annotations are expensive to acquire and integrate, sparse feedback presents a structural design choice between ratings (e.g., score Response A on a scale of 1-7) and rankings (e.g., is Response A better than Response B?). In this work, we analyze the effect of this design choice for the alignment and evaluation of LLMs. We uncover an inconsistency problem wherein the preferences inferred from ratings and rankings significantly disagree 60% for both human and AI annotators. Our subsequent analysis identifies various facets of annotator biases that explain this phenomena, such as human annotators would rate denser responses higher while preferring accuracy during pairwise judgments. To our surprise, we also observe that the choice of feedback protocol also has a significant effect on the evaluation of aligned LLMs. In particular, we find that LLMs that leverage rankings data for alignment (say model X) are preferred over those that leverage ratings data (say model Y), with a rank-based evaluation protocol (is X/Y's response better than reference response?) but not with a rating-based evaluation protocol (score Rank X/Y's response on a scale of 1-7). Our findings thus shed light on critical gaps in methods for evaluating the real-world utility of language models and their strong dependence on the feedback protocol used for alignment. Our code and data are available at https://github.com/Hritikbansal/sparse_feedback.