Neural retrievers are trained to estimate query-document relevance from annotated query-document pairs. Yet annotation protocols may not purely reflect relevance: they select only a subset of documents for labeling, and this selection can favor certain document types over others. We investigate whether supervised bi-encoder retrievers implicitly learn a document-level relevance prior: a query-independent signal encoded in their representation space as a side effect of training on annotated data. We estimate this prior by training simple classifiers on frozen document embeddings and evaluate three state-of-the-art retrievers across multiple IR benchmarks. We find that supervised neural retrievers encode relevance priors that generalize to unseen documents and are consistent across models. These priors are associated with a findability gap: documents with lower prior are systematically harder to retrieve, even when genuinely relevant. This effect appears in supervised dense retrievers but is weaker and less consistent in BM25, and it persists under controlled matched-document comparisons. Using LLM-based explanations, we find that judged-relevant documents tend to be comprehensive, self-contained summaries of mainstream topics, while niche, fragmentary, or highly technical content is often left unjudged. Retrievers appear to internalize this bias, ranking documents with these favored features higher than documents that lack them, independently of their actual relevance. Our findings expose a structural limitation of supervised retrieval: models trained on annotated data do not just learn relevance, but also the implicit document preferences in their training data.

This paper introduces OARelatedWork: a dataset for related work generation from open-access sources. It is the first large-scale multi-document summarization dataset for related work generation, containing whole related work sections and full texts of cited papers. Its validation and test splits are constructed so that every source paper is available in full text, enabling controlled evaluation of full-text related work generation. The dataset includes 94 450 papers and 5 824 689 unique referenced papers from multiple domains. With OARelatedWork, we aim to shift the field from generating parts of related work sections from abstracts only to generating entire related work sections from all available content. We (i) benchmark a wide spectrum of models, highlighting that synthesizing massive full-text contexts remains a challenge for tested cost-effective Large Language Models (LLMs): under our statement-level judge, GPT-4o-mini’s evidence-grounded True rate drops from 92.9% with abstracts to 83.8% with full texts. We (ii) empirically analyze human writing behavior through a human evaluation over 40 papers and 408 factual statements, revealing that authors frequently introduce abstractive claims ungrounded in localized source texts; consequently, tested LLMs actually surpass human baselines in strict, evidence-grounded factuality. Finally, we (iii) conduct a fine-grained meta-evaluation, revealing that standard reference-based metrics are inadequate for evaluating such long-form structured outputs, and introduce a robust statement-level evaluation framework to address this gap.

While Transformer models can learn algorithmic tasks and generalize reliably to unseen, in-distribution data, they often fail catastrophically when required to extrapolate to sequence lengths beyond their training regime. Using AttentionSpan, a suite of synthetic diagnostic tasks designed to enable inspection of internal model computations, we analyze the model’s behavior on length extrapolation. Our findings indicate that this failure is not fully explained by an inability to induce general rules for the task. Instead, we identify inconsistent attention patterns as a major contributing factor: information retrieval strategies learned by individual attention heads, which fail to remain stable as sequence length increases. Reinforcing a reference attention pattern at inference time, using no out-of-distribution data, largely recovers accuracy on String Reversal and partially on Value Assignment, while leaving multi-digit arithmetic unresolved. We also study length adaptation, how a model trained on one length adjusts to another by fine-tuning its Key and Query parameters, and find that a single dimension in every attention head is enough, across 4 algorithmic tasks and the RULER benchmark. Adaptation does not carry past the fine-tuned lengths, and our interventions are diagnostic rather than a proposed remedy. We make our benchmark and code publicly available.

We present BenCzechMark (BCM), the first comprehensive Czech language benchmark designed for large language models, offering diverse tasks, multiple task formats, and multiple evaluation metrics. Its duel scoring system is grounded in statistical significance theory and uses aggregation across tasks inspired by social preference theory. Our benchmark encompasses 50 challenging tasks, with corresponding test datasets, primarily in native Czech, with 14 newly collected ones. These tasks span 8 categories and cover diverse domains, including historical Czech news, essays from pupils or language learners, and spoken word. Furthermore, we collect and clean BUT-Large Czech Collection, the largest publicly available clean Czech language corpus, and use it for (i) contamination analysis and (ii) continuous pretraining of the first Czech-centric 7B language model with Czech-specific tokenization. We use our model as a baseline for comparison with publicly available multilingual models. Lastly, we release and maintain a leaderboard with existing 50 model submissions, where new model submissions can be made at https://huggingface.co/spaces/CZLC/BenCzechMark.

We present Claim-Dissector: a novel latent variable model for fact-checking and analysis, which given a claim and a set of retrieved evidence jointly learns to identify: (i) the relevant evidences to the given claim (ii) the veracity of the claim. We propose to disentangle the per-evidence relevance probability and its contribution to the final veracity probability in an interpretable way — the final veracity probability is proportional to a linear ensemble of per-evidence relevance probabilities. In this way, the individual contributions of evidences towards the final predicted probability can be identified. In per-evidence relevance probability, our model can further distinguish whether each relevant evidence is supporting (S) or refuting (R) the claim. This allows to quantify how much the S/R probability contributes to final verdict or to detect disagreeing evidence. Despite its interpretable nature, our system achieves results competetive with state-of-the-art on the FEVER dataset, as compared to typical two-stage system pipelines, while using significantly fewer parameters. Furthermore, our analysis shows that our model can learn fine-grained relevance cues while using coarse-grained supervision and we demonstrate it in 2 ways. (i) We show that our model can achieve competitive sentence recall while using only paragraph-level relevance supervision. (ii) Traversing towards the finest granularity of relevance, we show that our model is capable of identifying relevance at the token level. To do this, we present a new benchmark TLR-FEVER focusing on token-level interpretability — humans annotate tokens in relevant evidences they considered essential when making their judgment. Then we measure how similar are these annotations to the tokens our model is focusing on.

In this paper, we describe our participation in the subtask 1 of CASE-2022, Event Causality Identification with Casual News Corpus. We address the Causal Relation Identification (CRI) task by exploiting a set of simple yet complementary techniques for fine-tuning language models (LMs) on a few annotated examples (i.e., a few-shot configuration).We follow a prompt-based prediction approach for fine-tuning LMs in which the CRI task is treated as a masked language modeling problem (MLM). This approach allows LMs natively pre-trained on MLM tasks to directly generate textual responses to CRI-specific prompts. We compare the performance of this method against ensemble techniques trained on the entire dataset. Our best-performing submission was fine-tuned with only 256 instances per class, 15.7% of the all available data, and yet obtained the second-best precision (0.82), third-best accuracy (0.82), and an F1-score (0.85) very close to what was reported by the winner team (0.86).

In this paper, we describe our shared task submissions for Subtask 2 in CASE-2022, Event Causality Identification with Casual News Corpus. The challenge focused on the automatic detection of all cause-effect-signal spans present in the sentence from news-media. We detect cause-effect-signal spans in a sentence using T5 — a pre-trained autoregressive language model. We iteratively identify all cause-effect-signal span triplets, always conditioning the prediction of the next triplet on the previously predicted ones. To predict the triplet itself, we consider different causal relationships such as cause→effect→signal. Each triplet component is generated via a language model conditioned on the sentence, the previous parts of the current triplet, and previously predicted triplets. Despite training on an extremely small dataset of 160 samples, our approach achieved competitive performance, being placed second in the competition. Furthermore, we show that assuming either cause→effect or effect→cause order achieves similar results.

This work demonstrates that using the objective with independence assumption for modelling the span probability P (a_s , a_e ) = P (a_s )P (a_e) of span starting at position a_s and ending at position a_e has adverse effects. Therefore we propose multiple approaches to modelling joint probability P (a_s , a_e) directly. Among those, we propose a compound objective, composed from the joint probability while still keeping the objective with independence assumption as an auxiliary objective. We find that the compound objective is consistently superior or equal to other assumptions in exact match. Additionally, we identified common errors caused by the assumption of independence and manually checked the counterpart predictions, demonstrating the impact of the compound objective on the real examples. Our findings are supported via experiments with three extractive QA models (BIDAF, BERT, ALBERT) over six datasets and our code, individual results and manual analysis are available online.

This work presents a novel four-stage open-domain QA pipeline R2-D2 (Rank twice, reaD twice). The pipeline is composed of a retriever, passage reranker, extractive reader, generative reader and a mechanism that aggregates the final prediction from all system’s components. We demonstrate its strength across three open-domain QA datasets: NaturalQuestions, TriviaQA and EfficientQA, surpassing state-of-the-art on the first two. Our analysis demonstrates that: (i) combining extractive and generative reader yields absolute improvements up to 5 exact match and it is at least twice as effective as the posterior averaging ensemble of the same models with different parameters, (ii) the extractive reader with fewer parameters can match the performance of the generative reader on extractive QA datasets.

This paper describes BUT-FIT’s submission at SemEval-2020 Task 5: Modelling Causal Reasoning in Language: Detecting Counterfactuals. The challenge focused on detecting whether a given statement contains a counterfactual (Subtask 1) and extracting both antecedent and consequent parts of the counterfactual from the text (Subtask 2). We experimented with various state-of-the-art language representation models (LRMs). We found RoBERTa LRM to perform the best in both subtasks. We achieved the first place in both exact match and F1 for Subtask 2 and ranked second for Subtask 1.

We participated in all three subtasks. In subtasks A and B, our submissions are based on pretrained language representation models (namely ALBERT) and data augmentation. We experimented with solving the task for another language, Czech, by means of multilingual models and machine translated dataset, or translated model inputs. We show that with a strong machine translation system, our system can be used in another language with a small accuracy loss. In subtask C, our submission, which is based on pretrained sequence-to-sequence model (BART), ranked 1st in BLEU score ranking, however, we show that the correlation between BLEU and human evaluation, in which our submission ended up 4th, is low. We analyse the metrics used in the evaluation and we propose an additional score based on model from subtask B, which correlates well with our manual ranking, as well as reranking method based on the same principle. We performed an error and dataset analysis for all subtasks and we present our findings.

This paper describes our system that was designed for Humor evaluation within the SemEval-2020 Task 7. The system is based on convolutional neural network architecture. We investigate the system on the official dataset, and we provide more insight to model itself to see how the learned inner features look.

This paper describes our system submitted to SemEval 2019 Task 7: RumourEval 2019: Determining Rumour Veracity and Support for Rumours, Subtask A (Gorrell et al., 2019). The challenge focused on classifying whether posts from Twitter and Reddit support, deny, query, or comment a hidden rumour, truthfulness of which is the topic of an underlying discussion thread. We formulate the problem as a stance classification, determining the rumour stance of a post with respect to the previous thread post and the source thread post. The recent BERT architecture was employed to build an end-to-end system which has reached the F1 score of 61.67 % on the provided test data. Without any hand-crafted feature, the system finished at the 2nd place in the competition, only 0.2 % behind the winner.