Jan Šnajder

Josip Jukić, Fran Jelenić, Miroslav Bićanić et al.

Supervised machine learning has become the cornerstone of today's data-driven society, increasing the need for labeled data. However, the process of acquiring labels is often expensive and tedious. One possible remedy is to use active learning (AL) -- a special family of machine learning algorithms designed to reduce labeling costs. Although AL has been successful in practice, a number of practical challenges hinder its effectiveness and are often overlooked in existing AL annotation tools. To address these challenges, we developed ALANNO, an open-source annotation system for NLP tasks equipped with features to make AL effective in real-world annotation projects. ALANNO facilitates annotation management in a multi-annotator setup and supports a variety of AL methods and underlying models, which are easily configurable and extensible.

Zoran Medić, Jan Šnajder

Recently introduced transformer-based article encoders (TAEs) designed to produce similar vector representations for mutually related scientific articles have demonstrated strong performance on benchmark datasets for scientific article recommendation. However, the existing benchmark datasets are predominantly focused on single domains and, in some cases, contain easy negatives in small candidate pools. Evaluating representations on such benchmarks might obscure the realistic performance of TAEs in setups with thousands of articles in candidate pools. In this work, we evaluate TAEs on large benchmarks with more challenging candidate pools. We compare the performance of TAEs with a lexical retrieval baseline model BM25 on the task of citation recommendation, where the model produces a list of recommendations for citing in a given input article. We find out that BM25 is still very competitive with the state-of-the-art neural retrievers, a finding which is surprising given the strong performance of TAEs on small benchmarks. As a remedy for the limitations of the existing benchmarks, we propose a new benchmark dataset for evaluating scientific article representations: Multi-Domain Citation Recommendation dataset (MDCR), which covers different scientific fields and contains challenging candidate pools.

Josip Jukić, Jan Šnajder

Developed to alleviate prohibitive labeling costs, active learning (AL) methods aim to reduce label complexity in supervised learning. While recent work has demonstrated the benefit of using AL in combination with large pre-trained language models (PLMs), it has often overlooked the practical challenges that hinder the effectiveness of AL. We address these challenges by leveraging representation smoothness analysis to ensure AL is feasible, that is, both effective and practicable. Firstly, we propose an early stopping technique that does not require a validation set -- often unavailable in realistic AL conditions -- and observe significant improvements over random sampling across multiple datasets and AL methods. Further, we find that task adaptation improves AL, whereas standard short fine-tuning in AL does not provide improvements over random sampling. Our work demonstrates the usefulness of representation smoothness analysis for AL and introduces an AL stopping criterion that reduces label complexity.

Josip Jukić, Martin Tutek, Jan Šnajder

A popular approach to unveiling the black box of neural NLP models is to leverage saliency methods, which assign scalar importance scores to each input component. A common practice for evaluating whether an interpretability method is faithful has been to use evaluation-by-agreement -- if multiple methods agree on an explanation, its credibility increases. However, recent work has found that saliency methods exhibit weak rank correlations even when applied to the same model instance and advocated for the use of alternative diagnostic methods. In our work, we demonstrate that rank correlation is not a good fit for evaluating agreement and argue that Pearson-$r$ is a better-suited alternative. We further show that regularization techniques that increase faithfulness of attention explanations also increase agreement between saliency methods. By connecting our findings to instance categories based on training dynamics, we show that the agreement of saliency method explanations is very low for easy-to-learn instances. Finally, we connect the improvement in agreement across instance categories to local representation space statistics of instances, paving the way for work on analyzing which intrinsic model properties improve their predisposition to interpretability methods.

Fran Jelenić, Josip Jukić, Martin Tutek et al.

Effective out-of-distribution (OOD) detection is crucial for reliable machine learning models, yet most current methods are limited in practical use due to requirements like access to training data or intervention in training. We present a novel method for detecting OOD data in Transformers based on transformation smoothness between intermediate layers of a network (BLOOD), which is applicable to pre-trained models without access to training data. BLOOD utilizes the tendency of between-layer representation transformations of in-distribution (ID) data to be smoother than the corresponding transformations of OOD data, a property that we also demonstrate empirically. We evaluate BLOOD on several text classification tasks with Transformer networks and demonstrate that it outperforms methods with comparable resource requirements. Our analysis also suggests that when learning simpler tasks, OOD data transformations maintain their original sharpness, whereas sharpness increases with more complex tasks.

Josip Jukić, Iva Vukojević, Jan Šnajder

Expressing attitude or stance toward entities and concepts is an integral part of human behavior and personality. Recently, evaluative language data has become more accessible with social media's rapid growth, enabling large-scale opinion analysis. However, surprisingly little research examines the relationship between personality and evaluative language. To bridge this gap, we introduce the notion of evaluative topics, obtained by applying topic models to pre-filtered evaluative text from social media. We then link evaluative topics to individual text authors to build their evaluative profiles. We apply evaluative profiling to Reddit comments labeled with personality scores and conduct an exploratory study on the relationship between evaluative topics and Big Five personality facets, aiming for a more interpretable, facet-level analysis. Finally, we validate our approach by observing correlations consistent with prior research in personality psychology.

Domagoj Pluščec, Jan Šnajder

Data scarcity is a problem that occurs in languages and tasks where we do not have large amounts of labeled data but want to use state-of-the-art models. Such models are often deep learning models that require a significant amount of data to train. Acquiring data for various machine learning problems is accompanied by high labeling costs. Data augmentation is a low-cost approach for tackling data scarcity. This paper gives an overview of current state-of-the-art data augmentation methods used for natural language processing, with an emphasis on methods for neural and transformer-based models. Furthermore, it discusses the practical challenges of data augmentation, possible mitigations, and directions for future research.

Laura Majer, Jan Šnajder

The increasing threat of disinformation calls for automating parts of the fact-checking pipeline. Identifying text segments requiring fact-checking is known as claim detection (CD) and claim check-worthiness detection (CW), the latter incorporating complex domain-specific criteria of worthiness and often framed as a ranking task. Zero- and few-shot LLM prompting is an attractive option for both tasks, as it bypasses the need for labeled datasets and allows verbalized claim and worthiness criteria to be directly used for prompting. We evaluate the LLMs' predictive and calibration accuracy on five CD/CW datasets from diverse domains, each utilizing a different worthiness criterion. We investigate two key aspects: (1) how best to distill factuality and worthiness criteria into a prompt and (2) what amount of context to provide for each claim. To this end, we experiment with varying the level of prompt verbosity and the amount of contextual information provided to the model. Our results show that optimal prompt verbosity is domain-dependent, adding context does not improve performance, and confidence scores can be directly used to produce reliable check-worthiness rankings.

David Dukić, Marin Petričević, Sven Ćurković et al.

TakeLab Retriever is an AI-driven search engine designed to discover, collect, and semantically analyze news articles from Croatian news outlets. It offers a unique perspective on the history and current landscape of Croatian online news media, making it an essential tool for researchers seeking to uncover trends, patterns, and correlations that general-purpose search engines cannot provide. TakeLab retriever utilizes cutting-edge natural language processing (NLP) methods, enabling users to sift through articles using named entities, phrases, and topics through the web application. This technical report is divided into two parts: the first explains how TakeLab Retriever is utilized, while the second provides a detailed account of its design. In the second part, we also address the software engineering challenges involved and propose solutions for developing a microservice-based semantic search engine capable of handling over ten million news articles published over the past two decades.

Josip Jukić, Jan Šnajder

Enhancing generalization and uncertainty quantification in pre-trained language models (PLMs) is crucial for their effectiveness and reliability. Building on machine learning research that established the importance of robustness for improving generalization, we investigate the role of representation smoothness, achieved via Jacobian and Hessian regularization, in enhancing PLM performance. Although such regularization methods have proven effective in computer vision, their application in natural language processing (NLP), where PLM inputs are derived from a discrete domain, poses unique challenges. We introduce a novel two-phase regularization approach, JacHess, which minimizes the norms of the Jacobian and Hessian matrices within PLM intermediate representations relative to their inputs. Our evaluation using the GLUE benchmark demonstrates that JacHess significantly improves in-domain generalization and calibration in PLMs, outperforming unregularized fine-tuning and other similar regularization methods.

Jana Juroš, Laura Majer, Jan Šnajder

News headlines often evoke sentiment by intentionally portraying entities in particular ways, making targeted sentiment analysis (TSA) of headlines a worthwhile but difficult task. Due to its subjectivity, creating TSA datasets can involve various annotation paradigms, from descriptive to prescriptive, either encouraging or limiting subjectivity. LLMs are a good fit for TSA due to their broad linguistic and world knowledge and in-context learning abilities, yet their performance depends on prompt design. In this paper, we compare the accuracy of state-of-the-art LLMs and fine-tuned encoder models for TSA of news headlines using descriptive and prescriptive datasets across several languages. Exploring the descriptive--prescriptive continuum, we analyze how performance is affected by prompt prescriptiveness, ranging from plain zero-shot to elaborate few-shot prompts. Finally, we evaluate the ability of LLMs to quantify uncertainty via calibration error and comparison to human label variation. We find that LLMs outperform fine-tuned encoders on descriptive datasets, while calibration and F1-score generally improve with increased prescriptiveness, yet the optimal level varies.

Ivan Rep, David Dukić, Jan Šnajder

While BERT produces high-quality sentence embeddings, its pre-training computational cost is a significant drawback. In contrast, ELECTRA provides a cost-effective pre-training objective and downstream task performance improvements, but worse sentence embeddings. The community tacitly stopped utilizing ELECTRA's sentence embeddings for semantic textual similarity (STS). We notice a significant drop in performance for the ELECTRA discriminator's last layer in comparison to prior layers. We explore this drop and propose a way to repair the embeddings using a novel truncated model fine-tuning (TMFT) method. TMFT improves the Spearman correlation coefficient by over $8$ points while increasing parameter efficiency on the STS Benchmark. We extend our analysis to various model sizes, languages, and two other tasks. Further, we discover the surprising efficacy of ELECTRA's generator model, which performs on par with BERT, using significantly fewer parameters and a substantially smaller embedding size. Finally, we observe boosts by combining TMFT with word similarity or domain adaptive pre-training.

Josip Jukić, Martin Tutek, Jan Šnajder

Large language models (LLMs) often seamlessly adapt to new tasks through in-context learning (ICL) or supervised fine-tuning (SFT). However, both of these approaches face key limitations: ICL is inefficient when handling many demonstrations, and SFT incurs training overhead while sacrificing flexibility. Mapping instructions or demonstrations from context directly into adapter parameters offers an appealing alternative. While prior work explored generating adapters based on a single input context, it has overlooked the need to integrate multiple chunks of information. To address this gap, we introduce CompAs, a meta-learning framework that translates context into adapter parameters with a compositional structure. Adapters generated this way can be merged algebraically, enabling instructions, demonstrations, or retrieved passages to be seamlessly combined without reprocessing long prompts. Critically, this approach yields three benefits: lower inference cost, robustness to long-context instability, and establishes a principled solution when input exceeds the model's context window. Furthermore, CompAs encodes information into adapter parameters in a reversible manner, enabling recovery of input context through a decoder, facilitating safety and security. Empirical results on diverse multiple-choice and extractive question answering tasks show that CompAs outperforms ICL and prior generator-based methods, especially when scaling to more inputs. Our work establishes composable adapter generation as a practical and efficient alternative for scaling LLM deployment.

David Dukić, Goran Glavaš, Jan Šnajder

Sequence labeling (SL) tasks, where labels are assigned to tokens, are abundant in NLP (e.g., named entity recognition and aspect-based sentiment analysis). Owing to the intuition that they require bidirectional context, SL tasks are commonly tackled with encoder-only models. Recent work also shows that removing the causal mask in fine-tuning enables decoder-based LLMs to become effective token classifiers. Less work, however, focused on (supervised) generative SL, a more natural setting for causal LLMs. Due to their rapid scaling, causal LLMs applied to SL are expected to outperform encoders, whose own development has stagnated. In this work, we propose supervised in-context fine-tuning (SIFT) for generative SL. SIFT casts SL tasks as constrained response generation, natural to LLMs, combining in-context learning (ICL) from demonstrations with supervised fine-tuning. SIFT considerably outperforms both ICL and decoder-as-encoder fine-tuning baselines on a range of standard SL tasks. We further find that although long context hinders the performance of generative SL in both ICL and SIFT, this deficiency can be mitigated by removing the instruction, as instructions are shown to be largely unnecessary for achieving strong SL performance with SIFT. Our findings highlight strengths and limitations of SL with LLMs, underscoring the importance of a response-based generative task formulation for effective SL performance.

Marija Anđelić, Dominik Šipek, Laura Majer et al.

Online news outlets operate predominantly on an advertising-based revenue model, compelling journalists to create headlines that are often scandalous, intriguing, and provocative -- commonly referred to as clickbait. Automatic detection of clickbait headlines is essential for preserving information quality and reader trust in digital media and requires both contextual understanding and world knowledge. For this task, particularly in less-resourced languages, it remains unclear whether fine-tuned methods or in-context learning (ICL) yield better results. In this paper, we compile CLIC, a novel dataset for clickbait detection of Croatian news headlines spanning a 20-year period and encompassing mainstream and fringe outlets. We fine-tune the BERTić model on this task and compare its performance to LLM-based ICL methods with prompts both in Croatian and English. Finally, we analyze the linguistic properties of clickbait. We find that nearly half of the analyzed headlines contain clickbait, and that finetuned models deliver better results than general LLMs.

David Dukić, Jan Šnajder

Pre-trained language models based on masked language modeling (MLM) excel in natural language understanding (NLU) tasks. While fine-tuned MLM-based encoders consistently outperform causal language modeling decoders of comparable size, recent decoder-only large language models (LLMs) perform on par with smaller MLM-based encoders. Although their performance improves with scale, LLMs fall short of achieving state-of-the-art results in information extraction (IE) tasks, many of which are formulated as sequence labeling (SL). We hypothesize that LLMs' poor SL performance stems from causal masking, which prevents the model from attending to tokens on the right of the current token. Yet, how exactly and to what extent LLMs' performance on SL can be improved remains unclear. We explore techniques for improving the SL performance of open LLMs on IE tasks by applying layer-wise removal of the causal mask (CM) during LLM fine-tuning. This approach yields performance gains competitive with state-of-the-art SL models, matching or outperforming the results of CM removal from all blocks. Our findings hold for diverse SL tasks, demonstrating that open LLMs with layer-dependent CM removal outperform strong MLM-based encoders and even instruction-tuned LLMs.

Josip Jukić, Jan Šnajder

Pre-trained language models (PLMs) have ignited a surge in demand for effective fine-tuning techniques, particularly in low-resource domains and languages. Active learning (AL), a set of algorithms designed to decrease labeling costs by minimizing label complexity, has shown promise in confronting the labeling bottleneck. In parallel, adapter modules designed for parameter-efficient fine-tuning (PEFT) have demonstrated notable potential in low-resource settings. However, the interplay between AL and adapter-based PEFT remains unexplored. We present an empirical study of PEFT behavior with AL in low-resource settings for text classification tasks. Our findings affirm the superiority of PEFT over full-fine tuning (FFT) in low-resource settings and demonstrate that this advantage persists in AL setups. We further examine the properties of PEFT and FFT through the lens of forgetting dynamics and instance-level representations, where we find that PEFT yields more stable representations of early and middle layers compared to FFT. Our research underscores the synergistic potential of AL and PEFT in low-resource settings, paving the way for advancements in efficient and effective fine-tuning.

David Dukić, Kiril Gashteovski, Goran Glavaš et al.

Event detection is a crucial information extraction task in many domains, such as Wikipedia or news. The task typically relies on trigger detection (TD) -- identifying token spans in the text that evoke specific events. While the notion of triggers should ideally be universal across domains, domain transfer for TD from high- to low-resource domains results in significant performance drops. We address the problem of negative transfer in TD by coupling triggers between domains using subject-object relations obtained from a rule-based open information extraction (OIE) system. We demonstrate that OIE relations injected through multi-task training can act as mediators between triggers in different domains, enhancing zero- and few-shot TD domain transfer and reducing performance drops, in particular when transferring from a high-resource source domain (Wikipedia) to a low(er)-resource target domain (news). Additionally, we combine this improved transfer with masked language modeling on the target domain, observing further TD transfer gains. Finally, we demonstrate that the gains are robust to the choice of the OIE system.

Zoran Medić, Jan Šnajder

Citation recommendation (CR) models may help authors find relevant articles at various stages of the paper writing process. Most research has dealt with either global CR, which produces general recommendations suitable for the initial writing stage, or local CR, which produces specific recommendations more fitting for the final writing stages. We propose the task of paragraph-level CR as a middle ground between the two approaches, where the paragraph's topic sentence is taken as input and recommendations for citing within the paragraph are produced at the output. We propose a model for this task, fine-tune it using the quadruplet loss on the dataset of ACL papers, and show improvements over the baselines.

Fran Jelenić, Josip Jukić, Nina Drobac et al.

Active learning (AL) aims to reduce labeling costs by querying the examples most beneficial for model learning. While the effectiveness of AL for fine-tuning transformer-based pre-trained language models (PLMs) has been demonstrated, it is less clear to what extent the AL gains obtained with one model transfer to others. We consider the problem of transferability of actively acquired datasets in text classification and investigate whether AL gains persist when a dataset built using AL coupled with a specific PLM is used to train a different PLM. We link the AL dataset transferability to the similarity of instances queried by the different PLMs and show that AL methods with similar acquisition sequences produce highly transferable datasets regardless of the models used. Additionally, we show that the similarity of acquisition sequences is influenced more by the choice of the AL method than the choice of the model.

Damir Korenčić, Strahil Ristov, Jelena Repar et al.

Topic models are widely used unsupervised models capable of learning topics - weighted lists of words and documents - from large collections of text documents. When topic models are used for discovery of topics in text collections, a question that arises naturally is how well the model-induced topics correspond to topics of interest to the analyst. In this paper we revisit and extend a so far neglected approach to topic model evaluation based on measuring topic coverage - computationally matching model topics with a set of reference topics that models are expected to uncover. The approach is well suited for analyzing models' performance in topic discovery and for large-scale analysis of both topic models and measures of model quality. We propose new measures of coverage and evaluate, in a series of experiments, different types of topic models on two distinct text domains for which interest for topic discovery exists. The experiments include evaluation of model quality, analysis of coverage of distinct topic categories, and the analysis of the relationship between coverage and other methods of topic model evaluation. The paper contributes a new supervised measure of coverage, and the first unsupervised measure of coverage. The supervised measure achieves topic matching accuracy close to human agreement. The unsupervised measure correlates highly with the supervised one (Spearman's $ρ\geq 0.95$). Other contributions include insights into both topic models and different methods of model evaluation, and the datasets and code for facilitating future research on topic coverage.

Martin Tutek, Jan Šnajder

The attention mechanism has quickly become ubiquitous in NLP. In addition to improving performance of models, attention has been widely used as a glimpse into the inner workings of NLP models. The latter aspect has in the recent years become a common topic of discussion, most notably in work of Jain and Wallace, 2019; Wiegreffe and Pinter, 2019. With the shortcomings of using attention weights as a tool of transparency revealed, the attention mechanism has been stuck in a limbo without concrete proof when and whether it can be used as an explanation. In this paper, we provide an explanation as to why attention has seen rightful critique when used with recurrent networks in sequence classification tasks. We propose a remedy to these issues in the form of a word level objective and our findings give credibility for attention to provide faithful interpretations of recurrent models.

Matej Gjurković, Mladen Karan, Iva Vukojević et al.

Personality and demographics are important variables in social sciences, while in NLP they can aid in interpretability and removal of societal biases. However, datasets with both personality and demographic labels are scarce. To address this, we present PANDORA, the first large-scale dataset of Reddit comments labeled with three personality models (including the well-established Big 5 model) and demographics (age, gender, and location) for more than 10k users. We showcase the usefulness of this dataset on three experiments, where we leverage the more readily available data from other personality models to predict the Big 5 traits, analyze gender classification biases arising from psycho-demographic variables, and carry out a confirmatory and exploratory analysis based on psychological theories. Finally, we present benchmark prediction models for all personality and demographic variables.

Ivan Sekulić, Matej Gjurković, Jan Šnajder

Bipolar disorder, an illness characterized by manic and depressive episodes, affects more than 60 million people worldwide. We present a preliminary study on bipolar disorder prediction from user-generated text on Reddit, which relies on users' self-reported labels. Our benchmark classifiers for bipolar disorder prediction outperform the baselines and reach accuracy and F1-scores of above 86%. Feature analysis shows interesting differences in language use between users with bipolar disorders and the control group, including differences in the use of emotion-expressive words.

Martin Tutek, Jan Šnajder

Natural language processing has greatly benefited from the introduction of the attention mechanism. However, standard attention models are of limited interpretability for tasks that involve a series of inference steps. We describe an iterative recursive attention model, which constructs incremental representations of input data through reusing results of previously computed queries. We train our model on sentiment classification datasets and demonstrate its capacity to identify and combine different aspects of the input in an easily interpretable manner, while obtaining performance close to the state of the art.

Jan Šnajder

Argumentation mining from social media content has attracted increasing attention. The task is both challenging and rewarding. The informal nature of user-generated content makes the task dauntingly difficult. On the other hand, the insights that could be gained by a large-scale analysis of social media argumentation make it a very worthwhile task. In this position paper I discuss the motivation for social media argumentation mining, as well as the tasks and challenges involved.