Alejandro Benito-Santos

We surveyed 171 visualization design studies in the humanities to examine how their evaluation practices address both analytical utility and interpretive meaning. The study identifies recurring methodological weaknesses and recommends combining diverse forms of evidence to evaluate how visualizations support humanities research.

This work investigates capturing and modeling disagreement in Semantic Textual Similarity (STS), where sentence pairs are assigned ordinal similarity labels (0–5). Conventional STS systems average multiple annotator scores and focus on a single numeric estimate, overlooking label dispersion. By leveraging the disaggregated SemEval-2015 dataset (Soft-STS-15), this paper proposes and compares two disagreement-aware strategies that treat STS as an ordinal distribution prediction problem: a lightweight truncated Gaussian head for standard regression models, and a cross-encoder trained with a distance-aware objective, refined with temperature scaling. Results show improved performance in distance-based metrics, with the calibrated soft-label model proving best overall and notably more accurate on the most ambiguous pairs. This demonstrates that modeling disagreement benefits both calibration and ranking accuracy, highlighting the value of retaining and modeling full annotation distributions rather than collapsing them to a single mean label.

NLP research frequently grapples with multiple sources of variability—spanning runs, datasets, annotators, and more—yet conventional analysis methods often neglect these hierarchical structures, threatening the reproducibility of findings. To address this gap, we contribute a case study illustrating how linear mixed-effects models (LMMs) can rigorously capture systematic language-dependent differences (i.e., population-level effects) in a population of monolingual and multilingual language models. In the context of a bilingual hate speech detection task, we demonstrate that LMMs can uncover significant population-level effects—even under low-resource (small-N) experimental designs—while mitigating confounds and random noise. By setting out a transparent blueprint for repeated-measures experimentation, we encourage the NLP community to embrace variability as a feature, rather than a nuisance, in order to advance more robust, reproducible, and ultimately trustworthy results.

Song lyrics pose unique challenges for semantic similarity assessment due to their metaphorical language, structural patterns, and cultural nuances - characteristics that often challenge standard natural language processing (NLP) approaches. These challenges stem from a tension between compositional and distributional semantics: while lyrics follow compositional structures, their meaning depends heavily on context and interpretation. The Information Theory-based Compositional Distributional Semantics framework offers a principled approach by integrating information theory with compositional rules and distributional representations. We evaluate eight embedding models on Spanish song lyrics, including multilingual, monolingual contextual, and static embeddings. Results show that multilingual models consistently outperform monolingual alternatives, with the domain-adapted ALBERTI achieving the highest F1 macro scores (78.92 \textpm 10.86). Our analysis reveals that monolingual models generate highly anisotropic embedding spaces, significantly impacting performance with traditional metrics. The Information Contrast Model metric proves particularly effective, providing improvements up to 18.04 percentage points over cosine similarity. Additionally, composition functions maintaining longer accumulated vector norms consistently outperform standard averaging approaches. Our findings have important implications for NLP applications and challenge standard practices in similarity calculation, showing that effectiveness varies with both task nature and model characteristics.

The newly rediscovered frontier between data visualization and the digital humanities has proven to be an exciting field of experimentation for scholars from both disciplines. This fruitful collaboration is attracting researchers from other areas of science who may be willing to create visual analysis tools that promote humanities research in its many forms. However, as the collaboration grows in complexity, it may become intimidating for these scholars to get engaged in the discipline. To facilitate this task, we have built an introduction to visualization for the digital humanities that sits on a data-driven stance adopted by the authors. In order to construct a dataset representative of the discipline, we analyze citations from a core corpus on 300 publications in visualization for the humanities obtained from recent editions of the InfoVis Vis4DH workshop, the ADHO Digital Humanities Conference, and the specialized digital humanities journal Digital Humanities Quarterly. From here, we extract referenced works and analyze more than 1900 publications in search of citation patterns, prominent authors in the field, and other interesting insights. Finally, following the path set by other researchers in the visualization and Human-Computer Interaction (HCI) communities, we analyze paper keywords to identify significant themes and research opportunities in the field.

In this paper, we report the development of a model and a proof-of-concept visual text analytics (VTA) tool to enhance document discovery in a problem-driven visualization research (PDVR) context. The proposed model captures the cognitive model followed by domain and visualization experts by analyzing the interdisciplinary communication channel as represented by keywords found in two disjoint collections of research papers. High distributional intercollection similarities are employed to build informative keyword associations that serve as entry points to drive the exploration of a large document corpus. Our approach is demonstrated in the context of research on visualization for the digital humanities.

Nowadays, scholars dedicate a substantial amount of their work to the querying and browsing of increasingly large collections of research papers on the Internet. In parallel, the recent surge of novel interdisciplinary approaches in science requires scholars to acquire competencies in new fields for which they may lack the necessary vocabulary to formulate adequate queries. This problem, together with the issue of information overload, poses new challenges in the fields of natural language processing (NLP) and visualization design that call for a rapid response from the scientific community. In this respect, we report on a novel visualization scheme that enables the exploration of research paper collections via the analysis of semantic proximity relationships found in author-assigned keywords. Our proposal replaces traditional string queries with a bag-of-words (BoW) extracted from a user-generated auxiliary corpus that captures the intentionality of the research. Continuing along the lines established by other authors in the fields of literature-based discovery (LBD), NLP, and visual analytics (VA), we combine novel advances in the fields of NLP with visual network analysis techniques to offer scholars a perspective of the target corpus that better fits their research interests. To highlight the advantages of our proposal, we conduct two experiments employing a collection of visualization research papers and an auxiliary cross-domain BoW. Here, we showcase how our visualization can be used to maximize the effectiveness of a browsing session by enhancing the language acquisition task, which allows for effectively extracting knowledge that is in line with the users’ previous expectations.