Abstract A recent contribution to Digital Scholarship in the Humanities by Meng and Hilpert (‘Measuring the Semantic Headedness of English Blends with Token-Based Semantic Vector Space Modelling: A Corpus-Based Study’, 2024, 39 (4), 1075–1091) applied distributional semantics to English lexical blends but conflated three conceptually independent dimensions: semantic headedness, semantic overlap, and semantic salience, while relying on static word embeddings that cannot adequately handle polysemy. To address these gaps, the present study distinguishes the three metrics and quantifies them using the contextualized language model BERT on a dataset of eighty-seven English blends. The research findings are threefold: (1) semantic headedness is right-headed in 52.9% and left-headed in 47.1% of the eighty-seven blends; semantic overlap is left‑oriented in 43.7% and right‑oriented in 56.3% of the eighty-seven blends; (2) in terms of semantic salience, 56.3% of the eighty-seven blends are left-oriented and 43.7% are right-oriented; (3) the three metrics exhibit systematic directional inconsistency, revealing that English blends function as multi-dimensional semantic objects in which categorical inheritance, distributional similarity, and contextual prominence need not converge, which calls for a more empirically grounded and multi-faceted account of blend semantics. Theoretically, our findings corroborate the gradient view of Meng and Hilpert, moving blend research beyond taxonomic classification towards a usage-grounded, multi-dimensional model that requires a multivariate approach. Methodologically, the study demonstrates the necessity of cross-validating conclusions using multiple metrics in morphological research, illustrating how digital methods can address longstanding problems in word formation.
Abstract This article investigates the affective architecture underlying narrative polyphony in Milan Kundera’s The Joke through a computational analysis of emotion dynamics. The study addresses a central narratological question: how distinct narrative voices construct differentiated emotional domains within a polyphonic framework. To explore this issue, the text is segmented according to Kundera’s four first-person narrators, and each segment is examined through SenticNet-based emotion modeling. For every narrative voice, the four core affective dimensions of the Hourglass of Emotion (attitude, introspection, sensitivity, temper) are computed together with a polarity-based sentiment index. The resulting trajectories are visualized across the novel, and key turning points—such as Ludvik’s deportation to the labor camp—are marked to assess how major events shape affective modulation. Radar charts and comparative boxplots reveal clear divergences among the emotional profiles of the narrators, indicating that Kundera’s polyphony operates not only on stylistic or ideological levels but also through the distribution of distinct affective regions. These results demonstrate that computational emotion analysis can operationalize interpretive intuitions commonly addressed through qualitative close reading, offering an empirically grounded perspective on the internal architecture of polyphonic fiction. The study highlights the interpretive potential of affective computing in literary analysis and suggests possible extensions to broader corpora of Central European narrative.
Abstract This paper examines the challenge of meaning-based verse retrieval for Kannada literary works, including Vachanas, Tripadis, and Kaggas, using Kannada queries. These literary verses are, in a sense, a form of poem-cum-prose. Given their philosophical depth and metaphorical nature, retrieval was performed using interpretations provided by Kannada scholars rather than direct text matching. We evaluated five retrieval approaches: TF–IDF, TF–IDF with Latent Semantic Indexing (LSI), KannadaSBERT-STS sentence embeddings, a cluster-based tree model of interpretations, and a proposed Hybrid Verse Retrieval (HVR) approach that fuses TF–IDF and KannadaSBERT-STS similarity scores. Experiments were conducted on a refined subset of the KannadaLit4NLP corpus comprising 5,158 verses with 17,930 scholarly interpretations, evenly split for training and testing. The results show that the proposed HVR approach performed best overall, achieving 82.40% exact-match accuracy at Top 20, an improvement over KannadaSBERT-STS embeddings alone (78.75%), which in turn outperformed the classical TF–IDF and LSI baselines. The cluster-based approach offered competitive early-rank accuracy and the added advantage of grouping related themes, though its performance plateaued at greater retrieval depths. Overall, the study demonstrates that combining statistical and transformer-based semantic signals through a simple, fixed-weight fusion can meaningfully improve verse retrieval, highlighting the promise of hybrid natural language processing approaches for the retrieval and exploration of Kannada’s literary heritage.
In the last decades, European policies have highlighted the need to digitize cultural heritage material to enable Member States to be active actors in the long-term preservation, accessibility, and usage of heritage. So far, many institutions have grappled with diverse technologies to expand their digital offer, enrich the promotion of their collections, and enhance visitors' experiences. However, many others still fight to find their place in the digital space. In this article, I will discuss the level of digitization reached by museums in Cyprus. The work integrates traditional approaches with quantitative and qualitative methods to identify the needs and challenges organizations and professionals face in dealing with digitization. The results show that a variety of digitization projects are underway across several museums, yet the majority have not been completed due to financial, organizational, and technical constraints. Moreover, a set of needs such as digitizing the collection, investing in technology for collection management, and enhancing on-site visitor experiences has been identified as a priority by professionals. The comprehensive review of the results contributes to giving visibility to the recent achievements in the digitization of heritage in an underexplored context of the South-Eastern Mediterranean, while also highlighting areas for improvement that serve as a foundation for theoretical reflections and practical solutions about the museum digitization.
Magazine front covers serve as a powerful visual gateway to capturing attention, communicating editorial positions, and reflecting prevailing social values. Despite their recognized influence, front covers of children's magazines remain understudied, particularly concerning their role in the formation of children's gender identity through the visual presentation of role models. To address these gaps, this study analysed 150 front covers from Modern Children (1941-51), one of Hong Kong's most influential children's literary magazines. Anchored in visual critical discourse analysis, this study applied machine learning to assess the accuracy of gender classification based on visual features, namely the characters' behaviour, setting, and clothing colours. Ensemble learning achieved around 70 per cent accuracy, indicating notable differences in the representations of female and male characters. Logistic regression revealed the presence of colour stereotypes, whereas behaviour and setting were not significant predictors. The salient presence of females demonstrates strong female visibility, while the frequent portrayals of boys and girls in outdoor activities reflect the magazine's promotion of play-a departure from traditional education and a response to national crises. This study represents an innovative application of machine learning techniques to the analysis of gender representations. It offers insights into cultural and social norms in Hong Kong in the 1940s and sheds light on how visual representations on magazine covers may contribute to child socialization. The study also illustrates the new research perspective that machine learning brings to critical discourse analysis.
