In this paper, we evaluate two different natural language processing (NLP) approaches to solve a paradigmatic task for computational literary studies (CLS): the recognition of knowledge transfer in literary texts. We focus on the question of how adequately large language models capture the transfer of knowledge about family relations in German drama texts when this transfer is treated as a classification or textual entailment task using in-context learning (ICL). We find that a 13 billion parameter LLAMA 2 model performs best on the former, while GPT-4 performs best on the latter task. However, all models achieve relatively low scores compared to standard NLP benchmark results, struggle from inconsistencies with small changes in prompts and are often not able to make simple inferences beyond the textual surface, which is why an unreflected generic use of ICL in the CLS seems still not advisable.

Current top-performing coreference resolution approaches are limited with regard to the maximum length of texts they can accept. We explore a recursive merging technique of entities that allows us to apply coreference models to texts of arbitrary length, as found in many narrative genres. In experiments on established datasets, we quantify the drop in resolution quality caused by this approach. Finally, we use an under-explored resource in the form of a fully coreference-annotated novel to illustrate our model’s performance for long documents in practice. Here, we achieve state-of-the-art performance, outperforming previous systems capable of handling long documents.

The metaphorical framing of refugees, asylum seekers, and immigrants (RASIM) has been widely explored in academia, but mainly through close analysis. The present research outlines a large-scale computational investigation of RASIM metaphors in UKs media discourse. We experiment with a method that facilitates automatic identification of RASIM metaphors in 21 years of RASIM-related news reports from eight popular UK newspapers. From the metaphors extracted, four overarching frames are identified. Further analysis reveals correlations between political bias and metaphor usage: overall, right-biased newspapers use RASIM metaphors more frequently than their left-biased counterparts. Within the metaphorical frames, water, disaster, and non-human metaphors are more prevalent in right-biased media. Additionally, diachronic analysis illustrates that the distinctions between left and right media have evolved over time. Water metaphors, for example, have become increasingly more representative of the political right in the past two decades.

Dehumanization is a pernicious process of denying some or all attributes of humanness to the target group. It is frequently cited as a common hallmark of incitement to commit genocide. The international security landscape has seen a dramatic shift following the 2022 Russian invasion of Ukraine. This, coupled with recent developments in the conceptualization of dehumanization, necessitates the creation of new techniques for analyzing and detecting this extreme violence-related phenomenon on a large scale. Our project pioneers the development of a detection system for instances of dehumanization. To achieve this, we collected the entire posting history of the most popular bloggers on Russian Telegram and tested classical machine learning, deep learning, and zero-shot learning approaches to explore and detect the dehumanizing rhetoric. We found that the transformer-based method for entity extraction SpERT shows a promising result of F 1 = 0.85 for binary classification. The proposed methods can be built into the systems of anticipatory governance, contribute to the collection of evidence of genocidal intent in the Russian invasion of Ukraine, and pave the way for large-scale studies of dehumanizing language. This paper contains references to language that some readers may find offensive.

This is a short paper describing the process of derivation of synthetic Judeo-French text. Judeo-French is one of a number of rare languages used in speaking and writing by Jewish communities as confined to a particular temporal and geographical frame (in this case, 11th- to 14th-century France). The number of resources in the language is very limited and its involvement in the contemporary domain of Natural Language Processing (NLP) is practically non-existent. This work outlines the compilation of a synthetic Judeo-French corpus. For the purpose, a pipeline of transformations is applied to Old French text belonging to the same general time period, leading to the derivation of text that is as reliable as possible in terms of phonological, morphological and lexical characteristics as witnessed in Judeo-French. Ultimately, the goal is for this synthetic corpus to be used in standard NLP tasks, such as Neural Machine Translation (NMT), as an instance of data augmentation.

In this study, we present a generalizable workflow to identify documents in a historic language with a nonstandard language and script combination, Armeno-Turkish. We introduce the task of detecting distinct patterns of multilinguality based on the frequency of structured language alternations within a document.

We introduce EmotionArcs, a dataset comprising emotional arcs from over 9,000 English novels, assembled to understand the dynamics of emotions represented in text and how these emotions may influence a novel ́s reception and perceived quality. We evaluate emotion arcs manually, by comparing them to human annotation and against other similar emotion modeling systems to show that our system produces coherent emotion arcs that correspond to human interpretation. We present and make this resource available for further studies of a large collection of emotion arcs and present one application, exploring these arcs for modeling reader appreciation. Using information-theoretic measures to analyze the impact of emotions on literary quality, we find that emotional entropy, as well as the skewness and steepness of emotion arcs correlate with two proxies of literary reception. Our findings may offer insights into how quality assessments relate to emotional complexity and could help with the study of affect in literary novels.

Multi-Word Expressions (MWEs) play a pivotal role in language use overall and in register formation more specifically, e.g. encoding field-specific terminology. Our study focuses on the identification and categorization of MWEs used in scientific writing, considering their formal characteristics as well as their developmental trajectory over time from the mid-17th century to the present. For this, we develop an approach combining three different types of methods to identify MWEs (Universal Dependency annotation, Partitioner and the Academic Formulas List) and selected measures to characterize MWE properties (e.g., dispersion by Kullback-Leibler Divergence and several association measures). This allows us to inspect MWEs types in a novel data-driven way regarding their functions and change over time in specialized discourse.

This paper introduces EventNet-ITA, a large, multi-domain corpus annotated full-text with event frames for Italian. Moreover, we present and thoroughly evaluate an efficient multi-label sequence labeling approach for Frame Parsing. Covering a wide range of individual, social and historical phenomena, with more than 53,000 annotated sentences and over 200 modeled frames, EventNet-ITA constitutes the first systematic attempt to provide the Italian language with a publicly available resource for Frame Parsing of events, useful for a broad spectrum of research and application tasks. Our approach achieves a promising 0.9 strict F1-score for frame classification and 0.72 for frame element classification, on top of minimizing computational requirements. The annotated corpus and the frame parsing model are released under open license.

The Moravians are a Christian group that has emerged from a 15th century movement. In this paper, we investigate how memoirs written by the devotees of this group can be analyzed with methods from computational linguistics, in particular sentiment analysis. To this end, we experiment with two different fine-tuning strategies and find that the best performance for ternary sentiment analysis (81% accuracy) is achieved by fine-tuning a German BERT model, outperforming in particular models trained on much larger German sentiment datasets. We further investigate the model(s) using SHAP scores and find that the best performing model struggles with multiple negations and mixed statements. Finally, we show two application scenarios motivated by research questions from religious studies.

We investigate the impact of the Plain English Movement (PEM) on the complexity of legal language in UK law reports from the 1950s-2010s, contrasting it with the evolution of scientific language. The PEM, emerging in the late 20th century, advocated for clear and understandable legal language. We define complexity through the concept of surprisal - an information-theoretic measure correlating with cognitive processing difficulty. Our research contrasts surprisal with traditional readability measures, which often overlook content. We hypothesize that, if the PEM has influenced legal language, there would be a reduction in complexity over time and a shift from a nominal to a more verbal style. We analyze text complexity and lexico-grammatical changes in line with PEM recommendations. Results indicate minimal impact of the PEM on both legal and scientific domains. This finding suggests future research should consider processing effort when advocating for linguistic norms to enhance accessibility.

The pressing need for digitization of historical documents has led to a strong interest in designing computerised image processing methods for automatic handwritten text recognition. However, not much attention has been paid on studying the handwritten text written in the margins, i.e. marginalia, that also forms an important source of information. Nevertheless, training an accurate and robust recognition system for marginalia calls for data-efficient approaches due to the unavailability of sufficient amounts of annotated multi-writer texts. Therefore, this work presents an end-to-end framework for automatic detection and recognition of handwritten marginalia, and leverages data augmentation and transfer learning to overcome training data scarcity. The detection phase involves investigation of R-CNN and Faster R-CNN networks. The recognition phase includes an attention-based sequence-to-sequence model, with ResNet feature extraction, bidirectional LSTM-based sequence modeling, and attention-based prediction of marginalia. The effectiveness of the proposed framework has been empirically evaluated on the data from early book collections found in the Uppsala University Library in Sweden. Source code and pre-trained models are available at Github.

In this paper, we bridge computational linguistics with historical methods to explore the potential of topic modeling in historical newspapers. Our case study focuses on British and American newspapers published in the second half of the 20th century that debate issues of Greek tourism, but our method can be transposed to any diachronic data. We demonstrate that Non-negative Matrix Factorization (NFM) can generate interpretable topics within the historical period under examination providing a tangible example of how computational text analysis can assist historical research. The contribution of our work is two-fold; first, the extracted topics are evaluated both by a computational linguist and by a historian highlighting the crucial role of domain experts when interpreting topic modeling outputs. Second, the extracted topics are contextualized within the historical and political environment in which they appear, providing interesting insights about the historical representations of Greek tourism over the years, and about the development and the hallmarks of American and British tourism in Greece across different historical periods (from 1945 to 1989). The comparative analysis between the American and the British press reveals interesting insights including similar responses to specific events as well as notable differences between British and American tourism to Greece during the historical periods under examination. Overall, the results of our analysis can provide valuable information for academics and researchers in the field of (Digital) Humanities and Social Sciences, as well as for stakeholders in the tourism industry.

The quality of automatic transcription of heritage documents, whether from printed, manuscripts or audio sources, has a decisive impact on the ability to search and process historical texts. Although significant progress has been made in text recognition (OCR, HTR, ASR), textual materials derived from library and archive collections remain largely erroneous and noisy. Effective post-transcription correction methods are therefore necessary and have been intensively researched for many years. As large language models (LLMs) have recently shown exceptional performances in a variety of text-related tasks, we investigate their ability to amend poor historical transcriptions. We evaluate fourteen foundation language models against various post-correction benchmarks comprising different languages, time periods and document types, as well as different transcription quality and origins. We compare the performance of different model sizes and different prompts of increasing complexity in zero and few-shot settings. Our evaluation shows that LLMs are anything but efficient at this task. Quantitative and qualitative analyses of results allow us to share valuable insights for future work on post-correcting historical texts with LLMs.

Recent approaches to automatically detect the speaker of an utterance of direct speech often disregard general information about characters in favor of local information found in the context, such as surrounding mentions of entities. In this work, we explore stylistic representations of characters built by encoding their quotes with off-the-shelf pretrained Authorship Verification models in a large corpus of English novels (the Project Dialogism Novel Corpus). Results suggest that the combination of stylistic and topical information captured in some of these models accurately distinguish characters among each other, but does not necessarily improve over semantic-only models when attributing quotes. However, these results vary across novels and more investigation of stylometric models particularly tailored for literary texts and the study of characters should be conducted.

This study extends previous research on literary quality by using information theory-based methods to assess the level of perplexity recorded by three large language models when processing 20th-century English novels deemed to have high literary quality, recognized by experts as canonical, compared to a broader control group. We find that canonical texts appear to elicit a higher perplexity in the models, we explore which textual features might concur to create such an effect. We find that the usage of a more heavily nominal style, together with a more diverse vocabulary, is one of the leading causes of the difference between the two groups. These traits could reflect “strategies” to achieve an informationally dense literary style.

Named Entity Recognition (NER) is an important step in many Natural Language Processing tasks. The existing state-of-the-art NER systems are however typically developed based on contemporary data, and not very well suited for analyzing historical text. In this paper, we present a comparative analysis of the performance of several language models when applied to Named Entity Recognition for historical Swedish text. The source texts we work with are documents from Swedish labour unions from the 19th and 20th century. We experiment with three off-the-shelf models for contemporary Swedish text, and one language model built on historical Swedish text that we fine-tune with labelled data for adaptation to the NER task. Lastly, we propose a hybrid approach by combining the results of two models in order to maximize usability. We show that, even though historical Swedish is a low-resource language with data sparsity issues affecting overall performance, historical language models still show very promising results. Further contributions of our paper are the release of our newly trained model for NER of historical Swedish text, along with a manually annotated corpus of over 650 named entities.

Part-of-speech tagging is foundational to natural language processing, transcending mere linguistic functions. However, taggers optimized for Classical Latin struggle when faced with diverse linguistic eras shaped by the language ́s evolution. Exploring 16th-century Latin from the correspondence and assessing five Latin treebanks, we focused on carefully evaluating tagger accuracy and refining Large Language Models for improved performance in this nuanced linguistic context. Our discoveries unveiled the competitive accuracies of different versions of GPT, particularly after fine-tuning. Notably, our best fine-tuned model soared to an average accuracy of 88.99% over the treebank data, underscoring the remarkable adaptability and learning capabilities when fine-tuned to the specific intricacies of Latin texts. Next to emphasising GPT’s part-of-speech tagging capabilities, our second aim is to strengthen taggers ́ adaptability across different periods. We establish solid groundwork for using Large Language Models in specific natural language processing tasks where part-of-speech tagging is often employed as a pre-processing step. This work significantly advances the use of modern language models in interpreting historical language, bridging the gap between past linguistic epochs and modern computational linguistics.

This paper discusses two approaches to the diachronic normalization of Polish texts: a rule-based solution that relies on a set of handcrafted patterns, and a neural normalization model based on the text-to-text transfer transformer architecture. The training and evaluation data prepared for the task are discussed in detail, along with experiments conducted to compare the proposed normalization solutions. A quantitative and qualitative analysis is made. It is shown that at the current stage of inquiry into the problem, the rule-based solution outperforms the neural one on 3 out of 4 variants of the prepared dataset, although in practice both approaches have distinct advantages and disadvantages.

Digital archive collections that have been contributed by communities, known as community-generated digital content (CGDC), are important sources of historical and cultural knowledge. However, CGDC items are not easily searchable due to semantic information being obscured within their textual metadata. In this paper, we investigate the extent to which state-of-the-art, general-domain entity linking (EL) models (i.e., BLINK, EPGEL and mGENRE) can map named entities mentioned in CGDC textual metadata, to Wikidata entities. We evaluate and compare their performance on an annotated dataset of CGDC textual metadata and provide some error analysis, in the way of informing future studies aimed at enriching CGDC metadata using entity linking methods.

The application of text mining methods is becoming more and more popular, not only in Digital Humanities (DH) and Computational Social Sciences (CSS) in general, but also in vocational education and training (VET) research. Employing algorithms offers the possibility to explore corpora that are simply too large for manual methods. However, challenges arise when dealing with abstract concepts like occupations or skills, which are crucial subjects of VET research. Since algorithms require concrete instructions, either in the form of rules or annotated examples, these abstract concepts must be broken down as part of the operationalisation process. In our paper, we tackle the task of identifying occupational titles in the plenary protocols of the German Bundestag. The primary focus lies in the comparative analysis of two distinct approaches: a dictionary-based method and a BERT fine-tuning approach. Both approaches are compared in a quantitative evaluation and applied to a larger corpus sample. Results indicate comparable precision for both approaches (0.93), but the BERT-based models outperform the dictionary-based approach in terms of recall (0.86 vs. 0.77). Errors in the dictionary-based method primarily stem from the ambiguity of occupational titles (e.g., ‘baker’ as both a surname and a profession) and missing terms in the dictionary. In contrast, the BERT model faces challenges in distinguishing occupational titles from other personal names, such as ‘mother’ or ‘Christians’.

We present a novel combination of dynamic embedded topic models and change-point detection to explore diachronic change of lexical semantic modality in classical and early Christian Latin. We demonstrate several methods for finding and characterizing patterns in the output, and relating them to traditional scholarship in Comparative Literature and Classics. This simple approach to unsupervised models of semantic change can be applied to any suitable corpus, and we conclude with future directions and refinements aiming to allow noisier, less-curated materials to meet that threshold.

Many collections of digitized newspapers suffer from poor OCR quality, which impacts readability, information retrieval, and analysis of the material. Errors in OCR output can be reduced by applying machine translation models to “translate” it into a corrected version. Although transformer models show promising results in post-OCR correction and related tasks in other languages, they have not yet been explored for correcting OCR errors in Swedish texts. This paper presents a post-OCR correction model for Swedish 19th to 21th century newspapers based on the pre-trained transformer model ByT5. Three versions of the model were trained on different mixes of training data. The best model, which achieved a 36% reduction in CER, is made freely available and will be integrated into the automatic processing pipeline of Sprakbanken Text, a Swedish language technology infrastructure containing modern and historical written data.

In this paper we present the Kronieken Corpus, a new digital collection of 204 chronicles written in Dutch/Flemish between 1500 and 1850, which have been scanned, transcribed and annotated with named entities, dates, pages and a smaller part with sources and attributions. The texts belong to 308 physical volumes and contain between 23 and 24 million words. 107 chronicles, or 178 chronicle volumes, collected from 39 different archives and libraries in The Netherlands and Belgium and transcribed by volunteers had never been transcribed or published before. The result is a unique enriched historical text corpus of original hand-written, non-canonical and non-fiction text by lay people from the early modern period.

Identifying direct speech in literary fiction is challenging for cases that do not mark speech segments with quotation marks. Such efforts have previously been based either on smaller manually annotated gold data or larger automatically annotated silver data, extracted from works with quotation marks. However, no direct comparison has so far been made between the performance of these two types of training data. In this work, we address this gap. We further explore the effect of different types of typographical speech marking and of using evaluation metrics of different granularity. We perform experiments on Swedish literary texts and find that using gold and silver data has different strengths, with gold data having stronger results on token-level metrics, whereas silver data overall has stronger results on span-level metrics. If the training data contains some data that matches the typographical speech marking of the target, that is generally sufficient for achieving good results, but it does not seem to hurt if the training data also contains other types of marking.

We present a novel corpus consisting of orthographically variant words found in works of 19th century U.S. literature annotated with their corresponding “standard” word pair. We train a set of neural edit distance models to pair these variants with their standard forms, and compare the performance of these models to the performance of a set of neural edit distance models trained on a corpus of orthographic errors made by L2 English learners. Finally, we analyze the relative performance of these models in the light of different negative training sample generation strategies, and offer concluding remarks on the unique challenge literary orthographic variation poses to string pairing methodologies.

Coreference annotation and resolution is a vital component of computational literary studies. However, it has previously been difficult to build high quality systems for fiction. Coreference requires complicated structured outputs, and literary text involves subtle inferences and highly varied language. New language-model-based seq2seq systems present the opportunity to solve both these problems by learning to directly generate a copy of an input sentence with markdown-like annotations. We create, evaluate, and release several trained models for coreference, as well as a workflow for training new models.

The automatic classification of stage directions is a little explored topic in computational drama analysis, in spite of their relevance for plays’ structural and stylistic analysis. With a view to start assessing good practices for the automatic annotation of this textual element, we developed a 13-class stage direction typology, based on annotations in the FreDraCor corpus (French-language plays), but abstracting away from their huge variability while still providing classes useful for literary research. We fine-tuned transformers-based models to classify against the typology, gradually decreasing the corpus size used for fine tuning, to compare model efficiency with reduced training data. A result comparison speaks in favour of distilled monolingual models for this task, and, unlike earlier research on German, shows no negative effects of model case-sensitivity. The results have practical relevance for computational literary studies, as comparing classification results with complementary stage direction typologies, limiting the amount of manual annotation needed to apply them, would be helpful towards a systematic study of this important textual element.