TL;DR
This document contains the proceedings of the 5th International Conference on Natural Language Processing for Digital Humanities (NLP4DH 2025). It showcases research at the intersection of computational linguistics and humanities, exploring how NLP techniques can be applied to historical, cultural, and social research, and vice-versa. The papers cover a wide range of topics, from corpus creation and analysis to the development of novel NLP methods for underrepresented languages and ethical considerations in AI.
What Problem Does This Paper Address?
The NLP4DH conference addresses the growing need for interdisciplinary dialogue at the intersection of computational methods and the humanities. As language technologies become more sophisticated, there's a critical need to explore their application in historical, cultural, and social research. This includes understanding how NLP can help uncover new insights from historical texts, analyze cultural artifacts, and address ethical challenges posed by AI in these domains. The conference serves as a platform for researchers, scholars, and practitioners to share their work, discuss unique challenges, and explore opportunities in this evolving field.
Methods
The proceedings showcase a diverse range of methods employed in NLP for Digital Humanities. These include:
- Traditional NLP techniques: Word segmentation, Part-of-Speech (POS) tagging, Named Entity Recognition (NER) applied to historical texts.
- Machine Translation (MT): Development and evaluation of MT systems, including RAG-enhanced approaches for ancient languages.
- Topic Modeling: Both classical and neural approaches for analyzing large corpora of historical documents.
- Sentiment Analysis: Applications to code-mixed languages and social media data.
- Argumentation Mining: Analyzing emotions and categories within arguments.
- Dataset Creation and Annotation: Development of novel datasets for specific tasks like irony detection, dialect dictionaries, and sentiment analysis in specialized domains.
- LLM Evaluation: Assessing LLMs for various capabilities such as creativity, narrative topic labeling, ethical gaps, and cognitive biases.
- Interdisciplinary Approaches: Combining NLP with historical linguistics, literary analysis, and social science methodologies.
Datasets
The proceedings mention or imply the use of various datasets, some of which are newly introduced or adapted for specific research questions:
- Historical Chinese Sources (1900-1950): Used for word segmentation, POS tagging, and NER.
- 20th Century Romanian Author Texts: Used for analyzing LLM pastiche ability.
- Ancient Egyptian Text: Used for RAG-enhanced Neural Machine Translation.
- Erzya and Moksha Dialect Dictionary Data: Used for restructuring and visualization.
- Rumble Podcast Data: Used for understanding podcast dynamics.
- Fanfiction Data: Used for analyzing maturity ratings and their effect on style and engagement.
- Hebrew Documents: A novel dataset for irony detection and neural classification.
- Historical Newspaper Articles: Used for genre and topic annotation, and for comparative studies of topic models.
- Finnish Healthcare SMEs: Used for a threefold model of AI readiness.
- Rigveda: Used for mapping hymns and organizing concepts.
- Czech Poetry: Used for automated analysis and generation.
- Classical Latin Texts: Used for augmented close reading with BERT.
- Philosophical Text: Used for evaluating NER tools.
- Old Irish Lexical Resources and Treebanks: Developed for diplomatic editions of Old Irish text.
- Icelandic Gigaword Corpus: Used for analyzing language use of political parties.
- Old Occitan Texts: Used for POS tagging.
- 19th Century US Newspapers: Used for data-driven investigation of euphemistic language.
- COVID-19 Vaccines Posts on X/Twitter (Brazilian Political Elites): Used for stance and sentiment annotation.
- Argument Data: Used for emotion category annotations.
- Images Spreading Hate Speech: Used for mitigating generation.
Key Insights and Ideas
- LLMs are powerful but require careful application in humanities: While LLMs show promise in tasks like creativity, translation, and analysis, their outputs need rigorous evaluation for accuracy, bias, and ethical implications, especially when dealing with sensitive historical or cultural data.
- The importance of specialized datasets: Many papers highlight the necessity of creating and utilizing domain-specific datasets (e.g., historical texts, dialectal data, code-mixed languages) to effectively apply NLP techniques to humanities research.
- Ethical considerations are paramount: The conference emphasizes the need to critically examine the ethical implications of NLP tools, including bias in LLMs, the potential for misuse, and the responsible development of AI for cultural heritage.
- Bridging the gap between NLP and humanities: The conference fosters collaboration, demonstrating how NLP can unlock new research avenues in history, literature, linguistics, and cultural studies, while also informing NLP development with humanities-driven questions and data.
- The evolution of language is a rich area for NLP: Analyzing language change over time, register variation, and stylistic features in historical and contemporary texts offers valuable insights into both linguistic evolution and societal trends.
Key Results
Not discussed in the paper. The document is a collection of proceedings, not a single research paper with specific quantitative results.
Limitations and Future Work
Not discussed in the paper. As this is a collection of conference proceedings, specific limitations and future work are detailed within individual papers, not in the overall document. However, the preface to the proceedings highlights the ongoing evolution of the field and the continuous need for dialogue and innovation at the intersection of NLP and the humanities.
Summary updated