Source-linked AI summary
Charting a Decade of Computational Linguistics in Italy: The CLiC-it Corpus
Chiara Alzetta, Serena Auriemma, Alessandro Bondielli, Luca Dini, Chiara Fazzone, Alessio Miaschi, Martina Miliani, Marta Sartor
TL;DR
This paper examines how Italian computational linguistics and NLP research and its community have evolved amid changing priorities. It compiles and analyzes ten years of CLiC-it proceedings, finding growth in submissions and participation, gradual internationalisation, and diversification of topics alongside collaborative networks.
Problem
The study examines how Italian CL/NLP research and its community have evolved, focusing on authorship, participation, collaboration, and changing topics.
Source: §1 · p. 3Method
The authors compiled a structured corpus of 693 CLiC-it papers from 2014 to 2024 and analyzed authorship, diversity, collaboration networks, and research priorities.
Source: §5 · p. 18Results
Submissions and participants increased, contributions gradually internationalised, topics diversified, and co-authorship networks reflected cross-institutional and interdisciplinary openness.
Source: §1 · p. 2, §5 · pp. 18–19Takeaways & Limitations
The corpus and analysis provide empirical evidence for tracking Italian NLP research trends, collaboration networks, institutional contributions, and community evolution.
Source: §1 · p. 3Collaboration weights favor more frequent institutional collaboration and proportionally reduce affiliation-pair weights in papers with many authors from different institutions.
Source: §3.1 · p. 6
Abstract
from arXiv · showhide
Over the past decade, Computational Linguistics (CL) and Natural Language Processing (NLP) have evolved rapidly, especially with the advent of Transformer-based Large Language Models (LLMs). This shift has transformed research goals and priorities, from Lexical and Semantic Resources to Language Modelling and Multimodality. In this study, we track the research trends of the Italian CL and NLP community through an analysis of the contributions to CLiC-it, arguably the leading Italian conference in the field. We compile the proceedings from the first 10 editions of the CLiC-it conference (from 2014 to 2024) into the CLiC-it Corpus, providing a comprehensive analysis of both its metadata, including author provenance, gender, affiliations, and more, as well as the content of the papers themselves, which address various topics. Our goal is to provide the Italian and international research communities with valuable insights into emerging trends and key developments over time, supporting informed decisions and future directions in the field.
1. Introduction and Background
CLiC-it has grown over a decade as a venue for Italian computational linguistics and NLP, while research priorities have shifted toward language modelling, multimodality, and applications. The study presents a proceedings corpus and analyzes community metadata and research topics over time.
1. Introduction and Background
- CLiC-it is Italy’s leading conference dedicated to computational linguistics and natural language processing, organised by the Associazione Italiana di Linguistica Computazionale.Its editions were hosted by several Italian universities, with the tenth edition returning to Pisa in 2024.Source: §1 · p. 2
- The conference grew in paper and participant numbers as research priorities shifted from lexical and semantic resources toward language modelling, multimodality, and applications.The paper identifies affect detection and fact-checking among the applications associated with this shift.Source: §1 · pp. 2–3
- The study introduces a curated corpus of ten years of CLiC-it proceedings and analyzes both metadata and the development of research topics.The analysis includes author affiliations, geographic and gender distributions, and content.Source: §1 · p. 3
- The corpus and analysis offer evidence for exploring research trends, collaboration networks, institutional contributions, and the Italian NLP community’s evolution.The authors note its potential value for newcomers seeking an overview and experienced researchers exploring questions about authorship and international participation.Source: §1 · p. 3
2. The CLiC-it Corpus
The CLiC-it Corpus collects paper metadata and text from 693 papers across ten conference editions. The authors combine automatic and manual extraction, normalize the data, and provide derived variables for analyses of the community and its research.
2. The CLiC-it Corpus
- The corpus contains metadata and textual content from 693 papers presented across CLiC-it’s ten editions.Its design supports longitudinal analysis and can be extended with papers from future editions.Source: §2 · p. 4
- The authors obtained papers from editable source files where available and used PDF extraction for the 2014, 2018, 2019, and 2023 editions.They tested extraction strategies and manually corrected PDF-derived text.Source: §2 · p. 4
- LaTeX files were parsed automatically, while metadata and content from Word or PDF-derived text were extracted manually; 47 Italian papers were translated into English.The LaTeX workflow used TexSoup to extract metadata and abstracts, introductions, and conclusions.Source: §2 · p. 4
- The authors reviewed and normalized extracted data, harmonized affiliations, and retained abstracts, introductions, and conclusions for comparability across papers and editions.Internal departments and laboratories were omitted from university affiliations.Source: §2 · pp. 4–5
- The dataset includes derived authorship, affiliation, international-participation, and corporate-research variables for quantitative community analyses.The CLiC-it Corpus is freely available for further study and integration with other research initiatives.Source: §2 · p. 5
3. Contributors and Collaborations
The section traces CLiC-it participation and collaboration across its ten editions, combining authorship statistics with affiliation-network analysis.
3.1 Methodology
- The authors analyze author affiliations and participation over time using descriptive statistics, and examine institutional collaboration through a weighted directed affiliation graph.The graph weights balance collaboration frequency against institutions’ relative contributions to multi-institution papers.Source: §3.1 · pp. 5–6
3.2 Authorship Trends
- 346 authors contributed in 2024, versus 167 in 2014; first-time authors averaged 51.78% of contributors after excluding 2014.The 2024 edition set a record for author participation.Source: §3.2 · pp. 6, 8, Figure 1 · p. 7
- Non-academic author numbers remained fairly stable, while international participation became more consistent from 2019 and included contributors from Asia and the Americas.The Netherlands had the most papers involving non-Italian-affiliated authors (38), followed by Germany (27), and Spain and Switzerland (22 each).Source: Figure 2 · p. 7, §3.2 · p. 8, Figure 3 · p. 9
3.3 Community Collaborations
- The affiliation network has 16 components, including a 241-node giant component, but its overall density is approximately 0.017 and average node degree is 9.64.The remaining components are much smaller, and many contain only a few or isolated nodes.Source: §3.3 · pp. 9–10, Figure 5 · p. 11
3.3.1 Key Players in the Affiliation Network
- The centrality analysis ranks institutions by degree, closeness, and betweenness, then averages their ranking positions to capture prominence across structural dimensions.The authors state that combining rankings mitigates potential bias from relying on a single centrality measure.Source: §3.3.1 · pp. 10–12
- The University of Torino ranks most central across measures, with a combined average rank of 1.33; Groningen and CNRS also appear among the top 10.The three centrality measures generally yield similar rankings; their rank standard deviation reaches at most 4.93.Source: §3.3.1 · p. 12, Table 2 · p. 12
4. Content Analysis
The content analysis uses topic modelling to map CLiC-it papers into thematic clusters and trace their development across conference editions. It finds strong resource-oriented foundations alongside newer neural, multimodal, and socially focused research.
4.1 Methodology
- BERTopic yielded 24 initial clusters, which were manually merged by semantic overlap into 15 labelled clusters.The analysis applies the framework to paper abstracts, introductions, and conclusions, then uses ChatGPT to generate cluster labels.Source: §4.1 · p. 14
4.2 Research Trends in the CLiC-it Community
- The clusters span foundational resources and parsing, domain-specific retrieval and extraction, multimodal NLP, specialized corpora, text simplification, and socially relevant misinformation and gender-bias research.The largest affective-computing cluster is associated with social-media discourse, while Latin resources, learner corpora, and literary linguistics represent more specialized areas.Source: §4.2 · pp. 14–16
- Lexical and Semantic Resources and Analysis is the most represented topic, with 189 papers; a high number of papers also fall under Sentiment, Emotion, Irony, Hate, with 123 papers.Source: §4.2 · pp. 14–15
4.2.1 Diachronic Analysis of Topics
- The trends parallel ACL Anthology shifts toward neural language modelling and topic diversification; multimodal work has also expanded recently despite its longstanding presence.The paper characterizes the community’s trajectory as combining methodological innovation with ethical, inclusive, and application-driven research.Source: §4.2 · p. 16, §4.2.1 · p. 18
- Editions from 2014–2018 were resource-oriented; sentiment analysis remained popular but dipped in 2018 and 2022, with a 2020 peak possibly linked to pandemic-era online activity.The early period emphasized lexical and semantic resources and corpus-driven approaches.Source: §4.2.1 · pp. 16–17, Figure 7 · p. 17
- As literary analysis diminished, fake-news detection and gender-bias studies grew, shifting from Word2Vec-era studies around 2017 toward Transformer-based approaches from 2019.The paper suggests this change might reflect increased interest in socio-cultural applications.Source: §4.2.1 · p. 17
- Chatbots and dialogue systems emerged in 2019–2022, moving from templates and graph databases toward neural systems before generative LLMs quickly supplanted them in 2022.New conversational resources were produced to train data-driven agents.Source: §4.2.1 · p. 17
- Multimodality, Machine Translation, syntax, and Latin Resources persisted, while Distributional Semantics, Information Retrieval and Extraction, and corpus construction declined as Language Models grew after 2018 and Transformers predominated from 2019.The Language Models trajectory moved from traditional architectures toward fine-tuned pre-trained models for downstream tasks.Source: §4.2.1 · pp. 17–18
5. Conclusions
The authors compile and analyze 693 CLiC-it papers from 2014 to 2024. They report rising participation, internationalization, and topic diversity, and present the corpus as a resource for further research and future updates.
5. Conclusions
- The structured, curated corpus contains metadata and textual content from 693 papers presented across ten CLiC-it editions from 2014 to 2024.Its assembly combined automatic and manual extraction methods.Source: §5 · p. 18
- The analysis reports increasing submissions and participants, gradual internationalization, and diversification toward areas including multimodality, LLMs, misinformation detection, and sentiment analysis.Source: §5 · p. 18
- Co-authorship and institutional networks illustrate collaboration and openness to cross-institutional and interdisciplinary research in the Italian NLP landscape.Source: §5 · pp. 18–19
- The released corpus supports further empirical studies of authorship, topic evolution, and collaboration, and is designed for extension with papers from future editions.The authors expect future updates to enable longitudinal monitoring.Source: §5 · p. 19
Appendix A: Fields in the CLiC-it Corpus
The corpus records paper-level bibliographic, authorship, affiliation, and text fields, alongside derived indicators for analyzing participation and topics.
Appendix A: Fields in the CLiC-it Corpus
- Core paper fields include a unique identifier, conference year, title, language, authors, abstract, introduction, and conclusion.Source: Appendix A · p. 20
- Authorship fields record total authors, women authors, whether the first author is a woman, and names of international or company-affiliated authors.Source: Appendix A · p. 20
- Affiliation fields capture institutions, distinct-affiliation counts, international-affiliation indicators, company authors, and countries and cities represented.Source: Appendix A · p. 20
- Topic fields provide the number of topics extracted with BERTopic and the extracted topics themselves.Source: Appendix A · p. 20
Appendix B: Participation to CLiC-it by Country
Figure 1 presents country-level counts of affiliations outside Italy for each conference edition.
Appendix B: Participation to CLiC-it by Country
- Figure 1 counts affiliations from countries besides Italy.Source: Figure 1 · p. 22
- The figure groups the counts by country of affiliation.Source: Figure 1 · p. 22
- The per-edition breakdown allows comparison of non-Italian affiliation counts across editions.Source: Figure 1 · p. 22