Italian Association for Computational Linguistics

• Italian Association for Computational Linguistics is a leading society uniting academics and industry experts to advance NLP and AI-driven language applications.
• Its flagship CLiC-it conference demonstrates robust growth with increasing paper submissions and diverse, international collaborations in computational linguistics.
• The CALAMITA initiative exemplifies rigorous evaluation methods by standardizing Italian language model benchmarks through centralized metrics and open collaboration.

The Italian Association for Computational Linguistics (Associazione Italiana di Linguistica Computazionale, AILC) functions as the principal learned society focused on advancing research and dissemination in computational linguistics (CL) and NLP within Italy. Serving both national and international communities, the AILC provides a unified platform for researchers, academics, and industry professionals to present advancements in NLP, linguistic resources, machine learning for language, and AI-driven linguistic applications (Alzetta et al., 23 Sep 2025).

1. Foundation, Mission, and Scope

AILC’s founding year is not specified in the literature; however, it is established that AILC organized the first edition of the Italian Conference on Computational Linguistics (CLiC-it) in Pisa in December 2014, marking its role as the preeminent national CL association. The mission, as articulated in recent academic analyses, is "to provide a platform for researchers, academics, and industry professionals to present advancements in NLP, linguistic resources, machine learning for language, and linguistic applications driven by AI" (Alzetta et al., 23 Sep 2025).

This scope extends to uniting the Italian CL community and fostering openness to both international and industrial partnerships.

2. Organizational Structure and Membership Dynamics

AILC’s formal governance structure is not detailed in the available data. Instead, its activity and reach are best inferred from CLiC-it conference metrics, which serve as a practical proxy for association membership and engagement:

• Over the 2014–2024 period, CLiC-it accepted 693 papers authored by 2,006 unique contributors, averaging 3.39 authors per paper.
• Each year, after 2014, an average of 51.8% of CLiC-it contributors were newcomers—i.e., authors publishing at CLiC-it for the first time ($g(t) = \Delta N_t / N_t \approx 0.518$).
• Gender and affiliation are tracked at the author level, enabling meta-analyses using fields such as num_women_authors, woman_as_first_author, num_company_authors, and at_least_one_international_affiliation.
• Corporate involvement remains stable across editions, while international contributions show a steady increase, indicating broadening global reach (Alzetta et al., 23 Sep 2025).
• Collaboration networks (280 institutions, 1,350 weighted edges) demonstrate strong clustering around national hubs such as Università di Torino, Fondazione Bruno Kessler, and CNR-ILC.

3. CLiC-it: Flagship Conference and Community Hub

CLiC-it is AILC’s annual conference and the primary venue for Italian CL/NLP scholarship. The conference reflects both the quantitative growth and qualitative evolution of the community:

Year	Location	Submissions	Acceptances	Acceptance Rate (%)
2014	Pisa	97	75	77.32
2015	Trento	64	52	81.25
2016	Napoli	69	55	79.71
2017	Roma	72	58	80.56
2018	Torino	70	63	90.00
2019	Bari	82	75	91.46
2020	Bologna*	80	69	86.25
2021	Milano	68	59	86.76
2023	Venezia	86	75	87.21
2024	Pisa	136	114	83.82

* Held online

Submissions increased from under 100 in 2014 to 136 in 2024, with acceptance rates maintained between 77% and 92%, indicating both sustained selectivity and growth. The conference also functions as a data source for analyzing trends in authorship, institutional participation, and thematic focus (Alzetta et al., 23 Sep 2025).

4. Research Trends and Thematic Shifts

BERTopic-based analyses of CLiC-it submissions reveal diachronic changes in research priorities (Alzetta et al., 23 Sep 2025):

• 2014–2018: Emphasis on resource-building (lexical/semantic resources, corpora, syntax).
• 2019–2022: Intensified focus on dialogue systems, automated verification (fact-checking), and reduced attention to pure corpus construction and distributional semantics.
• Post-2019: Rapid proliferation of Transformer-based LLMs (e.g., GePpeTto, BERTino) displacing legacy SVM, RNN, and LSTM approaches.

Fifteen research topics were identified, led by Lexical and Semantic Resources and Analysis (189 papers), followed by Sentiment, Emotion, Irony, Hate (123). Novel areas, such as Fact Checking and Inclusive Language Studies, emerge from 2018 onward, while Multimodal, Machine Translation, and Latin Resources demonstrate thematic persistence. A plausible implication is the growing integration of generative models and multimodal applications, as well as the reemergence of tasks like Text Simplification and Learner Corpora in light of LLMs.

5. Large-Scale Evaluation and Community Initiatives: CALAMITA

In response to the dominance of English-centric LLM evaluation and the limited representation of Italian data in model training corpora, AILC launched and coordinated CALAMITA—Challenging the Abilities of LLMs in ITAlian (Nissim et al., 4 Dec 2025). This initiative exemplifies AILC’s role in community-driven benchmarking and methodological rigor:

• Over 80 contributors from 31 institutions contributed to the design, documentation, and evaluation of 22 challenge tasks (95 subtasks) spanning eight meta-abilities: Factual Knowledge, Linguistic Competence, Commonsense, Formal Reasoning, Fairness & Bias, Machine Translation, Summarization, and Code Generation.
• Data collection employed a structured, multi-stage submission protocol, centralizing technical reports and dataset documentation via standardized GitHub Gist templates.
• All tasks were integrated into a common pipeline built atop the lm-evaluation-harness library, supporting both classification and generative evaluation protocols, with core metrics including Precision, Recall, and F1-score and task-specific measures such as BLEU, ChrF, COMET, BLEURT, Pearson’s r, and BERTScore.
• Four open-weight LLMs were systematically benchmarked, with LLaMA3.1-70B-Instruct consistently outperforming smaller or less-tuned models across most ability categories.
• Methodological best practices included a harmonized evaluation team, centralized infrastructure leveraging national HPC (CINECA Leonardo), and robust reproducibility controls. Noted challenges were standardization of generative tasks and parsing of structured outputs.
• CALAMITA is designed as a rolling benchmark with ongoing rounds, scheduled data releases, and a “public/private” split to prevent training contamination, providing a blueprint for association-facilitated evaluation in other language communities.

6. Impact, Collaboration Networks, and Strategic Directions

AILC’s ecosystem exhibits increased scholarly output, broader internationalization (authors from 40+ countries), and stable industrial participation over the past decade (Alzetta et al., 23 Sep 2025). Core collaboration networks are tightly clustered around national research institutions, with Università di Torino, Fondazione Bruno Kessler, and CNR-ILC ranking as structural hubs based on combined centrality measures (Degree, Closeness, Betweenness). This networked structure supports cross-institutional and multi-sector synergy.

Strategic recommendations include the maintenance and extension of the CLiC-it Corpus post-2024, intensified meta-scientific analysis of inclusion and gender, and targeted support for emerging research priorities, including the study of Baby-LLMs for language acquisition and advanced multimodal processing. CALAMITA’s federated model is proposed as sustainable infrastructure for future LLM benchmarking, both in Italy and as a paradigm for non-English language communities (Nissim et al., 4 Dec 2025).

7. Resources for the Italian and International Community

AILC, through initiatives like CLiC-it and CALAMITA, provides longitudinal, meta-labeled corpora and empirically grounded frameworks to support decision-making and agenda-setting in Italian CL/NLP research and education (Alzetta et al., 23 Sep 2025, Nissim et al., 4 Dec 2025). These resources underpin strategic planning for research, instruction, and industrial collaboration, as well as facilitating rigorous, interpretable benchmarking of AI systems tailored to Italian and multilingual environments.

---

References:

• "Charting a Decade of Computational Linguistics in Italy: The CLiC-it Corpus" (Alzetta et al., 23 Sep 2025)
• "Challenging the Abilities of LLMs in Italian: a Community Initiative" (Nissim et al., 4 Dec 2025)