Skip to main content Link Menu Expand (external link) Document Search Copy Copied
ATE-IT Shared Task

Participation Guidelines

How to participate

Participation is open to all NLP practitioners worldwide: students, researchers, and industrial teams. You may join as an individual or as part of a team.

Registration deadline: 21 November, 2025

To register, please send an email to ateit.info@gmail.com with the subject line “ATE-IT Team Registration” and include the following details:

  • Group name and acronym
  • Participants’ full names and affiliations
  • Contact email address(es)
  • Subtask(s) you wish to participate in
  • Optional: Discord username(s)

👉 Important: At least one author of each accepted paper must register for the EVALITA conference in Bari (with a workshop fee) and present their work during the ATE-IT workshop.

Program

Detailed program: Key Dates

The shared task consists of four consecutive phases.

  • Phase 1: Development (22 September – 23 November, 2025)
    Training and development data will be released on the Data page. Teams should design, implement, and test their system(s) for one or both subtasks using the provided data.

  • Phase 2: Evaluation (24 November – 1 December, 2025)
    Test data will be released on the Data page. Teams must generate predictions and submit results in the format required for each subtask (Subtask A - Output, Subtask B - Output). We also encourage participants to include a link to the source code repository on GitHub.

  • Phase 3: Reporting (15 December, 2025 – 16 February, 2026) Teams must write and submit a short technical report following the provided format. The report should describe the system(s) and include:
    • System architecture and methods
    • Resources and pretrained models used
    • Results and error analysis
  • Phase 4: Final Workshop (26 – 27 February, 2026; in Bari)
    Teams will present and discuss their work during the ATE-IT workshop at EVALITA.

Rules

There are no restrictions on techniques: supervised and unsupervised machine learning, rule-based systems, deep learning, prompting, etc. Every technique is allowed.
However, please note the following recommendations:

  • We discourage simple fine-tuning of the latest models without additional qualitative analysis.
  • We strongly encourage the development of techniques that not only achieve good results but also provide insights into Automatic Term Extraction (ATE) and terminology research, thereby contributing to advancing the field. In particular, we encourage:
    • the use of open-source LLMs
    • the exploitation of linguistic knowledge

Allowed external resources include (but are not limited to):

  • pretrained embeddings
  • large language models (LLMs)
  • tokenisers, lemmatisers, POS taggers, etc.
  • other pretrained models
  • lexical resources
  • external corpora
  • terminology resources

All external resources must be clearly documented in the team’s technical report.

Participation options:
Teams may choose to participate in:

  • Subtask A (Term Extraction) only, or
  • Both Subtask A and Subtask B
  • Participation in Subtask B only (Term Variants Clustering) is not allowed.

Submission rules:

  • Each team may submit up to 2 runs per subtask.
  • For ranking, only the best run per evaluation metric will be considered.

Both training and development data may be used to train models.
Using test data for training is strictly forbidden.

Manual runs

We also accept runs where the test data is manually annotated.

Such runs must be thoroughly documented in the technical report. The report should include:

  • the annotators’ level of expertise
  • the annotation guidelines provided
  • the tools used for annotation
  • inter-annotator agreement measures

They will be used to compare ATE systems against human performance.

👉 Important: Manual runs will not be used for official ranking.

Submission

Submition deadline: (to be announced)

All submissions should be sent by email to ateit.info@gmail.com with the subject line “ATE-IT Submission”. Please attach your files in CSV or JSON format, following the specifications required for each subtask (Subtask A - format specification, Subtask B - format specification). Include the following information in the body of the email:

  • Group name and acronym
  • Subtask(s) addressed
  • Short description of your system (about 300 words)
  • Contact email address(es)
  • Optional: Link to the GitHub repository

Please ensure that your files are named consistently using the format:

[group_acronym]_[subtask_letter]_[runID].csv
[group_acronym]_[subtask_letter]_[runID].json

Table of contents