BM-TOK
Balancing multilingual token vocabularies
How to build the vocabulary of a multilingual language model so that less-resourced languages and non-Latin scripts are represented fairly.
Fabio Rinaldi
Group leader (former member, 2019–2026)
Fabio founded and led the NLP group at IDSIA from 2019. Before that he was a lecturer and senior researcher at the University of Zurich. His work focused on biomedical text mining: the scientific literature, clinical reports and health-related social media.
Projects
LPHI-mRAG
Language preference and hallucination in multilingual retrieval-augmented generation
Do language models prefer some languages over others when answering from documents? We test this for chatbots that answer questions from a collection of documents in several languages.
GURU (ENERGENIUS)
Conversational AI for energy-efficiency advice, a component of the ENERGENIUS project
A conversational interface where citizens ask in natural language about energy savings, costs and regulation.
QUADRATIC
NLP in support of pharmacovigilance: Quality Adverse Drug Reaction Active Control
Finds reports of adverse drug reactions in clinical text, to support drug-safety monitoring.
Swiss AI Initiative
Coordination of IDSIA's participation in the Swiss AI Initiative
Coordinated IDSIA's part in the national effort to build and apply large AI models on the Alps supercomputer at CSCS.
Brisk.AI
Simplified and translated scientific literature for education
Used AI to produce simplified and translated versions of scientific papers for teaching.
DAG-MTSM
Detecting misinformation and machine-generated text
Developed and tested ways to detect misleading content and text written by language models. Produced the BUST benchmark.
ELDI
Dialect identification in short texts
Developed methods to recognise the dialect of short texts such as social media posts, focusing on varieties of Arabic.
Mini-MUSE
AI-based visualisation methods to explore digitised publications
Built interactive visual tools to explore large collections of digitised historical publications.
PREGAMUS
Reducing the complexity of scientific literature
Used AI to make scientific papers easier to read, as a support against misinformation.
WRSD
Identifying the risk of, and preventing, work-related stress disorders
Explored how text and data can help identify the risk of work-related stress.
ArthroTraitMine
Mining arthropod traits from the literature
Built text-mining workflows and a database of traits of arthropods (insects, spiders, crustaceans and relatives) collected from the scientific literature.
BERGAMOS
Open, machine-readable annotations of biomedical literature
Connected our entity recognition tool to PubAnnotation, so researchers can automatically annotate collections of biomedical papers.
INCdid
Dialect identification in noisy text
Developed methods to identify dialects in short, noisy texts, especially between closely related language varieties.
MisInfoCOV
Misinformation in social media during COVID-19
Studied how to identify misinformation on social media, and public stance towards health policies, during the pandemic.
COVID-19 Twitter Monitor
Social media mining for health
Combined ten methods for analysing COVID-19 tweets into one public platform with interactive visualisations.
LifeLike / BOOST
Voice-based training simulations
With SkillGym, moved a leadership-training simulator from clicking options to speaking, using language understanding to follow the conversation.
TalentScout
Entity recognition for CVs and job ads
With a pharmaceutical company, built entity recognition for job titles, organisations and places in CVs, outperforming off-the-shelf models.
Patient insights from social media
Mining patient insights in social media conversations
With Roche, extracted information from patient conversations online to support patient-centred drug development.
StageAI
Conversational recommendation of performers
With Stagend, improved a platform that matches performers to events, using multilingual question answering to fill in missing information.
SwissMADE
Automated detection of adverse drug events from older inpatients' records
Used NLP and data mining on electronic medical records from five Swiss hospitals to detect adverse drug events in older patients. Part of the National Research Programme NRP 74 'Smarter Health Care'.
Publications with the group
- CANE is a Format: Rethinking Output Formats for Few-Shot LLM-Based Named Entity Recognition
- Automatic Medical Discharge Summary Evaluation
- Are Language Models Script-Aware?
- Script Choice in LLMs: Evidence for Late-Layer Commitment
- Are the LLMs Capable of Maintaining at Least the Language Genus?
- Presenting LLMs' Collective Intelligence Approach for Multilingual Hallucination Detection
- Cross-lingual Transfer-learning Approach to Negation Scope Resolution