📄 Language Varieties of Italy: Technology Challenges and Opportunities

Summary

This position-and-survey paper looks at how NLP handles Italy's endangered local languages and dialects. Italy has one of the densest concentrations of linguistic diversity in Europe, and UNESCO lists more than 30 of its varieties as endangered. The author first describes the linguistic situation, grouping the varieties into macro-areas based on Pellegrini's classic dialect map. He then reviews existing monolingual and multilingual NLP work on these varieties.

The central argument is that current research is machine-centric. It treats the varieties as interchangeable, low-resource versions of standardized languages and assumes they have a stable written form and the same needs as Italian or English. In practice, most varieties are mainly spoken, have no standard orthography, mix with Italian, and fill different social roles.

The paper calls instead for speaker-centric NLP. This means engaging with communities on terms of reciprocity and building technology around what speakers actually need. To support this, the author launched a community initiative and outlines related research directions, such as modeling regional Italian. The work matters as a framework for responsible NLP on endangered and oral languages beyond Italy.

Key points

• Table 1 lists Italy's UNESCO-endangered varieties with their ISO 639-3 codes, linguistic branch, endangerment level, speaker counts, and whether a standard written form exists. The branches covered are Romance, Germanic, Slavic, Albanian, and Hellenic.
• Usage figures from ISTAT (2017):
  • 45.9% of people speak mostly Italian at home.
  • 32.3% mix Italian with a local language.
  • 14.1% speak mainly a local language.
  • For context, fewer than 10% spoke Italian when the country unified in 1861.
• The paper identifies three machine-centric assumptions:
  • an overemphasis on machine-readable written data;
  • little attention to whether that data represents the speech communities;
  • a uniform view of the functions and needs of different varieties.
• The author launched the community "Varieties of the Boot" (varietiesoftheboot.github.io), along with a GitHub repository that lists language and culture institutes.
• Proposed alternative directions include:
  • NLP for regional Italian, including fairness of models across regional variants;
  • large-scale study of language variation that complements the ALI and AIS linguistic atlases;
  • research on code-switching in its social context.