Old dialects are gold worth for AI

How can some Halland guys make AI technology better? When American speech-to-text models are to learn Swedish, old dialect recordings have proven to be a goldmine.

» Published:

Old dialects are gold worth for AI
Photo: Uppsala universitet

Share this article

At the Institute for Language and Folk Culture (Isof), they have worked extensively with dialect collection, primarily between 1935 and 1970, when researchers traveled around the country recording how people spoke.

The initial purpose was largely to preserve traditional dialects, explains Annette C Torensjö, head of the Department of Archives and Research in Uppsala at Isof.

The recordings were used for many years mainly by researchers, but last year, the KB-Lab at the Royal Library got in touch. Leonora Vesterbacka is a senior data scientist and leads a project to train speech-to-text models, which "translate" speech into written text.

A Small Part of Swedish

American AI models are often trained on massive multilingualism, and then the Swedish part becomes very small.

There is a model from Open AI called Whisper, which is trained on 680,000 hours of various languages, of which around 2,000 hours are Swedish. That's 0.3 percent Swedish, explains Leonora Vesterbacka.

The models work well on standard Swedish, but worse on dialects. That's where the Halland dialects, and others who had their dialects recorded, come in.

If someone had told me in the 1980s that the new gold would be an old weird roll of recordings in an archive, I would have thought they were joking.

They also use protocols and recordings from the Riksdag Administration.

They are responsible for recording what is said in the Riksdag and then making it available and releasing protocols. This has been going on for a long time.

In Sweden, we are so good at preserving everything. It's amazing to see that it can be used in the future as well.

Making Available

When the models are fully trained, they can, for example, be used to transcribe medical records and meetings or write subtitles for TV broadcasts. They can also be used to make spoken material, such as podcasts and TV broadcasts from authorities, available.

At Isof, they had not believed that their old dialect recordings would become an important step into the future.

I'm very pleased that our dialect recordings really have such great relevance now, that this is something that can be worked with as part of social development, says Annette C Torensjö.

A dialect is a language variant spoken by the inhabitants of a geographically defined area. It differs from the standard language as well as from adjacent dialects. The differences can concern language features at all levels: phonological, morphological, lexical, and syntactic.

Sociolect, or social dialect, is a language variant that is characteristic of a particular social group. Sociolects are often limited to pronunciation, grammar, and vocabulary, and are also regionally restricted, so that members of the same social class in different parts of a language area vary from each other.

Source: ne.se

Tags

Author

TTT
By TTEnglish edition by Sweden Herald, adapted for local and international readers

More news

Pandemic Face Masks Pose Environmental Risk with Microplastics and Chemicals

Pandemic Face Masks Pose Environmental Risk with Microplastics and Chemicals

Teen Girl Stabbed in Borås; Suspect of Same Age Arrested

Teen Girl Stabbed in Borås; Suspect of Same Age Arrested

Algal Bloom Crisis Worsens at British Isles' Largest Lake Lough Neagh

Algal Bloom Crisis Worsens at British Isles' Largest Lake Lough Neagh

Man in His 70s Dies in Drowning Accident at Motala Beach

Man in His 70s Dies in Drowning Accident at Motala Beach

Government Plans to Add Chickenpox Vaccine to Child Vaccination Program

Government Plans to Add Chickenpox Vaccine to Child Vaccination Program

Gang Leader Among Two Escapees from Borås Detention Center

Gang Leader Among Two Escapees from Borås Detention Center

Stolen Swedish Data from Miljödata Leaked on Darknet, Expert Warns

Stolen Swedish Data from Miljödata Leaked on Darknet, Expert Warns

How to Avoid Salmonella: Tips for Safe Egg and Food Handling

How to Avoid Salmonella: Tips for Safe Egg and Food Handling

Södermalm Explosion Deemed a Misfire by Police

Södermalm Explosion Deemed a Misfire by Police

Hackers Threaten to Release Personal Data After Miljödata Breach

Hackers Threaten to Release Personal Data After Miljödata Breach

Gotland Braces for Heavy Rain as SMHI Issues Yellow Warning

Gotland Braces for Heavy Rain as SMHI Issues Yellow Warning

Swedish Potato Farmers Face PFAS Dilemma Amid Consumer Demands

Swedish Potato Farmers Face PFAS Dilemma Amid Consumer Demands

Head-on Collision on E4 in Huskvarna Sends Five to Hospital

Head-on Collision on E4 in Huskvarna Sends Five to Hospital

Kristersson Criticizes Stegrud's Actions in Stockholm Incident

Kristersson Criticizes Stegrud's Actions in Stockholm Incident

Kristersson Invites Center Party to Join Growth Pact

Kristersson Invites Center Party to Join Growth Pact

Swedish Citizens Extradited from Dubai Arrested for Serious Crimes

Swedish Citizens Extradited from Dubai Arrested for Serious Crimes

SMHI Issues New Rain Warning for Flood-Hit Västernorrland

SMHI Issues New Rain Warning for Flood-Hit Västernorrland

Man Arrested for Murder of Elderly Man in Skurup

Man Arrested for Murder of Elderly Man in Skurup

Ebba Busch Calls for Stricter Measures Against Disruptive Demonstrators

Ebba Busch Calls for Stricter Measures Against Disruptive Demonstrators

Red-Green Coalition Uncertain Despite Opposition Lead in Polls

Red-Green Coalition Uncertain Despite Opposition Lead in Polls