NirvanLabs
NirvanLabs studies how languages are kept and changed by the people who use them, and builds software that works from the language a community speaks.
No single author
Language is the one invention that the whole of humanity has worked on. No one designed it and no one is in charge of it. Every language spoken today is the combined work of the people who have spoken it, over thousands of years, and none of them planned the result.
A language has no finished state. There is no complete edition of Dutch, English or Hindi held somewhere and handed out to new speakers. There is the language as it is used today by the people using it, and there is the record of how it was used before.
Changed by use
Language has never stopped changing, and it changes through use. A word is used in a slightly new way in one household, one workplace, one town. If the new use serves the people there, it is repeated, and after a while it is what the word means in that place.
This happens first in ordinary speech, in particular places, among particular people. Dictionaries and grammars record the change afterwards, if they record it at all. Two towns a short drive apart can use the same word in different ways for a generation, and neither of them is using it wrongly.
Who keeps a language
The community that uses a language is the one that keeps it alive. A word means what that community has settled on. That settlement belongs to a place and a time. In the next region, or twenty years later, it may be different.
A standard variety is useful. It lets people who will never meet write to one another. It is still one variety among many, and much of everyday speech happens outside it.
Software and the language in use
Much of what a community writes now passes through software. Software suggests words, corrects them, translates them and replies in them. Most of it works from one standard variety for each language. A spell checker marks a regional word as an error. A translation takes the usage of the capital. A generated reply is phrased in a way people in the reader’s town would not phrase it.
Software used by a community should work from the language that community uses, in the place where it uses it. That requires a description of how the language is used in each region, built from what people there write.
The work
NirvanLabs works on how language is used and how it is understood. Its two products are Terrior and Kithe.
Terrior
terrior.app
Terrior localises software at the level of a region rather than a language. For each region it builds a regional language profile: a structured description of how people there use a language.
A profile records which of several competing words the region uses, and how often it uses a borrowed word where a native one exists. A profile is a document rather than a trained model. When text produced from it reads wrong, the fault can be traced to one entry and that entry corrected.
How a Terrior profile is built and tested
Findings from German and Hindi
Kithe
Kithe is a social platform. It is built around four problems with current platforms: people are misunderstood, the context of what they said is lost, their earlier posts stay fixed after their views have moved on, and access and reach are uneven, with some accounts privileged over everyone else.
The lab
- Address
- Binnendongenstraat 4L
5707 TA Helmond
the Netherlands - Chamber of Commerce
- KvK 85770833
- Correspondence
- hello@nirvanlabs.eu
- Products
- terrior.app, kithe.social