Meaning Without a Home Language
Identifiers, not translations
This event is part of a demonstration of what Lera offers for running knowledge events. Its content is written with care and draws on real sources, but the conference has not taken place, and apart from the convener its participants are fictional.
In Wikidata, Douglas Adams is ‘Q42’. The identifier belongs to no language: the English, Persian or Amharic labels are only names hung on it. Since Wikidata was launched in 2012, millions of subjects have been given identifiers in the same way. Denny Vrandečić and Markus Krötzsch (2014) described it as a free, collaborative knowledge base. I want to draw attention to an aspect of it that is less often noticed: an identifier separates meaning from the language in which it was born.
This separation is no small achievement. Most of our information systems still treat each translation as a separate page, loosely linked to an ‘original’. The English text is updated, the Persian text goes stale, and no one knows which version makes which claim. In 2016 the FAIR principles (findable, accessible, interoperable and reusable) made a unique and persistent identifier the first condition for data to be findable. My proposal is that we take the same condition seriously for knowledge itself, not only for datasets: every claim, every concept and every text should have an identifier that remains independent of language.
But an identifier is not neutral. If we look at Wikidata’s labels, a large number of items have an English label and no Amharic or Persian one. The properties and classes themselves often derive from the categories of the English Wikipedia. In other words, the identifier has no language, but the structure in which identifiers are connected still has an accent. The Abstract Wikipedia project (Vrandečić 2021) takes a bolder step: to hold content in an abstract form and generate the text of each language from it. Its promise is great, and so is its risk. If the grammars of low-resource languages are written by people who do not know those languages, the voice of those languages will once again be made from outside.
So two things have to be kept apart. The anchor of meaning should have no language: an identifier that no language owns. But the right to name, that is, to write the label, the definition and the explanation, must remain with each language community. The Persian definition of a concept should not be a translation of the English definition; it should be an equivalent representation of the same object, with its own author and its own versions.
This is where the Language-Agnostic Knowledge framework begins: a knowledge object has a stable identity, and translations are representations of that same object, not copies that drift away from it. In this talk I show what this distinction means in the design of a system: from what a reference actually points to, to how a translation knows that it has become out of date when the original changes.
Questions for discussion
- Is a language-free identifier possible, or does every identification scheme carry the accent of its makers?
- Who should have the right to write the definition of a concept in a given language?
- Does generating text from abstract content empower low-resource languages, or shape them from outside?
References
Vrandečić, Denny, and Markus Krötzsch. 2014. “Wikidata: A Free Collaborative Knowledgebase.” Communications of the ACM 57 (10): 78–85.
Vrandečić, Denny. 2021. “Building a Multilingual Wikipedia.” Communications of the ACM 64 (4): 38–41.
Wilkinson, Mark D., et al. 2016. “The FAIR Guiding Principles for Scientific Data Management and Stewardship.” Scientific Data 3: 160018.
Berners-Lee, Tim, James Hendler, and Ora Lassila. 2001. “The Semantic Web.” Scientific American 284 (5): 34–43.