Data science is built on the ability to identify items precisely, using numbers. Books, for example, all have an ISBN so that particular editions can be found in bookstores and libraries worldwide. A great challenge for informatics is ascertaining when things are the same across systems - whether the goods leaving the supplier are the same as the goods arriving at the warehouse. Sometimes pieces of information can link databases effectively; "July 19, 2010" will always refer to the same moment in history, even if systems render it variously as "19/7/10" or "19 juillet 2010". Words, however, are clouds that do not have meanings fixed in a standard system. A 🐕 dog could be "a mammal, Canis lupus familiaris, that has been domesticated for thousands of years"
, or "a domestic mammal, related to wolves and foxes, that is often kept as a pet"
, making it impossible to automatically tie together information in different systems that try to reference the same concept.
The problem is further compounded when expressing the same basic 💭 concept in different languages, since the shapes
of both the terms and the words used to define them are inherently different computationally, even if the idea is identical.
For example, 👂 is written in English as "ear", but the only fact the computer knows is the binary code for e-a-r, "01100101 01100001 01110010". In Romanian, 👂 is written as "ureche", but the computer sees "01110101 01110010 01100101 01100011 01101000 01100101". For the computer to know that two terms are equivalent, that 👂 in one language = 👂 in another, we need a set of digits that is the same for any term that expresses the same thought.
Kamusi is implementing a Universal Concept Identifier, a single number that can be assigned to a given idea. Any term that matches that idea - 🐘, éléphant in French, ndovu in Swahili, ゾウ名 in Japanese - is joined to that ID. Using our differentiator
, we can split similar things with different numbers where appropriate, yet still show their close ontological relationships
- a freeway and a thruway get different numbers, but are both linked through our graph architecture to the idea of limited access auto routes, and the German
translation "autobahn". We will use the json "synset" numbers WordNet
has established as a starting point, and integrate with the IDs created in the limited expanded vocabulary set proposed for CILI, the Collaborative Interlingual Index.
UCIs will be extended through Kamusi to cover millions of additional concepts that fall outside of the 100,000 concepts identified by WordNet, including words other than nouns, verbs, adverbs, and adjectives, and untreated forms such as gerunds, 650,000 additional concepts gleaned from Wiktionary
, millions of named entities
from the Joint Research Council
, 1.6 million species
from the Catalogue of Life
, 8 million domain-specific terms
in 25 languages from IATE
as well as term sets from additional sources, and items indigenous
to languages other than English, . The UCI will thus be available to codify open data across numerous languages, projects, and data systems, with the intent that the world's linguistic data can play together in ways that are not currently possible when links can only be inferred by guesswork based in spelling.