Suffeffix Research · v0.2 · the T.H.E.F.T. framework
One meaning. Five languages. Two families.
Suffeffix is an explainable lexical knowledge graph. It takes words apart — stem, affix, function — lines them up across Telugu, Hindi, English, French and Tamil, and traces where they came from, with a source for every etymology and a trace for every explanation.
Meaning
the state of being goodSTATE_OF(GOOD)
Hindi
Telugu
Tamil
- 564
- words
- 115
- aligned meanings
- 151
- affixes
- 50
- semantic atoms
- 127
- sourced etymology edges
- 0
- model calls
Every record in v0.2 is a draft, authored against reference works and awaiting expert review. The meanings were chosen to showcase derivation and etymology, so these counts describe the dataset, not the languages.
The framework
T.H.E.F.T.: inherited inside a family, stolen across it
Telugu and Tamil show one family giving two answers to Sanskrit. English and French show one long loan. English, French and Hindi are distant cousins. Between the two families the validator allows only one kind of edge: borrowing.
Read about the frameworkResearch
Technical report · v0.2 · October 2026
Suffeffix: an explainable lexical knowledge graph for morphology, semantic decomposition, etymology and affix alignment
How we model five languages across a family boundary the data refuses to blur: the validator rejects cognates between Dravidian and Indo-European, every etymology carries a source and a confidence, and explanations are generated by named rules rather than by a language model.
Read the reportExplore
Four ways into the graph
Concordance
Every meaning with its word in each language side by side, filterable by the affix function it uses.
Affix Atlas
Every affix filed by the job it does. Here is the privative row — the affixes that mean “without”.
Etymology
Sugar left Sanskrit three ways. Colour is family; every change of colour is a theft.
Semantic atoms
Fifty atoms, and not one more. An engineering interlingua, not a theory of cognition. Filled cells are NSM primes.
Browse the atomsPrinciples
A knowledge graph that shows its working
Explainable by construction
Every explanation is assembled from graph facts by a named template rule, and every sentence opens to the nodes, rules and sources behind it. There is no language model in the loop.
Honest about uncertainty
Each fact carries an epistemic status — established, engineering, hypothesis — and each etymology a source and a confidence. Where scholarship disagrees, the edge is stored contested and left unresolved.
Respectful of two families
Telugu and Tamil are Dravidian; English, French and Hindi are Indo-European. The validator rejects any cognate or inheritance edge across that boundary, so the resemblances that remain are modelled as the borrowings they are.
Build on it
Cite it. Download it. Correct it.
The dataset is CC BY-SA 4.0 and the code is Apache-2.0. Every record is a draft — if you can review one against a reference work, or spot an error, we want to hear it.
@techreport{suffeffix2026,
title = {Suffeffix: an explainable lexical knowledge graph for morphology,
semantic decomposition, etymology and affix alignment},
author = {{Suffeffix contributors}},
institution = {Suffeffix},
type = {Technical report},
number = {v0.2},
year = {2026},
month = oct,
url = {https://suffeffix.com/research/suffeffix-v0-2/},
note = {Dataset records are drafts pending expert review}
}