தேடல் · Thedal · retrieval
தமிழ் இலக்கணம் அறிந்த தேடல்
Search that knows Tamil inflects
தமிழ் ஒட்டுநிலை மொழி. ஒரு பெயர்ச்சொல் பல வடிவங்களாக மாறும் — கோயில், கோயிலில், கோயிலின், கோயிலுக்கு, கோயில்கள். சரியான சொல்லை மட்டும் தேடும் முறை இவற்றில் பெரும்பாலானவற்றைத் தவறவிடும்.
Turn the toggle off and on: the same query over the same corpus, matched as an exact string and then matched across inflected forms. The gap between the two numbers is why searching Tamil archives mostly fails today.
0 found · exact string alone would find 0
ஏற்றுகிறது…
இது எப்படி வேலை செய்கிறதுHow this works
Tamil inflects by suffixing, so every inflected form of a noun begins with the noun — with two systematic exceptions this handles explicitly. When a case ending attaches, the stem loses its final pulli (கோயில் + இல் = கோயிலில், so the literal word is not a prefix of its own inflection). And nouns ending in ம் change it before an ending: oblique மரம் → மரத்தில், plural மரம் → மரங்கள், neither of which shares a prefix with மரம் at all.
A suffix-stripping stemmer was tried first and discarded: it turns கோயில் into கோய, because the root itself ends in the letters that spell the locative. Guessing where a suffix ends needs the morphology, and the morphology is not a list.
This is prefix matching with two language rules, not a morphological analyser. It over-matches at the edges — searching தமிழ் also finds தமிழகம், which is தமிழ் + அகம் and arguably what you wanted. For search, recall is worth more than precision at that margin; for anything that must be exact, it would not be.
Corpus: Iyal AI cleaned Tamil corpus, filtered to documents discussing கல்வெட்டு, each kept with its licence. Runs entirely in the browser against a static index — no server, works offline. Target corpus is the transcribed inscriptions in South Indian Inscriptions; this demo does not contain inscription text and does not pretend to.