When p became h
Kannada milk is haalu and Tamil milk is paal. That is not a coincidence, it is a regular sound change, and it is the best tool anyone has for telling an inherited Kannada word from a borrowed one, or from an accident.
Line up one word across the Dravidian family and something odd happens. Milk is paal in Tamil, paalu in Telugu, and haalu ಹಾಲು in Kannada. Ten is pattu, padi, and hattu ಹತ್ತು. Tooth is pal, pallu, and hallu ಹಲ್ಲು. Tiger is puli, puli, and huli ಹುಲಿ. Pig is panri, pandi, and handi ಹಂದಿ. Name is peyar, peru, and hesaru ಹೆಸರು. Two languages keep a p and the third keeps swapping in an h, over and over, in words that have nothing else in common.
That is not Kannada being erratic. It is a regular sound change, and its direction is not in doubt. Word-initial p in older Kannada became h, passing through an f-like stage on the way. Old Kannada texts are written with the p still in place: paal, pattu, posa. The modern language has haalu, hattu, and hosa ಹೊಸ. The shift belongs to the passage from Old Kannada to Middle Kannada, and while the exact centuries are argued over, it is finished in the standard modern language. Regular is the important word. It did not happen to a handful of words that felt like it; it happened to the class.
Regularity is what makes it useful, because a regular change is a timestamp no borrowing can fake. A word already in Kannada when the shift came through was carried along and shows h today. A word that entered later, or that came in through the written language rather than the spoken one, kept its p. So an h in Kannada standing opposite a p in Tamil and Telugu is a certificate of inheritance, and this is the single most reliable test in the whole dictionary for telling native Kannada stock from a later loan.
Watch it work on a word that exists twice. Kannada has hoovu ಹೂವು for flower, and its relatives are Tamil pū and Telugu puvvu: inherited Dravidian, in the language before the shift, carried through it. Kannada also has pushpa ಪುಷ್ಪ for flower, and it keeps its p because it is a Sanskrit word taken whole from the written language, and the change ran through the inherited vocabulary rather than the learned borrowings. Two words for one thing, and their first letters give away their separate histories without a dictionary being consulted.
Tooth makes the same point and finishes it. Kannada hallu ಹಲ್ಲು is native, with Tamil pal and Telugu pallu beside it. Kannada danta ದಂತ is Sanskrit danta, taken whole. Hindi daant दाँत is that same Sanskrit word, inherited and worn down rather than borrowed. So Hindi daant and Kannada danta are genuinely connected, two treatments of one Sanskrit item, while Hindi daant and Kannada hallu are not connected at all. The pair a learner never notices is the real one. Flower runs the same way, although the Hindi side is less tidy: phool फूल is traced to Sanskrit, with more than one candidate proposed and the choice not settled, which is the sort of thing this project prefers to say out loud rather than round off.
The h shift also manufactures accidents, because Hindi has plenty of h words of its own and no relationship to any of these. Kannada haalu ಹಾಲು is milk; Hindi haal हाल is condition or state, and it is Arabic. Kannada saala ಸಾಲ is a debt; Hindi saal साल is a year, and it is Persian. Kannada aata ಆಟ is a game, from the verb aadu, to play; Hindi aata आटा is flour. Kannada kai ಕೈ is a hand; Hindi kaī कई is several. Four pairs, four coincidences, no connections whatsoever. Languages build tens of thousands of words out of a few dozen sounds, so collisions like these are not merely possible, they are guaranteed, and a dictionary that cannot resist them will be wrong constantly.
Then there are the cases nobody has settled, and this project keeps them unsettled on purpose. Kannada meenu ಮೀನು is a fish, and Sanskrit has mīna, the word behind Hindi meen मीन. A Dravidian source for the Sanskrit word has been proposed, the resemblance has been discussed for well over a century, and no account is agreed. Kannada akka ಅಕ್ಕ, elder sister, sits beside a Sanskrit akkā, and nursery words for family members are among the least reliable evidence in the field, because small children in unrelated languages make similar noises. Bhashegalu marks entries like these uncertain and stops there. A missing answer is honest; a tidy wrong one is a claim.
Out of all this falls the small field guide this project actually uses. First, look for family: does the word have relatives inside its own family, joined by a regular correspondence like the Kannada h? Relatedness is never two words sounding similar. It is whole classes of words changing in step. Second, look for a route: is there a documented channel, a Sanskrit register, a Deccan court, an English classroom, and does the word’s shape fit that channel, like the -u on kharchu or the -e on shaale? Persian on Kannada ground and The grammarian who named the layers walk through the two oldest channels in detail. Third, respect coincidence. Fourth, when none of that settles it, say so and stop.
Haalu and haal can share a syllable. They cannot share a history, and no amount of squinting will graft one onto the other. The good news is that the truth is better than the illusion. A false cognate is a dead end, but a borrowed word with a real route behind it carries actual history, courts and classrooms and coastlines, into ordinary speech. This dictionary would rather hand you one true story than ten plausible ones. Corrections and counterexamples are welcome, and from Kannada speakers doubly so.
