4 articles tagged “training”.
Armenian is printed in two orthographies with the same alphabet, and a book's title is no guide to which one its pages use. One letter is: the reform left ւ only one place to stand, so counting it anywhere else tells the two apart, on a page or in a single name
Our held-out registry knew every evaluation page and refused to train on any of them. It knew nothing about the other pages that print the same text: a commentary quoting the passage it annotates, a variant in another volume, an encyclopedia quoting a chronicle. An eight-word shingle check found 25 of them in the harvest, and nothing had failed
Two days after shipping v3 we built the evaluation that showed what it could not do: read anything but the encyclopedia it was tuned on. This is the story of v5 — a corpus three times wider, a filename bug that taught a model to hedge, and the first version to read Western Armenian at all
Three training runs after a model that read a quarter of the words on a page, our Apache-licensed Armenian recogniser reads three quarters of them — and most of what we learned on the way is about Armenian itself: letters with Latin twins, a full stop that looks like a colon, and punctuation that taught our model its own bad habits