fst-train(1)

💖 fst-train 💖

okay bestie, gather round, because this is fst-train from the fst-train manual, and she's literally all about learning transducer weights ✨ no cap, that's the whole NAME situation right there. it's giving "teach the machine by showing it examples" energy and honestly? we're obsessed 💅

---

✨ SYNOPSIS ✨

fst-train [ options ] file [ input-file ]

so you pass a file, and optionally an input-file, and sprinkle some options in there like the little garnish on a strawberry 🍓 that's the vibe, periodt.

---

🎀 OPTIONS 🎀

• -t file — pulls in multiple transducers, and she does it the exact same way fst-infl2 does. same energy, different girlie 🫶
• -b — this one's for supervised training with disambiguated data. yes bestie, it's a whole mode 💜
• -d — disambiguates the analyses symbolically, and the deets on how are literally sitting in the man pages of fst-infl2. go read those, iykyk 📖
• -q — quiet mode 🤫 (iconic behaviour honestly, sometimes you just don't wanna hear it)

---

🌸 DESCRIPTION 🌸

okay so here's the tea: fst-train is used to learn statistical weights for the transducers transitions based on training data. that's the whole main character moment right there 👑

and the training is either unsupervised — which is the default, btw, she's the standard — or supervised, which you get with option -b. two modes, one legend.

🦋 supervised mode (-b)

in supervised mode, your input contains fully disambiguated data — that means both the surface form AND the analysis form, side by side, no mystery 💖

the format restrictions here are identical to the ones that apply for lexicon entries. and this part matters so listen up: all operators other than the colon operator (:) are interpreted literally. literally literally. no games 🌈

💫 unsupervised mode (the default)

in unsupervised mode, the input data is just surface strings. that's it, that's the whole input. and the format is identical to the input format of fst-infl and fst-infl2. same vibe, same rules ✨

🧚 where the weights go

the transducer weights get stored in files whose names you get by appending .prob to the names of the transducer files. so like, automatic naming, very slay, no manual labour required 💅

---

💀 BUGS 💀

No bugs are known so far.

and honestly? not me being suspicious of that, but okay bestie, we'll take the win 🙈✨ flawless queen behaviour fr fr

---

👑 SEE ALSO 👑

• fst-infl2
• fst-compiler

go give them a lil visit, they're all in the same friend group 🫶

---

🌈 AUTHOR 🌈

Helmut Schmid, Institute for Computational Linguistics, University of Stuttgart, Email: schmid@ims.uni-stuttgart.de

This software is available under the GNU Public License. 🌱

---

and that's a wrap, bestie!! this page came to you courtesy of the fst-train(1) manual, October 2005 edition — yes, she's an elder but she still ate 💖🎀🦋 mwah, go train some transducers, love u bye 😭👑