Guided pathThis is part of Understand Transformers & BERTBack to the path

Do we already use models of language?

P10.orientation.02 · Audience: guest, it-ml, language-pro

Here is a reassuring idea before the machinery arrives: as a language professional you already build and run models of language every day — you just do it in your head. Grammar, register, and finishing other people's sentences are all the same move a language model makes. Once you see that, the technology stops being alien. No maths. Your glossary is one click away.

🗣️ From a linguist's perspective: you are already a prediction machine

When a colleague pauses mid-sentence — “Could you pass me the …” — you can already feel the likely next words (salt, report, phone) ranked by how well they fit the situation. You are not consulting a rulebook; you are drawing on a lifetime of examples. A language model does the same thing — rank the plausible next words given everything so far — only from far more examples and with no understanding of what any of it means.

Prompt: “…stepping into the ___”How plausible is each next word?boothvery likelyroomlikelymeetingpossiblebananaalmost never
Predicting the next word: a model ranks plausible continuations from the context, then picks a likely one — it aims for plausible, not necessarily true.
Step 1 / 3Grammar is a model

A grammar is a compact description of how a language tends to behave — enough to teach with, translate with, and correct with. It does not list every sentence anyone could ever say; it captures the patterns. That is precisely what makes it a model: a useful simplification, complete with the exceptions it quietly ignores (“i before e… except when it isn't”).

When you judge that “the cat sleeps” is well formed and “the cat sleep” is not, you are running that model — applying a pattern to a new case. You do it instantly and without arithmetic. Hold on to that: modelling language does not require numbers; numbers are just how a computer ends up doing the same job.

📚 Go Further

Everyday language work as modelling — nothing technical here.

  • Register — the way word choice shifts with audience, purpose and relationship; a model of the situation. See your glossary.
  • Notice yourself predicting today: pauses you finish, autocomplete you accept or reject, a translation where you chose words to fit the setting. Each one is you running a model of language.
Ask the mentor about this module

Ask a question about this content. The mentor explains and grounds its answer in what you are studying; asking is recorded as a learning signal, not a grade.

Ctrl/Cmd + Enter to send