> ## Documentation Index
> Fetch the complete documentation index at: https://www.latentcast.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Building a dictionary from your list

> Turn the company and person names in a recipient list into pronunciation rules in one pass, with a record of every decision.

**Where:** Personalization ▸ Canvases, then **Generate pronunciation dictionaries** on a canvas

Typing a rule per company for four hundred recipients is not work anyone should do. This run reads
the names in your recipient list, works out which ones are likely to be said wrong, and writes the
rules for you.

<Note>
  This is an organization administrator's task. If the action is not offered to you, ask whoever
  administers your LatentCast organization.
</Note>

## The run, end to end

<Steps>
  <Step title="Set the parameters">
    Which rows to read, the language the names will be spoken in, how many terms to put in each
    dictionary, and a base name for the dictionaries that come out.
  </Step>

  <Step title="Analyse">
    Press **Analyze pronunciation risk**. The run reads the names, judges each one, and proposes a
    say-as where it thinks one is needed. Progress is shown, you can leave the page and come back,
    and you can cancel.
  </Step>

  <Step title="Read the decisions">
    Every term appears in a table with what was decided and why:

    | Decision      | Meaning                                               |
    | ------------- | ----------------------------------------------------- |
    | **Added**     | A rule was written, and the system is confident       |
    | **Attention** | A rule was written, and a person should listen to it  |
    | **Skipped**   | No rule needed, or nothing useful could be determined |
  </Step>

  <Step title="Create the dictionaries">
    The dictionaries are created **switched off**, so nothing reaches a video before someone has
    looked. A link takes you straight into the new dictionaries in the voice studio.
  </Step>

  <Step title="Review, then enable">
    Work through the flagged terms, listen to them, then enable the dictionary. See
    [Reviewing flagged terms](/docs/guides/names/reviewing-flagged-terms).
  </Step>
</Steps>

## Choosing the language

The language you choose is the language the names will be **spoken in**, which is not the same as
where the company is from. A Finnish company name inside an English script needs the English run.

If you send in several languages, run it once per language.

## The audit

Every run keeps a record, and **Download the audit** gives you the whole decision table as a
spreadsheet: every term, what was decided, and the reasoning.

Two good uses. Send it to a native speaker in that market and have them mark the ones you got wrong.
And keep it, so that in three months you can answer why a particular name is pronounced the way it
is.

Past runs stay listed with their audits, so you can compare a re-run against the last one.

## When to re-run

* You added a batch of new accounts to the list.
* You moved into a new market and are sending in a new language.
* A colleague reported a name being said wrong and you would rather re-check the whole list than
  patch one rule.

Re-running does not silently overwrite what you reviewed. New dictionaries arrive alongside, off, and
waiting for you.

## Next

<CardGroup cols={2}>
  <Card title="Reviewing flagged terms" href="/docs/guides/names/reviewing-flagged-terms">
    The next step after every run.
  </Card>

  <Card title="Pronunciation dictionaries" href="/docs/guides/names/pronunciation-dictionaries">
    How a rule works, and how to write one by hand.
  </Card>
</CardGroup>
