Add tense-matched forms of formal verbs and a few phrases to the plain-language word list

[nonprofittechy]

What this changes

plain_language_replacements.yml only matches exact words. So "obtain" is flagged, but "obtained", "submitted" and "presented" are not, and suggesting "get" in place of "obtained" would read wrong anyway. This adds:

• Tense-matched forms of formal verbs, each with its own suggestion. For example obtained: [got, received], submitting: sending, expired: [ran out, ended], granted: [approved, given], provided: [gave, given], incurred: [had, owed], purchased: bought, terminated: [ended, stopped].
• Phrases the table was missing: set forth, aforementioned, in the event that, whereby.
• "employed", but only inside is employed, are employed and currently employed. The bare word would also match "self-employed", which is plain already.

Forms that are usually ordinary words or names were left out on purpose: "required", "benefits", "completed", "options", "exhibits", "deemed" (as in "Petition … to be Deemed Satisfied"), and "authorized" (as in "authorized representative").

The ALWeaver now runs this table over AI-drafted labels and screen text, so the two tools agree on which words to avoid.

Validation against published interviews

I ran the linter's style check (RuntimeOptions(style_enabled=True)) over all 180 data/questions/*.yml files from 50+ published Massachusetts AssemblyLine interviews, before and after this change, and compared the findings (scan script).

	plain-language findings
main	1555
this branch	1632 (+78, −1)

The one removed finding is in "Are you currently employed?". The new currently employed → working now replaces the old currently → now.

All 78 new findings were reviewed by hand. 72 are formal wording in text the litigant reads, where the suggestion fits. For example:

• "Do you live with a household member that is employed?"
• "So far, you have provided the following information about yourself."
• "the form of ID presented"
• "Has it been at least 7 years since the appeal period expired?"
• "What is the name of the dealership you purchased your vehicle from?"

The other 6 are false positives:

• quoted statute headings (3), such as "waiver in leases or other rental agreements prohibited";
• a quoted affidavit statement (1);
• a form section title, "Value of Documented or Anticipated Lost Wages…" (1);
• a term the author already glossed, "A judge ended (terminated)…" (1).

That's about 8%. For comparison, a random sample of 40 existing findings had about 8 false positives (~20%) of the same kinds: names and defined terms ("housing authority", "Assistance animal"), quoted affidavit and lease text, and suggestions that don't fit ("correct → exact"). So the new entries are no noisier than the table already is.

While testing I also found that a duplicate key makes the table fail to load (ruamel rejects duplicates), and the plain-language check then silently reports nothing. This change adds no duplicates (hereby and in the event of were already there, so I left them out), but a test that loads the table and fails on duplicates might be worth adding separately.

Tests

• Parametrized tests check that each kind of new entry is matched and suggests the right form.
• A test checks that "self employed" and "self-employed" are not flagged.
• pytest tests: 655 passed.

🤖 Generated with Claude Code