Skip to main content

Language

See what you are really saying.

Seymour reads the words a website uses about disability, gender, race and age. It finds candidates, records their context, and hands them to a person.

How it works

Rules, not guesses.

  1. ExtractPull the visible text from the captured page.
  2. DetectMatch terms and patterns. No AI, so it is repeatable.
  3. ClassifyRecord the context around each match.
  4. ReviewA person confirms, rejects or amends.

What it looks for

Three families live. Age is not configured yet.

Examples show wording Seymour flags and the current language it expects instead.

terms
164
patterns
365
categories
29
  • Disability

    75 terms, 198 patterns, 7 categories.

    Seymour flags

    • wheelchair bound
    • special needs
    • suffers from
    • differently abled
    • deaf and dumb

    Current language

    • people with disability
    • disabled people
    • wheelchair user
    • access needs
    • Person-first and identity-first are both current. Neither is flagged.
    • A capital-D “Deaf” (the community) is kept distinct from lower-case “deaf”.
  • Gender

    42 terms, 83 patterns, 10 categories.

    Seymour flags

    • mankind
    • manpower
    • chairman
    • both genders
    • “guys” for a mixed group

    Current language

    • chairperson
    • workforce
    • flight attendant
  • Race, nationality and ethnicity

    47 terms, 84 patterns, 12 categories.

    Seymour flags

    • half-caste
    • ATSI
    • illegal alien
    • hard to reach

    Current language

    • First Nations peoples
    • Aboriginal people
    • Torres Strait Islander peoples
    • Umbrella terms such as CALD are tracked, not failed.
    • Where jurisdictions differ, the result is “unknown”, never a pass.
  • Age

    Not configured

    • Not configured yet. It is always reported as “not configured”, never as zero.

Context

The same words, read in place.

Seymour records the context of every match. It never scores it.

  • Own voice

    The organisation speaking for itself. This is what matters most.
  • Quotation

    Someone else’s words, quoted. Sent to review, not judged.
  • Historical

    Describing the past, such as an old name or a law.
  • Legal title

    A statute, programme or organisation’s official name.
  • User-generated

    Words written by the public, such as a comment.
  • Guidance

    Explaining which words to avoid. Recorded as not relevant.
  • Unknown

    Seymour could not tell the context, so it goes to a person.

Good and not so good

Both are counted, separately.

Good language

Current terms get their own tally. They show what an organisation does well and never cancel out a problem.

Candidates to review

Flagged terms become candidates, each with an evidence snippet, its context and, where one exists, a preferred alternative.

A match is not an accusation

A match means a person should look. Nothing is published as a finding until it has been reviewed.

Severity, proposed

Severity is not applied yet.

The rule source holds a default severity for each rule. Seymour does not assign severity or a pass or fail to a match today, and no combined score is approved.

  • Critical
  • High
  • Medium
  • Low
  • Informational

Counts always travel with their denominator, the number of pages analysed.

Curious what your language says?

Tell us your sector and we will talk through what Seymour can show.