Disfluency and revision patterns
Maze Analysis for Language Samples
Maze analysis separates disfluent or revised material from the rest of the language sample. ConductSpeech counts maze words, mazed utterances, maze words per 100 words, and categories such as filled pauses, repetitions, revisions, and abandoned starts.
This matters because maze material can reveal planning and fluency patterns, but it should not inflate MLU, word totals, or morpheme totals.
Sample result
Maze Analysis for Language Samples
Maze words
1
Mazed utterances
1
Rate
2.56 maze words per 100 words
Summary
Maze words and mazed utterances
Categories
Filled pause, repetition, revision
Rate
Maze words per 100 words
Counts
Excluded from MLU totals
Summary
Maze words and mazed utterances
Categories
Filled pause, repetition, revision
Rate
Maze words per 100 words
Counts
Excluded from MLU totals
How it fits into a speech workflow
1
Collect
Start from a recording, transcript, or saved session.
2
Review
Check speaker turns and make clinical edits before relying on results.
3
Measure
See the language measures and notes that matter for this feature.
4
Use
Bring the output into reports, progress review, or research exports.
What counts as a maze?
Mazes include words or phrases that the speaker abandons, repeats, revises, or uses as filled pauses. In SALT-style transcription, many mazes are written in parentheses so the intended utterance remains readable.
- `(um)` marks a filled pause.
- `(he was) he was running` marks a repetition.
- `(I goed) I went home` marks a revision.
- `(I wanted to)` marks an abandoned start.
Why mazes stay separate
If maze words were counted as ordinary words, MLU and other language metrics would be harder to interpret. ConductSpeech keeps maze metrics visible while excluding coded maze material from the core word and morpheme totals.
How clinicians use it
Maze summaries can support observations about fluency, organization, self-correction, and planning load. The categories help distinguish a student who uses many filled pauses from a student who repeatedly revises sentence starts.
What users see
Maze summary output
A compact result view turns the feature into reviewable language, not a technical readout.
Maze words
1
Mazed utterances
1
Rate
2.56 maze words per 100 words
Category
Filled pause
Clinical interpretation notes
- Maze classification is most reliable when the transcript uses clear parenthesized coding.
- Clinicians should review maze-heavy samples before using the report language.
Related pages
SALT Transcript Editor
Edit AI-drafted transcripts with SALT-style codes, review suggestions, and re-analyze language sample metrics instantly.
Clinical Language Sample Reports
Generate IEP-ready language sample reports with MLU, PGU, SI, C-units, maze summaries, norms, and fluency context.
SALT-Compatible Language Sample Analysis
AI language sample analysis with SALT-style coding, C-units, SI, mazes, grade norms, reliability, and clinical reports.
SALT-compatible analysis methodology
Read how ConductSpeech documents conventions, validation, and limitations.
Ready to try it
Start with a real language sample.
Create an account, upload or review a sample, and see how this feature appears inside the ConductSpeech workflow.