Glossary

Status: Current Last updated: 2026-05-12 04:22 EDT

CHAT Format Terms

CHAT (Codes for the Human Analysis of Transcripts): The transcription format used by TalkBank. Files have .cha extension.
Main tier: The primary transcription line beginning with *SPK: (e.g., *CHI:). Contains the spoken text.
Dependent tier: Lines beginning with % that provide annotations for the preceding main tier (e.g., %mor: for morphology, %gra: for grammar).
Speaker code: Short identifier for a participant — up to seven characters from A-Z, 0-9, _, -, '; three uppercase letters is the convention (e.g., CHI = child, MOT = mother, FAT = father, INV = investigator).
%mor tier: Morphological analysis tier containing POS tags and morpheme breakdowns.
%gra tier: Grammatical relations tier containing dependency structure.
Gem: A marked segment of transcript bounded by @Bg (begin gem) and @Eg (end gem) headers. See Filtering gems.
Terminator: Utterance-ending punctuation (. period, ? question, ! exclamation) that appears at the end of main tier lines.

MLU (Mean Length of Utterance): Average number of morphemes (or words) per utterance. The primary measure of grammatical development in child language.
TTR (Type-Token Ratio): Ratio of unique words (types) to total words (tokens). A basic measure of lexical diversity.
NDW (Number of Different Words): Count of unique word types in a sample.
D statistic: Vocabulary diversity measure computed by VOCD. More stable than TTR because it accounts for sample size.
DSS (Developmental Sentence Scoring): Clinical scoring system for grammatical complexity (Lee & Canter).
IPSyn (Index of Productive Syntax): Syntactic complexity score based on the presence of specific grammatical structures.
SLD (Stuttering-Like Disfluencies): Disfluency types characteristic of stuttering: prolongations, blocks, part-word repetitions.
TD (Typical Disfluencies): Normal disfluency types: phrase repetitions, revisions, filled pauses.

AST (Abstract Syntax Tree): The typed, structured representation of a CHAT file produced by the parser. All analysis operates on the AST, not raw text.
Countable word: A word with genuine lexical content, as determined by is_countable_word(). Excludes fillers, fragments, untranscribed markers, and omissions.
NormalizedWord: Canonical lowercased word form used as frequency map key. Implements Borrow<str> for zero-allocation lookups.
Golden test: A test that compares output against the original CLAN C binary to verify parity.
CUTT (CLAN Utility Template Toolkit): The 17,926-line C framework underlying original CLAN. Replaced by the Rust framework in src/framework/.
Framework: The shared infrastructure (src/framework/) that handles filtering, command lifecycle, and output formatting for all commands.