Glossary
These terms are used consistently throughout the documentation. The point is
not to be exhaustive but to avoid calling the same thing two names on two
pages.
Library concepts
| Term |
Meaning |
| feature |
A single number extracted from a text (ttr, atesman) |
| key |
A feature's name; the dict key in the returned mapping |
| group |
One of the 13 sets features belong to (lexical, readability…; 14 with custom_ngrams) |
| citation |
The short source pointer — the parenthetical in a methods section |
| reference |
The full bibliographic record — what goes in a bibliography |
| registry |
The single place all feature metadata lives |
| pipeline |
The chain of steps from raw text to numbers |
| scale |
The kind of value: ratio_0_1, score, count… |
| segment |
A slice of text produced by segment_text |
| label |
The folder name in a corpus; author, class or experimental arm |
Measurement terms
| Term |
Meaning |
| word |
A whitespace-separated piece with edge punctuation stripped, containing a letter or digit; the word the library counts |
| sentence |
A unit ending in . ? ! …; : only when a new sentence follows |
| spaCy token |
The unit the model splits text into; punctuation is a token of its own. The dependency features count these |
| lemma |
A word's dictionary form; from Zeyrek for Turkish, from spaCy for English |
| POS tag |
One of the 17 UD part-of-speech tags (NOUN, VERB, ADJ…) |
| n-gram |
A sequence of consecutive words; counted with custom_ngrams |
| type |
One of the distinct words in a text |
| token |
In the type-token ratio: one of the word instances in a text |
| type-token ratio (TTR) |
Number of types ÷ number of tokens |
| hapax legomenon |
A word occurring exactly once in the text |
| stroke |
Every non-space character — the input to ARI |
| window |
A run of consecutive units measured at a time |
| threshold |
The boundary value that triggers a classification |
| percentile |
nth percentile: n% of the values are equal to it or smaller |
| calibration |
Deriving a threshold from data rather than picking it |
Verification terms
| Term |
Meaning |
| exact |
Within tolerance of the source's number (✅) |
| documented deviation |
A difference whose reason is written down (🟡) |
| mismatch |
An unexplained difference (❌) |
| end-to-end |
Tested by pushing the source's own text through the pipeline |
| formula |
Tested by feeding inputs directly to the function |
| as cited in |
The primary source could not be obtained; taken from a secondary |
| release gate |
A condition that blocks a release if unmet |