AI-for-Education/English-Linguistic-Knowledge-Benchmark

English Linguistic Knowledge Benchmark

A 20-item public sample of the English Linguistic Knowledge Benchmark, measuring how much an LLM knows about the linguistics of English (phonics/ orthography, morphology, syntax, vocabulary, and phonological awareness), rather than its ability to converse in the language.

This is a curated sample, not the full item set: the full 137-item benchmark (99 multiple_choice, 38 short_form) is kept private to avoid leaking into future models' training data. The full dataset is available on request. Items here were selected to spread across question domain, answer format, and difficulty.

Domain and grade definitions

Domain (fixed, language-agnostic -- every language's items are filed against the same 9 domains) and grade (the content level a fact is first introduced at, not the question's difficulty) definitions, side by side:

Domain	Covers		Grade	Content level
D1	Phonics/Orthography		G0	Reception / below Year 1
D2	Phonology		G1	Year 1
D3	Nominal morphology		G2	Year 2
D4	Verbal morphology		G3	Year 3
D5	Syntax		G4	Year 4
D6	Semantics/Lexicon		G5	Year 5
D7	Pragmatics/Discourse		beyond_primary	Above Year 5
D8	Sociolinguistic/Register
D9	Metalinguistic reasoning

Columns

Column	Description
question_id	Question identifier in the full benchmark
domain	Domain code (D1-D9) -- see Domain and grade definitions above
subdomain	Free-text sub-topic within the domain
grade	Grade code (G0-G5, or beyond_primary) -- see Domain and grade definitions above
answer_type	multiple_choice (answer is a single letter A-D) or short_form (1-3 words)
question	Question text (MCQ options are a separate options field)
options	{"A": ..., "B": ..., "C": ..., "D": ...}, present for multiple_choice only
expected_answers	List of acceptable answers (short_form can have more than one)
linguistic_feature	The specific linguistic point the question targets
svr_strand	Simple View of Reading strand the item serves: decoding, language_comprehension, or outside_flr_construct
geeap_component	GEEAP core instructional skill an error in this item would corrupt: systematic_phonics, phonological_awareness, oral_language, reading_fluency, reading_comprehension, writing, or none
sources	Source materials the question/answer is grounded in
grade_placement_evidence	How the grade assignment was justified in the pipeline

Source & methodology

Built by The Quality Assurance Facility for AI in Education.