About the Framework

How MetaWriteAI relates to Hyland’s interpersonal model, and what the platform adds on top of it.

MetaWriteAI is grounded in Hyland’s interpersonal model of metadiscourse. For pedagogical and computational purposes, the platform further operationalises the ten major categories into functional and linguistic subtypes. These extensions are explicitly identified as MetaWriteAI operational categories rather than original categories proposed by Hyland.

Three levels of annotation

Level 1Hyland category

One of the ten major metadiscourse resources: transitions, frame markers, endophoric markers, evidentials, code glosses (interactive); hedges, boosters, attitude markers, self-mention, engagement markers (interactional).

Interactional → Hedges

Level 2MetaWriteAI functional subtype

An operational, function-based subdivision used for teaching and for machine annotation. These subtypes are MetaWriteAI extensions, not categories proposed by Hyland.

Hedges → Possibility

Level 3Linguistic realisation

The form that carries the function in this text: a modal verb, an epistemic lexical verb, an adverb, a prepositional phrase, a clause pattern.

Modal verb → may

Contextual evaluation

A lexicon proposes candidates; context decides the classification. Every annotation therefore carries a contextual judgement alongside its three levels.

may

Hedges → Possibility → Modal verb

Function in context
qualifies a causal interpretation
Appropriateness
appropriate
Confidence
0.94

Overlap and double counting

Forms such as according to, must, should, see, we and clearly — and multifunctional Arabic forms such as و، فـ، بل — appear in more than one lexicon. A candidate is assigned to a single primary category unless genuine multifunctionality is analytically justified, in which case the secondary reading is recorded with its own confidence value.

Reference

Hyland, K. (2017). Metadiscourse: What is it and where is it going? Journal of Pragmatics, 113, 16–29. https://doi.org/10.1016/j.pragma.2017.03.007