Writing subskills are the distinct mental operations a writer performs while composing: generating ideas, organising them into a structure, turning them into sentences, and reviewing and repairing what has been produced. They are operations a writer recruits and recombines, not a fixed sequence of lesson stages. A skilled writer planning an email and a skilled writer drafting a research article draw on the same underlying repertoire, weighted differently for the task.
The modern account of these operations comes from Linda Flower and John Hayes, whose 1981 College Composition and Communication paper "A Cognitive Process Theory of Writing" reframed writing as goal-directed problem solving rather than a linear march from outline to fair copy. Working from think-aloud protocols, they identified three principal processes: planning (generating ideas, organising them, and setting goals), translating (converting that meaning into the linear string of written English), and reviewing (evaluating and revising). The decisive claim was recursion: writers loop back constantly, so revision is not a final stage but something that can interrupt and redirect planning at any moment.
Hayes substantially revised the model in 1996. Planning became one part of a broader reflection module; translating was renamed text production to credit its active role in generating content; and revision was recast not as a separate process but as a recombination of more basic operations (reading one's own text, reflecting, producing new text). Two additions matter most. Working memory moved to the centre, characterised through Baddeley's multicomponent model rather than as a single undifferentiated resource that processes merely compete for. And motivation and affect were integrated alongside cognition, with the writer's goals, beliefs, and cost-benefit estimates feeding bidirectionally into the task. The upshot for teaching is that a writer juggling content, organisation, and sentence form is loading a limited working-memory system, which is why doing everything at once is so hard for learners.
Several operations sit beneath these broad processes. Idea generation supplies content, whether through brainstorming, drawing on reading, or recalling experience. Discourse-level organising arranges that content into a shape a reader can follow: ordering paragraphs, sequencing arguments, signalling the relationship between parts. Controlling cohesion and coherence operates at the seams, using reference, conjunction, and lexical links so sentences hang together and the whole text reads as a unified line of thought rather than disconnected statements.
Above the sentence sits control of register and genre: matching formality, stance, and conventional structure to the purpose and audience, so a complaint letter does not read like a text message and a lab report follows the moves a reader of lab reports expects. This is where the social dimension foregrounded by a genre-based view enters the subskill inventory. Below the sentence sits grammatical and lexical accuracy (choosing the right words and forming structures correctly) and mechanics: spelling, punctuation, capitalisation, and layout. Mechanics are low-status but unforgiving, since errors here are the most visible to a reader and the quickest to undermine credibility.
Because working memory is finite, writers cannot give full attention to every subskill at once. A learner who fixes all available capacity on grammatical accuracy tends to produce halting, impoverished text; one who writes fluently to get ideas down often leaves errors uncorrected. Ken Hyland, in Second Language Writing (2003), notes that writing tasks are demanding precisely because they require simultaneous attention to content, coherence, fluency, and accuracy. The practical resolution is to separate the operations in time rather than expect them in parallel: generate and draft for fluency first, then return to accuracy in editing and revising. This is the rationale behind the staged drafting of process writing, and it explains why premature error-correction can stall composition. Crucially, separating operations for instruction does not mean the operations themselves are sequential; the writer still loops.
The cognitive-process tradition has been challenged for treating composing as a property of the individual mind, abstracted from the social and cultural setting in which texts are written and read. Hyland surveys five orientations to writing instruction (structure, function, expressivism, process, and content), and the genre and social-constructionist strands argue that what counts as good organisation, register, or even a "good idea" is community-specific, not a neutral cognitive universal. A second caution: the process model was built largely from skilled L1 writers thinking aloud in English, and second-language writers face added burdens (a smaller lexical store, slower retrieval, transfer from a first language) that a purely cognitive account can understate. The subskill inventory is a useful map, but it describes operations more cleanly than any real writer performs them.
Treat the subskills as operations to develop, not stages to march through. Design tasks that isolate one operation so learners can give it full attention: an idea-generation task with no demand for accuracy, a reformulation task focused only on cohesion, an editing task on a text whose content is already fixed. Sequence fluency before accuracy within a drafting cycle so composing is not strangled by premature correction, and reserve close attention to mechanics and grammar for a dedicated editing pass. Name the operations so learners build metacognitive control over their own composing, and remember that organisation, register, and genre conventions are learnable knowledge, not innate taste, which makes them legitimate objects of explicit instruction rather than something only "good writers" possess.