The present paper derives from a PhD study investigating the nexus between Grade 4 textbook vocabulary demands and Grade 3 isiXhosa-speaking learners' knowledge of that vocabulary to enable them to read to learn in Grade 4. The paper challenges the efficacy of the four current definitions of "word" for generating high frequency words (HFW) for such a study. The paper submits that the token and type conceptualisations are unrealistic in their disregard of the learning burden principle, and that the lemma and word family notions are too accommodative and untenable in their overextension of the learning burden principle. It critiques the arbitrary generation of word levels from a language corpus which is not cognisant of the natural order in which second language learners at different levels and from different first language backgrounds acquire English vocabulary. Based on research findings, the paper proposes, for the larger study, a unit-of-word quantification broader than the token but less accommodative than the lemma. The paper advocates further research into children's psychological processing of English word forms to constitute a taxonomy of word forms which merit treatment as single words at different levels of learners' competence.