Most commonly, following the work of phonetician John C. Wells, a lexical set is a class of words in a language that share a certain vowel phoneme.
Originally, Wells developed 24 such labels—keywords—for the vowel lexical sets of English, which have been sometimes modified and expanded by himself or other scholars for various reasons.
A further benefit is that people with no background in phonetics can identify a phoneme not by learned symbols or technical jargon but by its simple keyword (like STRUT or KIT in the above examples).
Wells defined each lexical set on the basis of the pronunciation of words in two reference accents, which he calls RP and GenAm.
[6] Wells classifies English words into 24 lexical sets on the basis of the pronunciation of the vowel of their stressed syllable in the two reference accents.
For example, the a in the stressed syllable of tomato is pronounced /ɑː/ in RP, and /eɪ/ in GenAm, a combination that is very unusual and is not covered by any of the 27 lexical sets above.
[11] Some words pronounced with /ɒ/ before a velar consonant in RP, such as mock and fog, belong to no particular lexical set because the GenAm pronunciation varies between /ɔ/ and /ɑ/.
[9] The standard lexical sets of Wells are widely used to discuss the phonological and phonetic systems of different accents of English in a clear and concise manner.
Although based solely on RP and GenAm, the standard lexical sets have proven useful in describing many other accents of English.
For example, some Irish and Scottish accents that have not undergone the fern–fir–fur merger split the NURSE lexical set into multiple subsets.
[23] Wells has stated that he didn't include a GOAL set because this should be interpreted as an allophone of GOAT that is sensitive to the morpheme boundary, which he illustrates by comparing the London pronunciations of goalie and slowly.