There is a universal trend to reduce the size of commonly used lexical items and proper names. For instance, English names are sometimes shortened to monosyllables to show intimacy, as in Pat < Patricia, Pete < Peter; commonly used words are clipped, e.g. lab < laboratory, prep < prepare. The process by which the number of syllables in a word is reduced is termed truncation.

The truncation of words is a productive and impromptu process which generates a variety of shortened words, the loose restriction on the process allows possible truncated forms for the same word. Despite the great flexibility in the truncatory process, there is always a minimal size beyond which a word cannot be further truncated.

Past literature and studies suggested that the minimal size is determined by the prosodic structure, and it has been proposed that the size of the minimal word matches exactly the size of a foot (F) (McCarthy and Prince 1986). Later formulations suggest that the size of a word, as well as other morphological categories (such as stem, root and affix) has a minimal size which can be described in terms of the phonological hierarchy (Prosodic Word – Foot – Syllable – Mora) (McCarthy and Prince 1995). Such correlations are, however, not well-established in some languages (Downing 2006). Cantonese is an example where the phonological categories do not correspond to minimal words and minimal stems.

The thesis aims to present and describe the Cantonese truncation data, collected from a corpus of English loanwords in Hong Kong and a spoken Hong Kong Cantonese Corpus (HKCanCor); and use the data to testify the general truncation theories. The data show (1) a discrepancy in size between verbs and nouns, and between nouns and stems, (2) sub-minimal (i.e. monosyllabic) verb and common noun truncations, and (3) words with high frequency are shorter. These peculiarities suggest that word size in Cantonese is sensitive to syntactic position and frequency information of the word. Local morphological information does not in itself determines the truncated size of a word.

It is proposed that size restrictions in Cantonese truncations can be explained by Duanmu (2007)’s information-stress principle which states that well-predicted elements should receive less stress, and Zipf’s law (1965), which suggests that the form of frequently used words will become more economical. An examination of Cantonese loanword truncation suggests that information correlates with word length. A word is best to have a length which correlates with its information load in an utterance. A stressed position is more likely to be disyllabic and lengthened while an unstressed position is more likely to be monosyllabic and shortened. An optimal word length can then be defined as the size of a word, with respect to its word class (or each syntactic position) which best matches with its information load.