Forms and Meanings of Lexical Reduplications in Cantonese: a corpus study

Forms and Meanings of Lexical Reduplications in Cantonese: a corpus study Charles Lam Department of English The Hang Seng University of Hong Kong Shatin, N.T., Hong Kong charleslam@hsu.edu.hk Abstract Lexical reduplications (LR) in Cantonese are multiword expressions (MWEs) that are frozen and unproductive. The meanings of LR are often non-compositional, such as hung4 bok1 bok1 ‘bright red’, where bok1 bok1 does not contribute any meaning to hung4 ‘red’. This can pose a challenge in parsing and nat- ural language understanding. LRs can be mis- interpreted to bear literal meaning, if they are mistakenly treated as decomposable chunks. Identifying LRs is therefore an important step to reduce errors in word segmentation and natural language understanding. This study discusses a collection of LRs extracted from two data sets. Some common patterns are also identified in this study, which may guide parsers to automatically identify items that are novel to the system. 1 Introduction Reduplication in general is a ubiquitous phe- nomenon in Cantonese. There are many types of reduplication and they are mostly productive. This means that reduplications are templatic, and their compatibility with the lexical items is rule-governed. For example, non-stative verbs (i.e., verbs involv- ing actions) may be reduplicated to show durative events, as in (1a). Reduplication of the classifier (or measure words) denotes the meaning of ‘every’, as shown in (1b). To express the short duration of an event, one may use the ‘V+one+V’ redupli- cation in (1c). Since these reduplication types are rule-governed, their meaning and structure are pre- dictable and cause no problem in NLP. (1) a. haang6 walk haang6 walk haa5 ‘while walking’PRT b. zek3 CL zek3 CL gau2 ‘every dog’dog c. mong6 look jat1 one mong6 look ‘to take a look’ d. coeng4 long coeng2 long dei2 DEI ‘long-ish; fairly long’ Unlike the productive reduplications described above, lexical reduplications (LR) in Cantonese are multiword expressions (MWEs) that cannot be formed productively. LRs are considered frozen and are not coined by individual users. Many LRs are idiomatic in that their meanings are non-transparent. In (2a), the single use of dai2 can only mean ‘un- der / beneath’, but not undergarment. Example (2b) does not have a corresponding base form (i.e. non- reduplicated) tiu3 zaat3 either. When the item is used, tiu3 must be reduplicated. (2) a. dai2 under dai2 under ‘undies’ (casual term for undergarment) b. tiu3 jump tiu3 jump zaat3 tie ‘bouncy and active’ (Attested example from words.hk) What makes LR tricky is that they may carry literal meaning in some cases. This can cause parsers to an- alyze them as compositional and miss the intended idiomatic use. Example (3) contains the parallel of haang4 ‘walk’ and kei2 ‘stand’: (3) haang4 walk haang4 walk kei5 stand kei5 stand ‘being idle and aimless’ (From Cheung, Ngai and Poon (2018)) Taken literally, the phrase could mean walking and pausing intermittently. However, it actually means being idle and not doing anything and is often used to describe laziness. Its meaning is therefore unpre- dictable and must be learned …

Content truncated to 500 words