Considering word token versus word type in the interpretation of Pillai scores: The case of the Cantonese AM/P∼OM/P merger

Stanley & Sneller (2023) have recently demonstrated the importance of large sample sizes in the interpretation of Pillai Scores in sociophonetic research. In this paper, I address whether larger is necessarily better if larger means more tokens of the same word types. The case study focuses on the AM/P∼OM/P merger in Cantonese, which involves /o/ becoming /ɐ/ in pre-labial contexts. The list of OM/P words is very small, though a few of these words are high frequency. The data come from spontaneous speech samples from 41 participants from the HLVC Corpus (Nagy 2011). Pillai Scores were calculated for each individual speaker based on both a smaller sample (five tokens maximum of the same word type, n = 1163) and a larger sample (more than five tokens of the same word type, n = 1659). Applying Stanley & Sneller’s (2023) threshold value formula, results show 10 merged speakers with the smaller sample but five merged with the larger sample. Overall, these results show that while some researchers may have previously overestimated the spread of this merger (Bauer & Benedict 1997), increasing sample size without increasing word type diversity may lead to underestimation of the extent of the merger.