Predicting semantic transparency of Chinese QIEs using distributional semantics and lexical frequency

Abstract Semantic transparency refers to the relationship between the meanings of whole-words and their constituent morphemes. Mandarin Chinese Quadrisyllabic Idiomatic Expressions (QIEs, also known as Cheng-Yu) have a similar property, in which the whole meaning is often more than the summed meanings of the component words. Using data from Wu (2016) , we analyze the semantic transparency of QIEs as a function computational semantic similarity, word frequency, and syntactic structure. Our results indicate that the probability of a QIE being labelled as transparent increases with its frequency, whereas computational measures of semantic similarity and structure were not strongly associated with one existing set of semantic transparency labels. We hypothesize that these results may be influenced by the nature of QIEs, where their meaning may sometimes be obscured and based on traditional stories rather than any properties of the constituents — semantic or otherwise. Therefore, we advocate for the inclusion of word frequency as a factor in transparency rating of QIEs. Additionally, we suggest exploring other variables that may improve the transparency rating models.