Beyond quantity of experience: Exploring the role of semantic consistency in Chinese character knowledge.

Abstract

Most printed Chinese words are compounds built from the combination of meaningful characters. Yet, there is a poor understanding of how individual characters contribute to the recognition of compounds. Using a megastudy of Chinese word recognition (Tse et al., 2017), we examined how the lexical decision of existing and novel Chinese compounds was influenced by two properties of individual characters: family size (the number of distinct words that embed a character) and family semantic consistency (the average semantic relatedness between a character and all words containing it). Results revealed that both variables influence word and nonword processing: Words are recognized more quickly and accurately when they contain characters that occur frequently across different words and that make consistent meaningful contributions to those words, while nonwords containing those types of characters are rejected more slowly. These findings suggest that the learning of individual characters is based not only on the quantity of experience with them but also on the reliability of the semantic information they communicate. In addition, readers are able to generalize character knowledge acquired from previous word experiences to their daily encounters with familiar and unfamiliar words. We close by discussing how word experience shapes character knowledge when different ways of calculating family properties are considered.

Compositional processing in the recognition of Chinese compounds: Behavioural and computational studies

“…We then sought to determine the relationship between these model-driven measures and various forms of human behavioural data to understand how the ease of incorporating character meanings into overall compound meaning impacts how adults process compounds (Günther & Marelli, 2019. We hypothesised that the proximity measures derived from the CAOSS model would perform better than those derived from the simple additive model due to the importance of positional information in Chinese (Hsieh et al, 2024).…”

“…Previous studies have shown that constituents are readily accessed during recognition of Chinese compounds (e.g. Hsieh et al, 2024;Tsang & Chen, 2013;Xiong et al, 2023). However, it remains unclear whether and how readers consider the meanings of individual characters in combination during the processing of compounds.…”

“…However, it remains unclear whether and how readers consider the meanings of individual characters in combination during the processing of compounds. The high degree of ambiguity in the meanings of Chinese characters (Hsieh et al, 2024) and the much greater degree of structural variability in Chinese compounds (Ceccagno & Basciano, 2007) raise questions about whether this type of compositional process would foster efficient understanding in Chinese. Likewise, while the CAOSS model provides an account of the influence of semantic composition in the recognition of English and German compounds Günther & Marelli, 2019, it is unclear whether there is sufficiently consistent structure in Chinese compounding for the generic CAOSS architecture to learn the meaningful functions of Chinese compound constituents.…”

“…Our compositional DSMs were built using the word2vec model (Mikolov et al, 2013) pretrained on traditional Chinese developed by Academia Sinica, Taiwan (available at https://ckip.iis.sinica.edu.tw/project/embedding). We chose this model as it shows good performance in simulating human semantic relatedness judgements (Hsieh et al, 2024). This model was trained using the skip-grams with negative sampling (SGNS) method, which predicts word contexts given a target character or word.…”

“…Ji & Gagné, 2007;Marelli, 2023). Moreover, the mapping between the meanings of characters and whole compounds is not consistently systematic, as Chinese characters frequently bear multiple meanings (Hsieh et al, 2024;Liu et al, 2007;Tsang & Chen, 2010). For instance, the character 公 /gōng/ can mean not only "male" (as in 公雞 /gōng jī/, "rooster") but also "public" (as in 公寓 /gōng yù/, literal: "public accommodation", or "blocks of flats").…”