From Frying to Speculating: Google Ngram evidence to the meaning development of ‘?’ in Mandarin Chinese

From Frying to Speculating: Google Ngram evidence to the meaning development of ‘炒’ in Mandarin Chinese Jing Chen, Chu-Ren Huang The Hong Kong Polytechnic University Department of Chinese and Bilingual Studies Yuk Choi Road 11, Hung Hom, Kowloon, Hong Kong jing95.chen@connect.polyu.hk, churen.huang@polyu.edu.hk Abstract This paper explores semantic change simply using the Ngram information, with the intu- ition that Ngrams as the direct neighbors offer collocation cues that further signal meaning changes. We specifically investigate the case of ‘炒’ in Mandarin Chinese on the basis of un- igrams extracted from Google Books Ngrams Corpus and reconstruct the meaning develop- ment of ‘炒’ with specific stages. The results indicated that the major meaning changes that occurred to ‘炒’ is from ‘frying’ to ‘speculat- ing’, which roughly started in the 1970s. The attested word types related to the latter sense denote economic events, such as stocks, for- eign currencies, and speculators. It further re- flects that social context plays an essential role in the process of semantic change. 1 Introduction The availability of large historical digital corpora and the recent advance in natural language pro- cessing have greatly facilitated empirical stud- ies on semantic change nowadays (Michel et al., 2011; Hamilton et al., 2016a,b; Tahmasebi et al., 2019). Beyond the world of historical linguistics, the computational community has shown a grow- ing interest in exploiting statistical and compu- tational models to automatically detect semantic change over the past two decades, from monitor- ing the fluctuation in the frequency of target words in historical texts (Michel et al., 2011; Hilpert and Gries, 2009; Kulkarni et al., 2014) to measur- ing their context differences using the state-of-art distributional models(Kim et al., 2014; Hamilton et al., 2016b; Giulianelli et al., 2020). Building off of the distributional hypothesis in linguistics, ‘You shall know a word by its com- pany’ (Firth, 1957), the distribution-based ap- proach has taken up the predominant position in the recent lexical semantic change detection task (Schlechtweg et al., 2020; Tahmasebi et al., 2019). In general, distributional differences over a period of time could be quantified by constructing and evaluating historical word embeddings (Hamilton et al., 2016a,b). However, this novel trend is still in its youth. Most existing studies focus on evalu- ating how much the overall distribution of a word form deviated over two or three intervals but tell little about how the meaning of a word devel- oped as a continuous process (Kutuzov and Pivo- varova, 2021; Rodina and Kutuzov, 2020; Tah- masebi et al., 2019; Kutuzov et al., 2018). More importantly, the popular word-type models, also known as static word embeddings, may not be sen- sitive to less contrastive usage drifts (Tahmasebi et al., 2019; Schlechtweg et al., 2020). In this paper, we present a simple and inter- pretable way to track the meaning development by collecting evidence directly from N-grams data extracted from Google Books Ngram dataset(the Chinese subpart) and manually checking the di- rect neighbours of target words over times. The intuition here is that N-grams provide collocation information …

Content truncated to 500 words