Productivity and semantic transparency: An exploration of word formation in Mandarin Chinese

We used word embeddings to study the relation between productivity and semantic transparency.We compiled a dataset with around 2700 two-syllable compounds that shared position-specificconstituents (henceforth pivots), and some 1100 suffixed words. For each pivot andsuffix, we calculated measures of productivity as well as measures of semantic transparency.For compounds, productivity (P) was negatively correlated with the number of types (V ) andwith the semantic similarity between non-pivot constituents and their compounds. Conversely,greater semantic similarity of the pivot with either the compound or the non-pivot constituentpredicted higher degrees of productivity. Visualization with t-SNE revealed clustering of suffixedwords’ embeddings, but no by-pivot clustering for compounds, except for a minority ofpivots whose regions in semantic space did not contain intruding unrelated compounds. Asubset of these pivots were found to realize a fixed shift in semantic space from base word tocompound, a property that also emerged for several suffixes. For these pivots, no correlationbetween P and V was present. Thus, Mandarin compounds appear to realize, at one extreme,motivated but unsystematic concept formation (where other pivots could just as well have beenused), and at the other extreme, systematic suffix-like semantics.