Semantic transparency measures for English compounds
Existing compounds
This dataset contains semantic transparency measures for a set of 1,865 English compound words, such as airport or ladybird. It is included as Supplemental Material to the following article:
Günther, F., & Marelli, M. (2018). Enter sand-man: Compound processing and semantic transparency in a compositional perspective. Journal of Experimental Psychology: Learning, Memory, and Cognition. doi: 10.1037/xlm0000677
The measures are obtained from an English distributional semantic space, where each word is represented as a high-dimensional numerical vector. Using this semantic space, modifier relatedness (simMO) is computed as the cosine similarity between modifier and compound, cos(air, airport) , and analogously head relatedness (simHO) as the cosine similarity between head and compound, cos(port, airport).
Furthermore, a compositional model was employed to obtain compositional meaning representations for the compounds (i.e., the meaning that would be predicted given the combination of their constituents). With this compositional meaning, modifier composition (simMC) is computed as the cosine similarity between modifier and compositional compound, cos(air,[air+port]), head composition (simHC) as the cosine similarity between headand compositional compound, cos(port, [air+port]), and compound compositionality (simOC) as the cosine similarity between actual and compositional compound, cos(airport, [air+port]).
For more details, see the article (Günther & Marelli, 2018)
Novel compounds
This dataset contains the same measures for a set of almost 2 million novel English compound words, that is words consisting of two constituents that are not observed in our source corpus (such as cottonwatch, zeroworker, or thunderdad). Please cite the following article when using this dataset (which includes additional details on the dataset), where it was the candidate set described at the beginning of the Material section of Experiment 1:
Günther, F., & Marelli, M. (2020). Trying to make it work: Compositional effects in the processing of compound “nonwords”. Quarterly Journal of Experimental Psychology, 73(7), 1082-1091. doi: 10.1177/1747021820902019
Note that measures relying on an observed compound representation (head and modifier relatedness, as well as compound compositionality) cannot be computed for novel compound (since they are not observed in a corpus). Therefore, the dataset only contains modifier composition (simMC), head composition (simHC), as well as constituent similarity (simMH, the similarity between the modifier and the head).
The article (Günther & Marelli, 2020) contains an analysis of processing times for a subset of the dataset presented here.