Headedness, overlap, and salience: the multi-dimensional tension in English blends

A recent contribution to Digital Scholarship in the Humanities by Meng and Hilpert (‘Measuring the Semantic Headedness of English Blends with Token-Based Semantic Vector Space Modelling: A Corpus-Based Study’, 2024, 39 (4), 1075–1091) applied distributional semantics to English lexical blends but conflated three conceptually independent dimensions: semantic headedness, semantic overlap, and semantic salience, while relying on static word embeddings that cannot adequately handle polysemy. To address these gaps, the present study distinguishes the three metrics and quantifies them using the contextualized language model BERT on a dataset of eighty-seven English blends. The research findings are threefold: (1) semantic headedness is right-headed in 52.9% and left-headed in 47.1% of the eighty-seven blends; semantic overlap is left‑oriented in 43.7% and right‑oriented in 56.3% of the eighty-seven blends; (2) in terms of semantic salience, 56.3% of the eighty-seven blends are left-oriented and 43.7% are right-oriented; (3) the three metrics exhibit systematic directional inconsistency, revealing that English blends function as multi-dimensional semantic objects in which categorical inheritance, distributional similarity, and contextual prominence need not converge, which calls for a more empirically grounded and multi-faceted account of blend semantics. Theoretically, our findings corroborate the gradient view of Meng and Hilpert, moving blend research beyond taxonomic classification towards a usage-grounded, multi-dimensional model that requires a multivariate approach. Methodologically, the study demonstrates the necessity of cross-validating conclusions using multiple metrics in morphological research, illustrating how digital methods can address longstanding problems in word formation.