Conflating Directional Association Measures―A Case Study on NP Constructions―

Abstract

This paper proposes a method to conflate directional association measures. In measuring associations of collocates, various measures are proposed (e.g., Dice, Jaccard, Mutual Information (MI)). These conventional association measures are said only to reflect limited aspects of collocates. One of the most frequently discussed aspects in collocational analysis is the direction of associations. For instance, when analysts wish to state that the possessive article “my” is likely to occur with the noun “sister”, they need to recognise the two questions regarding the two different directions of association: (i) ‘When the word “my” is used, how likely does the word “sister” collocate?’, and (ii) ‘When the word “sister” is used, how likely does the word “my” collocate?’. Collostructional analysis implements these directions as two types of ΔPs, treated as independent aspects of the association (cf. Gries 2019, 2023). This paper proposes a method to integrate these dimensions. Our approach can account for (i) the sparse distributions of two ΔPs and (ii) the positive correlations between constructions and words more smoothly.