Effects of online abstraction on adjective order preferences

Abstract

This study investigates whether the generalisation of prenominal adjective order preferences is best accounted for by a linear precedence relationship between persistent abstract adjective categories (category precedence) or by online abstraction over adjective exemplars. In a two-alternative forced-choice task, participants had to select which of two adjective orders they preferred for low-frequency adjective pairs. Online abstraction over exemplars, operationalised in a range of exemplar-based variables, explained variation in the participants' adjective order choices that category precedence could not account for. However, category precedence still explained variation in adjective order preferences that was not captured by the exemplar-based variables. Although these findings might support a dual-mechanism model of adjective order preferences, with additive roles for abstract adjective categories and online abstraction, an improved operationalisation of online abstraction and a stricter control of the covariates introduced through the implementation of both approaches may eliminate the effect of category precedence.

Keywords:

• adjective
• word order
• exemplar model
• categorisation
• syntax–semantics interface

Acknowledgement

This paper is a revised version of a chapter in the first author’s doctoral dissertation completed at the University of Antwerp, Belgium.

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplemental data

Supplemental data for this article can be accessed here.

Notes

1. Bache and Davidsen-Nielsen’s (1997) ordering scheme contained lexical categories that not all linguistic paradigms might regard as being subject to adjective ordering restrictions (i.e. quantifiers and nouns). According to such paradigms, changing the position of those categories relative to “true” modifiers invariably results in an ungrammatical order. The experiment reported in this paper does not address the question whether adjective order theories can also explain the order of quantifiers and prenominal nouns, as the items in our stimulus set did not contain words belonging to those categories (also see Footnote 2).

2. The taxonomy and corresponding precedence hierarchy of Bache and Davidsen-Nielsen (1997) is applicable to Dutch with only one minor change: because bare nouns do not function as premodifiers in Dutch, there is no need for a Mod3: nominal category. For a Dutch noun to function as a premodifier, it either needs to be derived into an adjectival modifier by appending the denominalisation suffix -en (e.g. het standvastige tinnen soldaatje [the steadfast tin soldier]) or merged with the head noun into an orthographic compound (e.g. cotton industry translates to katoenindustrie).

3. Available from http://mpqa.cs.pitt.edu/lexicons/subj_lexicon.

4. Available from http://ilk.uvt.nl/frog/.

5. Available from https://ilk.uvt.nl/timbl/.

6. The IB1 algorithm used the following settings: a neighbourhood size (k) of 4, tie resolution by random class selection from the exemplar neighbourhood, no feature weighting, and the Modified Value Difference Metric (MVDM) (Cost & Salzberg, 1993) (with w_{i + 1} as the exemplar class) for distance metric. Other parameters were kept at their default values.

7. We used the Python™ package Leanlex (Keuleers, 2007) to access CELEX.

8. Given the limited coverage of the order class set (a problem also noted for other category sets by Wulff, 2003) and the lack of order-independent heuristics to determine whether an adjective is primarily specifying, descriptive, or classifying, we did not implement a variable that purely relied on the linear precedence relationship between adjective-inherent order classes.

Additional information

Funding

This research was supported by a PhD fellowship from the Research Foundation – Flanders (FWO), awarded to the first author.