Predicting cross-linguistic adjective order with information gain

Abstract

Languages vary in their placement of multiple adjectives before, after, or surrounding the noun, but they typically exhibit strong intralanguage tendencies on the relative order of those adjectives (e.g., the preference for 'big blue box' in English, 'grande boîte bleue' in French, and 'als . undūq al'azraq alkabīr' in Arabic). We advance a new quantitative account of adjective order across typologically-distinct languages based on maximizing information gain. Our model addresses the left-right asymmetry of French-type ANA sequences with the same approach as AAN and NAA orderings, without appeal to other mechanisms. We find that, across 32 languages, the preferred order of adjectives mirrors an efficient algorithm of maximizing information gain.

Adjective Ordering Across Languages

“…PMI also offers an operationalization for meaning closeness. Dyer et al (2021) propose yet another information-theoretic predictor of ordering preferences: information gain, a measure of uncertainty reduction on the basis of the evidence, both positive and negative, provided by the occurrence of an adjective. For Dyer et al, speakers greedily maximize information gain, which means that adjectives that contribute greater information gain appear earlier than adjectives with less information gain.…”

“…Only one study has looked at information gain, but its scope dwarfs many of the previous investigations. Dyer et al (2021) investigate information gain using corpora of 32 languages; of these, information gain successfully predicts attested orderings in 29 languages. Notably, Dyer et al test the predictions of information gain in three different ordering templates: AAN, NAA, and ANA-the only study mentioned in this section to look at mixed ANA orderings.…”

“…Across all three paradigms, the authors find that preferences in Hebrew are weaker or less categorical than those in English. Moreover, both Dyer (2017) and Dyer et al (2021) find that their predictors-integration cost and information gain, respectively-at least trend such that they predict less well in Hebrew than in English. Given that Hebrew has postnominal adjectives while English adjectives are prenominal, we may wonder whether the different communicative status of pre-versus postnominal adjectives may affect the strength of ordering preferences.…”

“…Given that Hebrew has postnominal adjectives while English adjectives are prenominal, we may wonder whether the different communicative status of pre-versus postnominal adjectives may affect the strength of ordering preferences. 9 However, Dyer et al (2021) assess the relative flexibility of adjective ordering in 18 prenominal versus 13 postnominal languages and find that the rate of ordering flexibility (calculated as the number of adjective pairs attested in both possible orders) is comparable across templates. But, as the discussion above highlights, different measures may be more or less sensitive as assessments of a given research question regarding adjective ordering.…”

Investigating Strategies for Mitigating and Avoiding Expert Blind Spot Development in Teaching English Attributive Adjectives

“…Most common errors are committed in English attributive adjectival phrases include using the wrong adjective as in *I have bicycle one' instead of 'I have one bicycle' (Scontras, 2023), overusing adjectives, placing adjectives in the wrong order as in the adjectival phrase '*a box brown small' instead of 'a small brown box' as the order of listing is done from most important to least one (NDA, 2022). Dyer, Futrell, Liu & Scontras (2021) postulate that adjectives contributing greater information gain appear earlier than those with less information gain. Some errors involve using multiple adjectives unnecessarily, stringing numerous adjectives to form new words; and using words that are not real English attributive adjectives.…”