Adjective Order Restrictions and Perceptual Biases

A phenomenon that has been discussed much in the linguistic literature is the relative orders of stacked adjectives. In multiple unrelated languages, adjectives of different “semantic classes” display a tendency to show up in similar relative orders when jointly modifying a noun. For English, one formulation of the sequence, in abbreviated form, looks something like (1), modified after Scott (2002), but similar hierarchies can be found in other sources with sometimes very different analyses (e.g. Dixon, 1982), though some would describe them as trends rather than categorical (Bouchard, 2002; Truswell, 2009; among others).

(1) size > shape > texture > color > provenance > material > N

In other words, a stone that is both green and round would more naturally be described as

(2) a round green stone

than

(3) a green round stone

and similar effects are observed in many other languages (in languages with postnominal adjectives, the order is typically reversed, but what remains constant is that COLOR is closer to the noun than SHAPE in the unmarked case).

What I’m interested in is the cognitive basis for such ordering restrictions. Whether we take adjectival modification to be syntactically largely the product of free adjunction, with ordering governed primarily by semantic, communicative, and processing constraints, or whether we take them to reflect specific positions in a functional sequence (Cinque, 1994; Scott, 2002), we ultimately want to find an extralinguistic basis for the resultant sequence. Even if adult grammars possess an elaborate sequence of functional projections dedicated to specific classes of adjectives, it seems very unparsimonious to posit it as entirely situated within an innate grammar module.

What I’m not talking about

Some ordering facts receive relatively easy explanations in terms of the (formal) semantic contribution of the adjectives participating:

(4) a fake wooden spoon

is simply not the same thing as

(5) a wooden fake spoon

Similarly, when an A+N combination has a fixed idiomatic meaning, it is relatively easy to explain why it can’t be split up by other adjectives; thus, only (7), “violating” the sequence in (1), preserves the idiomatic meaning of “black bear”.

(6) a black Canadian bear (a dark-furred bear of any species from Canada)
(7) a Canadian black bear (a specimen of Ursus americanus from Canada)

Ordering restrictions among different intersective adjectives

But even if we exclude idiomatic readings and scope taking adjectives, we’re left with some empirically solid ordering facts that are unexplained by compositional semantics alone: There is no obvious logical reason for SHAPE > COLOR, for example. Both adjectives describe physical attributes, and both modify the noun phrase through set intersection: a “round green pebble” is an object that is in the intersection of pebbles, round objects, and green objects. Since this set is the same in whichever order the operations are performed, a “green round pebble” would be logically equivalent. So we have to turn elsewhere for an explanation of these ordering facts.

Intrinsicalness

One proposal that has been around since at least the 1960s especially in the psycholinguistic literature (exemplified by Martin, 1969) is that the perceived “intrinsicalness” of a property to the object it applies to determines order (other dimensions that have been proposed include the distinction between “relative” and “absolute” adjectives, or between objective and subjective properties, but those don’t apply to SHAPE vs. COLOR). What exactly that means, and what are the correlates of a properties (perceived) “intrinsicalness” outside of adjective order restrictions, remains less clear. Byrne (1979) hints that it might best be interpreted as feature permanence or typicality: In a forced choice paradigm, subjects consistently interpreted (8) as “an aging greyhound” and (9) as “a rocket powered Saint Bernard”, i.e. the adjective further from a noun was interpreted as a transient property.

(8) a slow fast dog
(9) a fast slow dog

While fascinating, a problem with this study is that all of the example sentences seem rather marginal in the first place, and would probably be most naturally pronounced with heavy focus or comma intonation. So the results may not directly translate to natural unbroken sequences.

Another hint is provided by Belke (2006), who shows, by asking participants to identify a target object in an array of objects that varied by size, color, or both, that as many as 87% of subjects mention color when it is redundant, i.e. when size alone is necessary and sufficient to uniquely identify a target object.

(created in tikz, after Belke (2006))

This may indicate that the processing of color information is more automatic and harder to inhibit than size. A potential problem is that she contrasted SIZE – a relative adjective – with COLOR, rather than e.g. SHAPE.

So, while there are some interesting threads to pick up, I think that a there’s still a lot of work to be done: The relevant notion of “intrinsicalness”, or whatever ultimately turns out to be the dimension that determines adjective orders, needs to be spelt out much more explicitly before we can systematically look for correlates outside of language, and possibly for parallels in other species.

__________________

*I’m currently in the process of formulating an exposé along these lines. I’m grateful for any ideas and pointers to further literature, especially recent(ish) theoretical treatments from outside of generative grammar, of which I’ve found very little. There’s a long version I’m working on but I hesitate to post it, too many unfinished threads.

Belke, E. (2006). Visual determinants of preferred adjective order. Visual Cognition 14(3), 261–294.

Bouchard, D. (2002). Adjectives, Numbers, and Interfaces: Why Languages Vary. North Holland Linguistic Series. Elsevier.

Byrne, B. (1979). Rules of prenominal adjective order and the interpretation of “incompatible” adjective pairs. Journal of Verbal Learning and Verbal Behavior 18(1), 73–78.

Cinque, G. (1994). On the Evidence for Partial N Movement in the Romance DP. In G. Cinque, J. Koster, J.-Y. Pollock, L. Rizzi, and R. Zanuttini (Eds.), Paths Towards Universal Grammar, pp. 85–110. Georgetown: Georgetown University Press.

Dixon, R. M. W. (1982). Where have all the adjectives gone?, Chapter 1: Where have all the adjectives gone? Berlin: Mouton.

Martin, J. (1969). Semantic determinants of preferred adjective order. Journal of Verbal Learn- ing and Verbal Behavior 8, 697–704.

Scott, G. (2002). Stacked Adjectival Modification and the Structure of Nominal Phrases. In G. Cinque (Ed.), Functional Structure in DP and IP, Volume 1 of The Cartography of Syntactic Structures, Chapter 4, pp. 91–120. Oxford: Oxford University Press.

Advertisements

One thought on “Adjective Order Restrictions and Perceptual Biases

  1. Hi, you might be interested in reading section 4 of this paper (that won an award in an AI conference in Germany) http://www.kuenstliche-intelligenz.de/archives/ki-journal-20160917/fileadmin/template/main/archiv/pdf/ki2009-02_page18-23_web_full.pdf

    I am now writing a longer but less technical paper that shows that logic, ontology and the computational notions of polymorphism and type casting are what is behind the adjective-ordering restrictions and from there i argue that purely quantitative (statistical, neural network and similar data-driven) approaches to natural language processing are not only flawed, but futile.

    Would love to discuss this further.

    Regards,
    Walid Saba
    https://www.linkedin.com/in/walidsaba

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s