Keynote Lecture

Noun Phrase Order: Grammar, Parser and Typological Frequency

Outline
The neutral order of demonstratives, numerals, adjectives and nouns has been a topic of typological research since the 1960s (Greenberg 1963, Cinque 2005, Dryer 2018, among others). The pattern of attested and unattested orders permits a simple analysis based on variation in the linearization of base-generated orders alongside the option of leftward movement of (constituents containing) the noun (Cinque 1996, 2005, Ackema & Neeleman 2002, Abels & Neeleman 2012;). It is not clear, however, why noun movement should be leftward, nor why the frequency of attested orders should display such striking asymmetries.

In this talk, I argue that typological frequency is determined by a combination of innate preferences and the effort required to parse structures characteristic of the target grammar. It is clear why innate preferences should play a role. Increased parsing effort leads to lower typological frequency because it reduces the intake available in language acquisition; Hawkins 1990, 2009; Kirby 1999).

Four ranked factors are enough for an empirically adequate model: (i) an innate preference for harmonic orders (Cinque 2005, Culbertson et al. 2012, Culbertson & Newport 2015, Dryer 2018); (ii) the processing cost of movement (associated with buffering and gap finding); (iii) the additional processing costs of rightward movement (Staub et al. 2006); and (iv) the processing costs of adjective-noun orders (as compared to noun-adjective orders; Culbertson et al. 2012, Dryer 2018). The four factors are ranked as follows:

(1)     No-Rightward-Movement > Harmony > N-before-A > No-Movement.

The strongest factor (the processing costs of rightward movement) has enough weight to reduce the frequency of orders derived by rightward noun movement to zero. The remaining three factors conspire to generate the ranking of attested orders in (2) (more frequent orders first):

(2)     N-A-Num-Dem > Dem-Num-A-N > {Dem-N-A-Num, Dem-Num-N-A, Num-N-A-Dem} > {N-A-Dem-Num, Dem-N-Num-A, N-Num-A-Dem, N-Dem-A-Num, N-Dem-Num-A} > {Dem-A-N-Num, Num-A-N-Dem, A-N-Num-Dem} > A-N-Dem-Num

How well does the ranking in (2) fit the data? There are two relevant data bases, collected by Dryer and Cinque respectively. The determination of frequency can be based on Dryer’s (2018) notion of adjusted frequency, which takes account of genealogical and geographical distance. Alternatively, one could take frequency of language type or frequency of genera containing languages of the relevant type as measure. The resulting five measures do not always agree (for some orders, the spread across measures is as much as six ranks). Nonetheless, the ranking in (2) is a good match with the rankings based on each of the individual measures (in each case, only one pair of orders does not fit), and a perfect match with the average rankings.