An opaque and diffuse system

New (I think) article on Ryukyuan from Thomas Pellard and Masahiro Yamada! “Verb morphology and conjugation classes in Dunan (Yonaguni)” (also redirected from academia.edu). Here’s the abstract, broken into two paragraphs by me for readability:

Most Japonic languages have a relatively simple and transparent morphology. Their verb morphology is usually characterized by a highly agglutinative structure that exhibits little morphophonology, with only a few conjugation classes and a handful of irregular verbs. In sharp contrast with its relatives, Dunan (Yonaguni), a highly endangered Japonic language of the Southern Ryukyuan branch, spoken by approximately 400 speakers located on Yonaguni Island (Okinawa prefecture, Japan), exhibits a unusually complex verb morphology for a Japonic language, mostly due to some drastic historical changes. The verb morphology of Dunan seems be the most complex one within the whole Japonic family, and a systematic description challenges in several interesting ways morpheme-based approaches.

The following presents an outline of the verb morphology of Dunan, limited to the basic synthetic forms of regular verbs. Focus is put on the partition of verbs into classes and its morphomic pattern. Three morphomic factors are identified as partitioning verbs into conjugation classes: stem alternation, suffix allomorphy, and metatony. The resulting system of paradigm classes is found to be opaque and to show little interpredictability between paradigm cells, i.e. few reliable inferences can be made from one inflected form about other forms. Morpheme-based approaches are not well-suited for the analysis of Dunan’s verb morphology, which rather calls for an abstractive Word and Paradigm approach.

To a hardened Indo-Europeanist, the morphology Pellard and Yamada describe is probably just barely sufficient to raise a single eyebrow, but (as they observe) in the highly agglutinative context of Japonic, it is quite off the chain:

[T]he number of distinct stem forms per verb does not exceed three. However, the distribution of stems within paradigms is not uniform for all verbs (Table 2.7). For instance, in the case of verbs with three stems, the shortest stem usually appears only in the perfect cell, but not in the case of sigmatic verbs, where the same stem is shared by the hortative, medial, and perfect cells. On the other hand, for sigmatic verbs, the shortest stem is used in the negative cell only, while for most other classes the negative shares the same stem as the present and imperative forms.

The number of stems needed to account for the whole pattern of stem alternations across the different classes amounts to seven, i.e. more than twice the number of distinct stem forms for any verb (Table 2.7). This leads to the unexpected conclusion that there are almost as many stems as basic forms.

(Seriously, check out table 2.7. It’s on page 37.)

Overall, a very satisfying read for a construction grammar zealot like me, although with a real gut-punch of a conclusion:

The most realistic hypothesis is that speakers memorize whole inflected forms (minimally, principal parts) as part of a network of interdependencies with an implicational structure. Unfortunately psycholinguistic tests are hardly feasible in the case of Dunan, a highly endangered language still spoken by elderly persons only. A more computational approach, based on entropy and complexity measures (Ackerman et al. 2009, Finkel & Stump 2009, Sagot & Walther 2011, Blevins 2013) is a more realistic goal for future research.