The morphology and phonology of exponence ed. by Jochen Trommer (review)

This book brings together a number of articles about how morphosyntactic feature bundles get turned into things (un)pronounceable. The book begins with an excellent overview by the editor, which is less opinionated than the one you are about to read (http://fds.oup.com/www.oup.com/pdf/13/9780199573738_chapter1.pdf).

Ch. 2, ‘The architecture of grammar and the division of labor in exponence’, by Ricardo Bermúdez-Otero takes on a large and difficult array of issues from reduplication and morpheme-specific phonology, to the productivity, variability, and exceptionality of phonological processes, to classical blocking and cyclicity, to frequency effects and questions of lexical storage. The result is a complicated grammatical processing model based on symbolic rules and an articulated lexicon that stores prosodic structure that most phonologists would derive; but it is this articulated lexicon that allows it to model such a wide array of facts. Bermúdez-Otero ties together the messy and the elegant using a dual-process model (storage plus processing) in a feed-forward grammar based on lexical redundancy rules (Jackendoff 1975), stratal optimality theory (Bermúdez-Otero 1999, Kiparsky 2000), and a dual-route processing model (Prasada & Pinker 1993). He argues for severe restrictions on how morphology and phonology interact and against any kind of morpheme-specific phonology, treating all apparently nonconcatenative morphology as concatenating defective phonological material—for example, treating reduplication as the affixation of empty prosodic structure.

Andrew Nevins’s ‘Haplological dissimilation at distinct stages of exponence’ (Ch. 3) collects evidence that haplological dissimilation takes place at major stages in language processing. Nevins assembles a formal model of spell-out based on Ackema & Neeleman 2003, Arregi & Nevins 2007, and Richards 2010, with four stages of processing: (i) syntactic linearization, (ii) prosodic phrasing, (iii) word formation, and (iv) vocabulary insertion. Dissimilation occurs when things are too similar at whatever level is relevant: for example, syntactically too similar, or prosodically too similar. I like the idea of the paper, but many of the details seem arbitrary: linear adjacency is relevant for stages two and four but not for stages one and three—why does it disappear and then come back? In what sense are French que ‘than’ and que ‘that’ ‘completely featurally identical’ (93)? Or Romanian -ul ‘def’ and -al ‘poss.m.sg’, which do not look ‘completely featurally identical’ but have to be for the analysis? Much of the data in this paper seemed to me cherry-picked or shoe-horned to fit the model; a separation of facts from model would have helped me appreciate the paper more.

The major result of Ch. 4, ‘Morphophonological polarity’, by Paul de Lacy is that it gets rid of the clearest, best-studied case of morphophonological polarity, one involving stem-final voicing alternations in Dholuo. The polarity is supposed to involve voicing in certain singular ~ plural pairs (alap~ ӕlӕbe ‘open space(s)’, gɔt ~ gɔdɛ ‘hill(s)’) but devoicing in others (kitӕbu ~ kitepe ‘book(s)’, kεdε ~ kεtε ‘twig(s)’). Building on unpublished work by Bye (2006) and Trommer (2008), de Lacy clearly shows that ‘root-final consonants devoice in the singular while in the plural the rightmost consonant devoices only if it is non-final ... in the root ... There is no morphophonological polarity’ (124–25). That much is clear and neat, and it casts huge doubt on whether any cases of morphophonological polarity will remain once good investigative work like this is done. The actual analysis that derives the forms above, however, is technical and dense and involves tricks of the trade that not every reader will find enlightening; we are probably done with morphophonological polarity but not with Dholuo.

Dieter Wunderlich in ‘Polarity and constraints on paradigmatic distinctness’ (Ch. 5) argues convincingly that morphological polarity (e.g. when -a marks masc.sg. but fem.pl., while -o marks masc.pl. and fem.sg.) is...