At the recent LIPP Symposium on parts of speech in Munich, one of the most interesting papers was Jennifer L. Smith’s paper on “Parts of speech in phonology” (see also Smith 2011). She examines 20 languages from 11 families and finds that in almost all cases where nouns, verbs and adjectives behave differently phonologically, the nouns have more phonological contrasts than adjectives, and these in turn have more contrasts than verbs. A well-known example is Spanish, where stress is contrastive in nouns (cf. 1a-b) and adjectives, but verbs show uniform stress for all verb classes.
(1) a. sábana ‘sheet’
b. sabána ‘savanna’
Hebrew is similar, but adjectives are intermediate between nouns (with variable stress) and verbs (with uniform stress) in that stress variability is more constrained than in nouns. So Smith formulates a “hierarchy of phonological privilege” N > A > V, and she observes that the word-class effects are overwhelmingly prosodic in nature, involving stress, pitch accent (as in Japanese), tone (as in Ewe), and prosodic shape (as in Chukchi). Another language where the prosodic shape of nouns and verbs is different is Oko (a Benue-Congo language of Nigeria), where verbs must begin with a consonant, and nouns must begin with a vowel, and where even loanwords must obey this, so that English-derived consonant-initial words take an epenthetic initial vowel (e.g. ifiriji ‘fridge’, Atoyebi 2010).
That the phonology can refer to grammatical information also means, according to Smith, that words cannot be category-less in the lexicon, as some recent generative syntactic proposals would have it (e.g. Embick & Marantz 2008): Speakers must store word-class information and phonological information in the same place. (Another very special example of this sort is the mixed language Michif, where nouns are derived from French and verbs are derived from Cree, and both keep the phonological patterns of the source languages.) Smith also notes that the hierarchy N > A > V is more in line with functional-typological clines than with generative binary-feature models. This also fits with the statistical nature of the generalization: There is at least one exception, Ewe, which has more tone distinctions in verbs than in nouns.
What could be the origin of these phonological distinctions? Smith notes that verbs show more inflection that nouns. Using several relevant chapters from WALS, she finds the following:
(2) a. Languages with noun inflection also have verb inflection 94% of the time.
b. Languages with verb inflection also have noun inflection only 76% of the time.
So verbs tend to me more inflected than nouns. (Even if I am skeptical concerning our ability to distinguish inflection from associated clitics, it does seem likely that we would get very similar results if we counted all associated function-items, rather than only inflection.) How can this lead to prosodic differences? Smith speculates about a bias in language acquisition:
“It may be the case that learners have more cognitive resources available to learn complex phonological patterns in nouns than in verbs”
But she does not say why the bias should come from acquisition rather than language use. So this could be a usage-bias affecting all speakers which can get grammaticalized in language change. In any event, it must be a very small bias, because apparently most languages do not show word-class differences in their phonology, and even when they do, the difference does not always mean that one word-class is more privileged than another (i.e. allows more contrasts). But world-wide cross-linguistic work can detect even fairly weak tendencies, and we might eventually find the right explanation.
One might ask further whether phonological differences can also be associated with smaller classes of words, e.g. transitive verbs, or mass nouns, or non-gradable adjectives. Probably examples of all these can be found, they just need to be systematically collected. Another example that Smith mentioned at the conference is proper names, which sometimes show different stress patterns, as is most famously found in Turkish.
References
Martin, thanks for the discussion! Thanks also to other commenters for further examples of category-specific phonology. Some of the languages mentioned here I have already included in the survey, but others are new leads that I am looking forward to following up.
Just one minor point of clarification:
What I actually intended to propose in the talk (see (24) on the handout) was that either this is true, and lexical entries contain category information, or it must be the case that morphosyntactic levels of whatever size it is that does determine word-class categories (phrases, functional projections, etc.) are able to interface with the phonological component. Either way, I think there are interesting implications for the morphosyntax/phonology interface.
If anyone is interested in the handout from last week’s conference, it’s available on my web site, here. There’s also a pre-publication version of the 2011 paper available here. Comments, and additional examples of relevant languages, are always welcome.
How very interesting. What springs to mind is that nouns may often be used in isolation, i.e. make up an utterance on their own, whereas verbs are probably used as part of a larger utterance to a much greater extent than nouns are. In other words, with nouns word prosody and sentence prosody may conflate, whereas they are distinct with verbs. Perhaps it is for this reason that nouns are phonologically less constrained than verbs.
Within nouns, proper names are probably even more likely to be used in isolation than common nouns, so they should enjoy even greater phonological freedom. This seems to be the case in Italian: among the handful of (otherwise blocked) proparoxytona with a heavy penult in Italian, most are place-names (Otranto, Lepanto, Levanto). But of course that may be related to the underlying dialects rather than to a genuinely phonological reason.
I think this is a very interesting proposal. But did you really mean that “with nouns word prosody and sentence prosody may conflate, whereas they are distinct with verbs”? I would think it’s the other way round: Since verbs are normally parts of clauses, their prosody is not as distinct as the prosody of nouns.
What I wanted to say is that if nouns are used in isolation, then word and utterance are co-extensive, hence no material difference between word prosody and sentence prosody. But I agree with your characterization.
I haven’t seen the Smith paper, but such phonological differences between form classes in some languages stand out. In Yoruba, verbs can be monosyllabic and must commence with a consonant; nouns must have at least two syllables and overwhelmingly start with a vowel (which may echo a long-abandoned noun class system). Consonant-initial nouns are largely loans. I remember seeing all this in EC Rowlands’ Teach Yourself Yoruba as a teen.
That’s very interesting. We have the same situation in Ancient Greek as well, where nouns have a fixed accent that only varies according to specific inflectional patterns. Conversely, verbal accentuation is recessive that stays as far from the end of the word as possible.
I wonder whether additional variation in the prosody of verbs could perhaps be a result of the fact that they tend to be a relatively popular host for clitics, which then shift the prosodic structure. Such shifts could then be solidified diachronically if the clitic grammaticalized as an inflectional marker.
Her observation holds for Ancient Greek as well, where the pitch accent for verbs is largely predictable (recessive) but has to be memorized for nouns. (Ancient Greek has a fairly weak distinction between nouns and adjectives.)
I haven’t seen Smith’s paper, but there is an important link here with work on phonology as a cue to word class status by Michael Kelly (Kelly 1992), as well as work on phonological and distributional cues in grammatical categorisation by Padraic Monaghan (Monaghan et al. 2005). I have observed a similar pattern in relation to nouns, verbs, and ideophones in Siwu — though I did not find obvious differences in phoneme frequencies, I did find that word length and certain phonotactic word shapes are statistically reliable cues to word class status in Siwu (Dingemanse 2011:104-5, 134-5, with apologies for citing myself).
Dingemanse, Mark. 2011. “The Meaning and Use of Ideophones in Siwu”. PhD dissertation, Nijmegen: Radboud University. http://thesis.ideophone.org/.
Kelly, Michael H. 1992. “Using Sound to Solve Syntactic Problems.” Psychological Review 99 (2): 349–364.
Monaghan, Padraic, Nick Chater, and Morten H. Christiansen. 2005. “The Differential Role of Phonological and Distributional Cues in Grammatical Categorisation.” Cognition 96 (2): 143–182.