fonolog

Spoiling the party

fonolog — Tue, 06 Feb 2007 09:48:00 GMT

Harry van der Hulst seemed afraid that his talk (mp3; handout) at the CUNY Phonology Forum on Precedence would spoil the party: the other kids wanted to play with this new toy they had just discovered, the arrow (of time), and he wanted to throw it out of the window.

In any case, this is a very clear presentation of his idea why we should get rid of the arrow: because it is superfluous if we posit enough structure; for some reason he includes also subsegmental structure in this, but it is not completely clear to me why, since this hardly seems temporally ordered. This is a valid point, although it raises many issues, e,g, with respect to autosegmental structure, constraints on adjacent coda-onset combinations (sonority sequencing) etc.

It would also be interesting to compare Van der Hulst's Radical CV approach to Lowenstam and Scheer's CVCV theory, which, especially in Scheer's version seems to have come to quite the opposite conclusion: given the redundancy of tree structures and linear ordering, CVCV tries to do away with the former. On which criteria could we decide which of those two is the right move?

Immediate precedence

fonolog — Sun, 04 Feb 2007 21:42:00 GMT

All papers from the CUNY Phonology Forum in January are now online. I will no doubt be reading several of them in the following days.

I have started by reading Paul de Lacy's contribution, The formal properties of phonological precedence, which I found quite exciting. De Lacy proposes that phonology does not incorporate the notion of transitive precedence, but only the intransitive version (immediate precedence). This has certain interesting implications, e.g. for our theory of constraints: it accounts for the fact that constraints of grammars never seem to refer to the fact that for two segments A and B, A needs to precede B, irrespective of the number of segments between those two. (E.g. a constraint which would say that an /n/ in a word needs to be preceded by a /t/, even if those segments are separated by a number of syllables. Adjacency is the only thing that counts.

Another interesting property is that it is much easier to implement this idea in a containment model of faithfulness than in a correspondence model. The reason for these is deletion: suppose we only have immediate precedence, and a string /xyz/ from which we delete /y/. Now all underlying immediate precedence relations are destroyed, but this means that /xz/ is equally faithful as /zx/, at least in Correspondence Theory. Under Containment, this does not necessarily hold, since /y/ will still be contained in the output, as are the precedence relations.

A similar problem arises for epenthesis: suppose we have /xy/ underlyingly, and apply epenthesis by /z/. Again, this destroys the underlying immediate precedence relation, so that /xzy/ and /yzx/ are equally faithful. This is actually also a problem for a containment implementation of this idea, although there might be a few ways to circumvent this. For instance, the problem only arises for strings of length 2: as soon as we add extra segments to the string, this is no longer true: /axyb/ -> [axzyb] is still more faithful than /axyb/ -> [ayzxb]. And suppose we have some form of marker for the initial and final segment of the word, then these can attract faithfulness even in the shorter cases. This idea also applies to a correspondence implementation of De Lacy's proposal.

There are several other problems, e.g. the fact that somehow in the end we need to make sure that we have a total order (since words need to be pronounced). De Lacy assumes that the reason for this is phonetics: if the output of phonology in unpronounceable (i.e. if it cannot be translated to a total order in the phonetics) the derivation crashes, or is sent back to the Eval function. This is not very elegant, and it raises a question of design: why would phonology have this relation of immediate precedence (only) if it needs to talk eventually to a module which needs total orderings? In what sense is the idea of 'crashes' in phonetics not a covert way of still using total orderings also in the phonology?

All in all, this is a very thought-provoking paper, one of the most interesting contributions to the theory I have seen in quite some time.

Universality and variation in syllable structure

fonolog — Sun, 14 Jan 2007 13:21:00 GMT

Shanti Úlfsbjörninn discusses Floating Consonants in French: the need for the skeleton in input (and related issues). The author warns: "this is nothing but an undergraduate essay with a suggestion, not the last word on any particular topic". The suggestion is that we need underlying syllabification in order to understand the behaviour of French liaison consonants as well as the Spanish diminutive. This is a good suggestion, but Úlfsbjörninn assumes that the idea of underlying syllabification is itself new or controversial to OT. I do not think this is true; the real problem is Richness of the Base, which would force us to assume an underlying syllabification /at.a/ for a monomorphemic form; since we would also need faithfulness constraints (if these do not exist, it does not make sense to assume underlying syllable structure in the first place), this would predict that there are languages in which [a.ta] could contrast with [at.a]. "If onsets always maximise over codas then an intervocal consonant word-medially will always be an onset (Kager 1999:95). To attribute this consonant to an onset by the use of freely rankable constraints is redundant and generally flawed as it would be nothing else in the universe anyway." This is not true: it is not redundant, since we will also otherwise need to clarify why some languages need onsets and some (other) languages disallow codas. So the two constraints are not redundant. However, this universal poses a real puzzle: how can we set up a theory which is strict enough to disallow languages which we know do not exist, while allowing for a natural analysis of facts such as those in French and Spanish.

Suffixes without constraint rankings

fonolog — Sun, 14 Jan 2007 12:14:00 GMT

Hideki Zamma applies Partial Ordering Theory to the facts of English stress in his Categorical and Non-categorical Variation in English Stress Assignment. An important observation of this paper is that some suffixes (-ary, -ate, -ize, -ite) show alternating behaviour:

(7)	a.	cólumbary, óctonary, sédentary, vóluntary
	b.	abecédary, annivérsary, conséctary, duodénary, quatérnary, septuagénary
	c.	ántiquary, plebíscitary, resíduary
	d.	compleméntary, documéntary, eleméntary, evangelístary, referéndary

These are interesting facts hitherto ignored. Zamma's proposal is that part of the constraints for English stress are unordered, for instance those differentiating between Strong and Weak Retraction. Some suffixes are specified for a ranking giving Strong Retraction, some are specified for a ranking giving Weak Retraction. -ary etc. do not have a specified ranking. Of course, individual forms still need to choose a ranking, even if Zamma does not specify how this works: there is no variation between vóluntary and volúntary.

Marked morphology expressed by marked phonology

fonolog — Sat, 30 Dec 2006 12:43:00 GMT

Cristian Iscrulescu's USC dissertation is called The Phonological Dimension of Grammatical Markedness — a title which has of course attracted my attention. Iscrulescu studies the phenomenon that grammatically marked structures in some languages allow more phonologically marked structures; e.g. the plural forms are systematically more complex than the singulars. ?is examples are mainly from Old Saxon, Romanian and Mayak.

In order to analyse this, Iscrulescu proposes that there are licensing constraints which say e.g. 'palatalised consonants need to occur in the plural' (if a consonant is palatalised, it needs to occur in a plural form).

Iscrulescu only discusses one possible consequence of this: the morphosyntactically unmarked forms disallow certain phonological complexities (and therefore we find reduction, etc.). Logically speaking, one might expect that a way of satisfying the licensing constraints would be to turn a phonologically complex singular into a morphological plural. This is an absurd result, but it is not entirely clear how it can be prevented. (Well, it is clear how this can be done, viz. by assuming absolute faithfulness of Morphological Structure, but Iscrulescu does not say so explicitly.)

My own preferred solution to this type of problem is based on an Items-and-Arrangement view of morphology, and the concept of morpheme expression: the plural is underlyingly marked and there is a constraint Express-plural (a special type of faithfulness: at least some part of the plural morpheme needs to be expressed in the phonology). Iscrulescu briefly discusses this alternative (on pp. 225–229), but this is not the most satisfactory part of the thesis. His arguments are:

Realize-Morpheme approaches are "blind as to the specific way in which the plural morpheme is realized, as long as there is detectable phonological material affiliated to it." I consider this almost to be an argument in favour of Realize-Morpheme: the specific way in which the morpheme is realized is now left to the (hopefully independently established) phonological well-formedness of the language in question.
The other objection is that RM "potentially overgenerate[s]". The idea is that if we have RM-singular, we could rank this constraint topmost, and generate a language in which the singular is more complex than the plural. Iscrulescu attributes the idea to my friend Nina Topintzi, but I think it does not hold. The fact that there are apparently no languages with more complex singulars than plurals only means that either (i) singular is not a feature (number features are monovalent) and need not be realised, or (ii) we need to posit a universal ranking RM-plural >> RM-singular, exactly parallel to a similar stipulation that Iscrulescu needs to make.

The most crucial evidence in favour of Iscrulescu's approach and against RM would come from a language in which the 'more marked' phonology means less structure; for instance schwa is arguably 'marked' in the sense that not all languages have it, so there needs to be a constraint against it. On the other hand, it has also arguably less structure than other vowels. So a language in which schwa can only occur in plurals but not in singulars could not be analysed in an RM approach, but it could be understood in terms of Licensing of marked structures in morphologically complex positions. Unfortunately, Iscrulescu does not provide such examples; in all of his examples, the morphologically complex structures also involve more phonological material.