GLOW 1998 -- The phonology talks

Marc van Oostendorp

This report I wrote for GLOT International

The emphasis of the GLOW session traditionally is on syntax rather than phonology. This yearâs conference featured two lectures on phonological structure and one on the interface between phonology and syntax. All three gave arguments in favour of an enriched theory of phonological representations rather than concentrating our research on the issue of rules versus constraints.

The talk by Marc van Oostendorp, Phonological feature domains and the content of epenthetic vowels, presented arguments that we need to take into account the morphological affiliation of phonological segments in order to account for root-controlled vowel harmony in Turkish and for the exceptional behaviour of epenthetic vowels in both Turkish and Icelandic.

Keren Rice and Trisha Causley concentrated on Asymmetries in Featural Markedness: Place of Articulation. The authors argued that there are two problems connected to the markedness of place features. On the one hand, there is a great deal of cross-linguistic consistency as to what patterns as unmarked phonologically. But on the other hand, it is not always the same feature that patterns as unmarked cross-linguistically. Rice and Causley solved this by presenting a theory of place both for the consonantal and for the vocalic domain that involved different types of unmarkedness. For instance, central vowels and velar consonants are represented by an empty place node. This representation is quite literally unmarked in the sense that it involves a minimal structure. For this reason central vowels and velar consonants count as unmarked in these segments. On the other hand, certain languages do not allow these segments, because they disallow completely empty place nodes. Rice and Causley presented their analysis in terms of Optimality Theory but it is reasonably clear that their approach could be transferred without too many problems to other theories of phonology. This does not mean that their proposal will be uncontroversial because it involves a few representational oddities that may not be acceptable to all scholars: e.g. the features [Labial] and [Dorsal] are subsumed under a class node [Peripheral].

'The `interface talk' by Gorka Elordieta dealt with The phonological import of syntactic features. Elordieta claims that prosodic phonology needs to be supplemented with another set of syntactic domains for phonological processes. Two adjacent syntactic heads which stand in a feature-checking relation can form such a domain. The data on which Elordieta bases his claims are vowel assimilation in Basque and liaison in an informal register of French. In the talk he showed how standard assumptions of prosodic phonology are insufficient: the phenomena take place between a lexical head and certain functional heads, but not others. According to Elordieta the reason behind the asymmetry is that in the latter cases there is no syntactic feature checking going on whereas in the former, there is.

Phonology Workshop: Opacity

The phonology-related talks in the main session of GLOW dealt mainly with issues of representation; the phonology workshop on opacity on the other hand went straight into the heart of the discussion between proponents of bistratal Optimality Theory and the defendants of polystratal generative phonology based on extrinsically ordered rules. Since the two extremes were both represented, the discussion was quite lively. As Morris Halle pointed out during this day, the phenomenon of opacity was one of the driving forces for setting up the mechanism of extrinsic rule ordering. On the other hand, the existence of these phenomena pose serious problems for Optimality Theory, because their analysis appears to crucially involve some additional levels of representation beyond the input and the output. Several ways of dealing with these phenomena have been proposed within the OT literature, and it was one of the purposes of this day to evaluate those proposals.

The contribution of David Odden was entitled Cyclicity and Counterfeeding in Kimatuumbi, but Odden was quick to point out that the paper didnât deal exclusively with either cyclicity, counterfeeding or Kimatuumbi. His data were drawn from several (Bantu) languages and he showed that an OT analysis of these facts would involve a quite heavy machinery, including Sympathy Theory, two-level constraints and output-output relations. Neither of these three `supplementsâ to OT are sufficient to handle all the relevant facts. An analysis of Bantu in terms of extrinsically ordered rules is less problematic, but Mr. Odden seemed hesitant to draw the conclusion out of this that such an analysis would therefore be preferable. From an empirical point of view, both theories seem to be equally able to handle the relevant facts.

Marc van Oostendorp claimed in his talk Non-derivational opacity in allomorph selection that there are certain types of opacity which can be handled in a sympathy approach but not in a rule-based approach; we have cases in Dutch where a form behaves phonologically as if it has chosen the unmarked allomorph, even though on the surface the form displays the marked allomorph of a given affix.

Ronald Sprouse unfortunately was not able to be physically present at the workshop. His paper was read by Orhan Orgun. Mr. Sprouses paper Enriched Input Sets as a source of opacity in Optimality presented a tristratal version of Optimality Theory, in which we have two passes of generation; the first generator function is only able to add material to the input, not to delete anything (this function is therefore comparable to the original definition of Gen in Prince and Smolensky 1993); the second function can apply all sorts of operations to the output of the first Gen. Furthermore there are sets of faithfulness constraints relating, input, enriched input and output. Mr. Sprouse demonstrates that this extension of Optimality Theory (which is more restricted than Sympathy Theory) can handle the data of Turkish epenthesis and deletion and the complicated facts of Yowlumne (=Yawelmani).

Orhan Orguns own paper Phonological Opacity and Synchronically Arbitrary Alternations: What Two-Level Phonology can contribute to OT also proposed to extend the Gen function of Optimality Theory, but in a completely different manner. Mr. Orgun tries to give an articulated, language-specific implementation of the generator, formulated in terms of the two-level theory of phonology that Mr. Orgun developed in his thesis, based on the work of Koskenniemi. Mr. Orgun proposes that the generator function of a given language may only generate those candidates that conform to this two-level system. He claims that this is only a minimal extension to the theory (we may need language-specific rules anyway) and that the two-level system is rich enough to account for complex cases of opacity such as the ones in Hebrew, Icelandic, Yowlumne and Kashaya. The question then arises what the division of labour is between the two-level generator and the evaluator function; Mr. Orgun left this question open to further research.

The talk by Ania Lubowicz was called Derived Environment Effects in OT. Ms. Lubowicz argued that Derived Environments are by definition in most cases also instances of opacity: certain segments do not undergo a given phonological process, because they do not occur in a derived context. Ms. Lubowicz wants to analyse this as the result of constraint conjunction; a faithfulness constraint is conjoined with a well-formedness constraint in the domain of a segment. If the faithfulness constraint is violated in a given (derived) context, we are not allowed to violate the well-formedness constraint at the same time. Therefore, the segments in derived environments are likely to undergo a certain change in order to become more well-formed. Yet segments in underived environments do not have to violate the same faithfulness constraint, and therefore it becomes irrelevant wether or not they violate the well-formedness constraint: they can stay the way they are. Ms. Lubowicz showed that this analysis works both for phonologically and for morphologically derived environment effects in languages like Polish and Slovak.

Bill Idsardi studied some formal properties of Opacity, sympathy and derivations. The emphasis was on the treatment of opacity within sympathy theory. Taking the analysis of Tiberian Hebrew vowel epenthesis and glottal stop deletion, and Icelandic j-deletion and vowel epenthesis as a starting point, Mr. Idsardi showed that current Sympathy accounts are less parsimonious (there are more constraints/rules at stake), more abstract (the analysis involves a larger numer of relevant but non-surfacing representations), more delicate (changing the grammar slightly may have dramatic consequences for candidates which seem unaffected by the basic analysis) than a rule-based accounts. Furthermore, Sympathy Theory also is not more restricted than rule-based theory according to Mr. Idsardi: also instances of the so-called Duke-of-York Gambit may be analysed using sympathy. The latter point was also made earler during the day by David Odden.

The last talk of the workshop was Morris Halle's English Stress 1968-1998. In this talk, Mr. Halle gave an overview of the developments of the theory of English stress since the appearance of The Sound Pattern of English. The main development is an enlarged insight into the representation of phonological structure. The idea of grouping segments into syllables, projecting syllables onto an independent stress plane and enriching that plane with a limited number of boundary symbols, allows us to formulate the relevant rules in a way that is maximally simple. The theory thus established is supplemented with a small number of exception markings. Mr. Halle furthermore adopted Burzio's proposal that certain word-final `super-light' syllables such as -ure and -y do not get projected to the stress plane at all. This further simplified the rule-based account of English word-level stress.