From: Przemyslaw Pawelec

                     The Trapped Sonorants in Polish

     The paper addresses the problem of the so called Trapped
 Sonorants in Polish, i.e. consonantal sonorants that appear in
 an onset between two consonants of lower sonority, as in e.g.
 krwi [krf'i] 'blood' and brwi [brv'i] 'brow' (both gen.sg). It
 is argued that they should not be treated on a par with other
 types of sonorants violating the Sonority Sequencing
 Generalization (SSG) in Polish. The argument is supported by a
 new observation about the structure of the offending onset
 clusters.

     The paper refers to a recent study of trapped sonorants in
 Rubach (1997), where a comparison is made of how two different
 theories, a derivational theory (DT) and the Optimality Theory
 (OT), handle the problem of sonorants that violate the SSG.
 When OT is found to be in some ways superior to DT, but at the
 same time incapable of accounting for the phenomenon of trapped
 sonorants, Rubach (1997) offers a new solution: Derivational
 Optimality Theory (DOT), which is a modified version of OT. The
 revision does not come without a cost, though - the
 introduction of a derivational step into OT undermines one of
 its basic tenets.

     The discussion in the present paper begins with the
 observation that in Polish there are four kinds of environment
 in which consonantal sonorants may occur with what appears to
 be an SSG violation:

 a)  word-initial, e.g. rtec 'mercury', lsnic 'shine';
      sonorants in this position are not transparent to Voice
      Assimilation, i.e. in Warsaw Polish there is no Voicing
      across word boundary before them.
 b)  word-final, e.g. lotr 'villain', organizm 'organism',
      which are transparent to Voice Assimilation.
 c)  word-medial between two syllable nuclei (and not adjacent
      to any of them, naturally)   are also transparent to Voice
      Assimilation.
 d)  in the onset of a word-initial syllable, between two
      consonants of smaller sonority; in this position the so
      called trapped sonorants are transparent to Voice
      Assimilation.

     In the DT analysis in Rubach (1997) sonorants in all four
 kinds of context are assumed to be extrametrical, i.e. they do
 not belong to any syllable and thus do not violate the SSG.
 Technically, they are not linked to the syllable node, but
 directly to the PW node. At the same time good use is made of
 the opportunities offered by the derivation: the difference in
 transparency between type (a) on the one hand and types (b),
 (c), (d) on the other is shown to be due to the difference in
 the stage of prosodic licensing, and, consequently, voice
 specification of sonorants: before the application of Final
 Devoicing and Voice Assimilation (Delinking and Spread) for
 type (a), and after it for types (b,c,d).

     As this is not available in OT (there is no derivation, so
 there is no way of distinguishing between a non-transparent
 sonorant in (a) and transparent sonorants in (b,c,d) by means
 of derivational stages) the desired effect is obtained by the
 appropriate ranking of constraints. In the OT analysis in
 Rubach (1997) word-initial sonorants (a) are forced into a
 syllable (a violation of the SSG), while final and medial
 sonorants (b and c) remain extrasyllabic, and are linked to PW
 node (a violation of Strict Layer). Unfortunately, type (d)
 poses a problem for this kind of analysis. Trapped sonorants
 are transparent, so they should be extrasyllabic and linked to
 PW node, like in types (b) and (c), but for reasons independent
 of the DT/OT distinction it is not possible: association lines
 would have to cross, which is not permitted by the principles
 of Autosegmental Phonology, a theory of representation that is
 assumed in both DT and OT. This leads the author to the
 postulation of a new theory, DOT, where the constraints are
 reranked after a derivational step.

     The argument rests on the assumption that there is nothing
 special about type (d), so no constraints that would refer
 specifically to (d) are justified. Thus it is argued that (d)
 should be accounted for by means of the reranking of the same
 constraints which account for types (a), (b) and (c), and this
 in turn is done by means of a derivational step.

     In this paper we will try to demonstrate that trapped
 sonorants in (d), in fact are different: their structure is
 very specific, strongly restricted, and what is more, it is
 contrary to what should be expected. Trapped sonorants in
 certain respects do not behave like transparent sonorants in
 (b) and (c), but like obstruents which impose stronger
 distributional restrictions on neighbouring sounds than "true"
 obstruents do in Polish. A new observation about the structure
 of the offending onset clusters is made and alternative
 solutions for a derivational theory and for the Optimality
 Theory are suggested.

 Reference:
 Rubach, Jerzy. 1997. Extrasyllabic Consonants in Polish:
 Derivational Optimality Theory. In: I. Roca (ed.) Derivations
 and Constraints in Phonology. Oxford: Clarendon Press: 551-581.