[HPSG-L] Selection of phonology in nonlocal dependencies and raising

Tue Mar 1 16:28:58 UTC 2016

Hi,

For what it’s worth, Geoff Pullum and colleagues have argued strenously against the idea that syntax is sensitive to phonology, e.g.:

	http://web.stanford.edu/~zwicky/4PPFS.pdf

and here’s an implementation of (a version of) their notion of shape constraints in HPSG:

	http://web.stanford.edu/group/cslipublications/cslipublications/HPSG/2/Ash-Klein-pn.pdf

--
Rob Malouf <rmalouf at mail.sdsu.edu>
Department of Linguistics and Asian / Middle Eastern Languages
San Diego State University

> On Feb 29, 2016, at 10:28 AM, Dan Flickinger <danf at stanford.edu> wrote:
> 
> Hi Stefan -
> 
> I have been discussing with Paul Kay the issues you raised before the new year about SBCG, regarding some consequences of dropping the SYNSEM feature and adding the MOTHER feature.  I hope you can roll back the time machine to let me pick up this thread.
> 
> First, on the phonology front, I think the prediction you note is a good one, that the PHON feature can be referred to in syntax.  Even in English, we see at least two phenomena that benefit from this visibility, one involving the choice of determiner "a" vs. "an", and the other the voicing of possessive "s" when analyzed as a phrasal clitic.  Without the PHON feature, it is hard to see how to ensure that "an" appears when the next word in the noun phrase has a vocalic onset, and "a" when the onset is consonantal.  With PHON available, the word "an" can constrain its SPEC value (constraints on the head of the head-specifier construction) to have a PHON which has a vocalic onset, and similarly for "a".  A second such phenomenon is the voicing of the "'s" marking the possessive of singular NPs: even though this contrast is not reflected in the orthography, this morpheme has to be voiced if the coda (the final phoneme) of the preceding NP is voiced, and voiceless otherwise.  This argument is relevant as long as the possessive morpheme is treated as a phrasal clitic, an analysis motivated well in Anderson (2013): http://cowgill.ling.yale.edu/sra/elsj.pdf.  Here again, as long as the PHON feature is present in what a selector can constrain, the voiced and unvoiced variant entries of the possessive clitic can constrain their SPR value (treating the clitic as the head) to have the appropriate voiced/unvoiced coda.
> 
> I see that these two examples do not satisfy your wish for a phenomenon that imposes a phonologoical constraint even through an unbounded dependency, but even the surface-level effects of these two examples involve syntactic elements that are not immediate lexical sisters, so the dependencies are non-local in these syntactic structures.
> 
> Regarding the addition of the MOTHER feature in SBCG, I see it as one reasonable approach to expressing the highly desirable property of locality in allowable syntactic constraints.  I agree with you that for this feature to achieve its intended purpose of hiding properties of daughters of a phrase from outside selection that are not identified with any features in MOTHER, one has to ensure that only the value of this MOTHER feature structure can be referenced, but that is how SBCG is set up: the selector features have as values lists of signs, and signs don't have access to daughters.   As you know, this locality requirement is expressed differently in grammars such as the English Resource Gramamr, which introduces an ARGS (`daughters') feature as a top-level attribute of `sign', and follows Pollard and Sag 94 in having the values of the selector features be lists of synsem objects (a stronger constraint than in SBCG), thus preventing access to that ARGS feature for selection.  I find this ARGS mechanism for constructions to be at least as convenient as the SBCG architecture with its MOTHER feature, but I don't see any important difference in intent, namely to make explicit the strong hypothesis that dependencies imposed by words and phrases are strictly local.
> 
> Regarding the treatment of phonology-driven selection phenomena in this more classical HPSG architecture used in the ERG, I have moved the PHON feature into SYNSEM, but a more conservative approach might aim to  identify particular elements of a sign's phonology (perhaps just `onset' and `coda') that are made reentrant with features within SYNSEM.  This more conservative exposure of limited properties of phonology would be harder to do in SBCG where the full PHON value is necessarily visible for selection, so it will be interesting to know if others have found other phenomena that motivate selector access to more complex phonological properties.
> 
> Dan
> 
> 
> ________________________________________
> From: HPSG-L <hpsg-l-bounces at listserv.linguistlist.org> on behalf of Stefan Müller <stefan.mueller at fu-berlin.de>
> Sent: Friday, December 11, 2015 2:35 AM
> To: hpsg-l at listserv.linguistlist.org
> Subject: [HPSG-L] Selection of phonology in nonlocal dependencies and raising
> 
> Hi everybody,
> 
> I recently got some comments on the section on SBCG in my GT textbook. I
> rethought everything and have some questions that I cannot answer but
> maybe somebody on the list knows the answer.
> 
> One motivation for the change in feature geometry with locality of
> selection. This braught MOTHER. Interestingly SYNSEM is gone now and
> PHON is grouped with SYN and SEM. LOCAL is gone too. Rather than
> selecting for synsem objects and sharing local objects in nonlocal
> dependencies, complete signs are selected for and shared in nonlocal
> dependencies.
> 
> MOTHER was introduced to exclude the selection for arguments of
> arguments of arguments. As with computational complexity, I think that
> this should not be hardwired in the grammar formalism, the fact that we
> do not select arguments of arguments is just a fact about what the
> theories do. We do not have to state explicitely everything that is
> impossible.
> 
> But let's assume we think that MOTHER should be there because of
> locality issues. Wouldn't it be a problem then that a head that is far
> away can select the phonology of one of its arguments?
> 
> In the approach to raising in SBCG the subject of the downstairs head is
> shared with the matrix subject. So "eat" can see the phonology of "Kim":
> 
> Kim can eat apples.
> 
> Of course we can have long chains of raising verbs. Question: Are there
> languages that show phonological effects accross several words? And if
> so, does it help to have a head that selects for the phonology of a
> phrase far away or should these phonological effects be treated on the
> phrasal level?
> 
> The second issue is nonlocal dependencies:
> 
> Bagels, I think that Sandy likes.
> 
> "likes" can see the phonology of "bagels", as can "that" as can "think".
> In principle there could be languages that require that the filler has
> three vowels in it or anything like this.
> 
> Are there languages that have phenomena in which the phonology of the
> filler affects elements at the extraction path?
> 
> 
> The stuff is written down more carefully here (Section 10.6.2) and will
> be updated depending on the outcome of this discussion.
> 
> http://hpsg.fu-berlin.de/~stefan/Pub/grammatical-theory.html
> 
> 
> Thank you very much for your feedback and possibly for references to
> work that describes relevant phenomena.
> 
> Best wishes
> 
>        Stefan
> 
> --
> PGP welcome
> 
> Stefan Müller       Tel: (+49) (+30) 838 52973
>                    Fax: (+49) (030) 838 4 52973
> Institut für Deutsche und Niederländische Philologie
> Deutsche Grammatik
> Habelschwerdter Allee 45
> 14 195 Berlin
> 
> http://hpsg.fu-berlin.de/~stefan/
> 
> http://langsci-press.org/
> 
> http://hpsg.fu-berlin.de/Projects/CoreGram.html
> _______________________________________________
> HPSG-L mailing list
> HPSG-L at listserv.linguistlist.org
> http://listserv.linguistlist.org/mailman/listinfo/hpsg-l
> 
> _______________________________________________
> HPSG-L mailing list
> HPSG-L at listserv.linguistlist.org
> http://listserv.linguistlist.org/mailman/listinfo/hpsg-l