From mcginnis at ucalgary.ca Wed May 22 15:00:48 2002 From: mcginnis at ucalgary.ca (Martha McGinnis) Date: Wed, 22 May 2002 09:00:48 -0600 Subject: Carson Schütze: gender copying and vocabulary insertion Message-ID: Hi everyone, I have a question about the treatment of gender concord in DM, which either wasn't answered in the earlier discussion of gender in DM more generally, or else I'm not able to extract the answer from the many larger issues in that discussion. Here's the context. I've been reading Roland Pfau's dissertation on analyzing speech errors in DM (which I highly recommend, by the way, especially for those interested in "architecture" issues). He reports a striking finding about noun substitution slips in German and their effect on gender marking in the containing DP (which I'll omit for brevity). He then proposes an explanation in DM that rests on the following claim that he attributes, I think correctly, as a principle endowed upon DM by its Creators: In the MS/SpellOut component, all operations on terminal nodes, including fusion, feature insertion (if that's how you do subject-verb agreement), merger, feature copying, etc., must precede Vocab Insertion. This has the consequence that if, just at the point of Vocab Insertion, the target noun root gets erroneously replaced with a different root of a different gender, it could not happen that the gender inflections elsewhere in that DP would be adjusted to match the gender of the intruder root. That's because once we begin Vocab Insertion, we aren't allowed to do feature copying anymore, so in particular we aren't allowed to copy the newly-arrived gender feature of the intruder noun onto the determiner & adj next to it. Thus, even if that D and A haven't yet themselves been spelled out (undergone Vocab Insertion), they'll be stuck spelling out as the forms appropriate to the gender of the original target noun, in this case mismatching the (intruder) noun actually uttered. There are all sorts of interesting issues here, but the one I'm focusing on at the moment is this: What was the motivation for prohibiting operations like feature copying after Vocab Insertion has begun? Was this simply an a priori way to limit the power of the system, perhaps conceptually motivated, or did they have in mind particular conceivable empirical phenomena that never occur that this prohibition was designed to rule out? Apologies if this is in the published literature somewhere. thanks, Carson From mcginnis at ucalgary.ca Thu May 23 15:04:47 2002 From: mcginnis at ucalgary.ca (Martha McGinnis) Date: Thu, 23 May 2002 09:04:47 -0600 Subject: Heidi Harley: gender copying and vocabulary insertion (reply to Carson Schutze) Message-ID: Hi all -- this is a v. interesting question, which in general i think has a straightforward answer, but in this specific is tricky. Straightforwardly, if Vocab Insertion discharges features by *replacing* them, then feature copying, fusion, etc, will not be prohibited but just impossible. The intuition would be that the point of Vocab Insertion is to rewrite bundles of morphosyntactic features as peices of phonology. Perhaps feature copying could happen after vocab insertion has begun, assuming VI applies bottom-up successive-cyclically, but it would only apply to and from nodes that have not yet been discharged -- effectively it could only apply to nodes that hadn't been involved in VI yet. On the other hand, though, if Jonathan B. is right (see 1999 Maryland Morphology Mayfest paper on Itel'men), things are a bit trickier particularly with respect to gender/class features: according to him, such features are fundamentally different from other morphosyntactic features. Because they belong to particular roots, he argues they must be, in effect, 'phonological', and do not exist before Vocab Insertion. If that's so, sensitivity to such features, copying 'em and so forth, can ONLY happen *after* VI. (He argues that these are the only cases of countercyclic feature sensitivity, i.e. 'inwards' feature sensitivity, and that this type of sensitivity is explained if class/gender features are phonological.) But seems like that's incompatible with Roland's nifty-sounding explanation for gender mismatches after mistakes in VI. If Jonathan's wrong, though, and all features are truly morphosyntactic, including gender&class, and if VI does discharge features by replacing 'em, then we can understand why feature copying isn't possible after VI has begun. I think. :) hh --------------------------------------------------------------------- Heidi Harley Department of Linguistics Douglass 200E University of Arizona Tucson, AZ 85721 Ph: (520) 626-3554 Fax: (520) 626-9014 hharley at u.arizona.edu From mcginnis at ucalgary.ca Thu May 23 18:48:32 2002 From: mcginnis at ucalgary.ca (Martha McGinnis) Date: Thu, 23 May 2002 12:48:32 -0600 Subject: Carson Schutze: gender copying and vocabulary insertion (reply to Heidi Harley) Message-ID: Thanks for the already very helpful info, I'll read that paper of Jonathan's. But meanwhile I perhaps should have given Roland's entire result, to make the nature of the problem clearer. Noun substitution errors can be of two kinds: either you slip to a semantically-related noun or to a phonologically similar noun. The finding is that if you slip to something semantically related, gender inflection on D, A etc. *does* get adjusted to match it, whereas if you slip to a phonologically similar noun, gender does *not* get adjusted. The intuition, which he works out, is that meaning-based word selection (hence confusion) happens 'early' but form expression (hence confusion) happens 'late', and somewhere between the two is when gender features propagate from the head N to the rest of the DP. He already has to propose a change to the DM architecture to make this work (namely to carry pointers to roots through the whole derivation, i.e. as part of the numeration), and I'm playing with whether we can do without that change, if we can monkey with the timing of the copying of gender features at MS (which he was assuming we can't, following the edict that I referred to in my first message). The intuition would be that the 'early' slips happen at the point of choosing among vocabulary items, after which gender feature propagation happens 'as normal' according to Jonathan's treatment, as I now understand it. The late slips happen after selection of vocab items, in the course of actually phonologically realizing the root, i.e. it's a slip from one string of phonemes to a nearby similar string of phonemes (very crudely speaking). This ph. realization happens after all the feature stuff, including gender feature copying. Does anyone see big problems in principle with this alternative approach? thanks, Carson From mcginnis at ucalgary.ca Fri May 24 03:22:20 2002 From: mcginnis at ucalgary.ca (Martha McGinnis) Date: Thu, 23 May 2002 21:22:20 -0600 Subject: Andrew Carstairs-McCarthy: gender copying and vocabulary insertion (reply to Heidi Harley) Message-ID: Hello Jonathan Bobaljik's 1999 Maryland Morphology Mayfest paper, mentioned by Heidi in connection with gender copying, contains ideas that bear on the general question: When a Vocabulary item is being chosen for insertion at a given word-structure node, what morphosyntactic information (if any) *not located at that node* can influence the choice? One answer, which uses the metaphor of 'discharge', is: 'Only information that has not yet been discharged through phonological spell-out'. Another answer is: 'Only information encoded by material already phonologically present in the wordform, e.g. by affixes already added'. DM of course prefers the first answer, but the second is implied in Shelly Lieber's approach and also, I guess, in Kiparskyan level-ordered morphology. Both have a ring of plausibility about them. Which is correct? Or are both in some degree correct -- which may suggest that what is really going on is something different from either? My article 'Grammatically conditioned allomorphy, paradigmatic structure, and the Ancestry Constraint' (Transactions of the Philological Society 99 (2001), 223-45) addresses this directly, with discussion of Jonathan's Itelmen paper. I propose a constraint on grammatically conditioned allomorphy which, though different from Jonathan's, is consistent with the spirit of DM, I think. Any comments and criticisms will be welcome. Andrew -- Andrew Carstairs-McCarthy Professor and Head of Department Department of Linguistics, University of Canterbury, Private Bag 4800, Christchurch, New Zealand phone (work) +64-3-364 2211; (home) +64-3-355 5108 fax +64-3-364 2969 e-mail a.c-mcc at ling.canterbury.ac.nz http://www.ling.canterbury.ac.nz/adc-m.html From mcginnis at ucalgary.ca Sun May 26 11:14:19 2002 From: mcginnis at ucalgary.ca (Martha McGinnis) Date: Sun, 26 May 2002 05:14:19 -0600 Subject: Julie Legate: gender copying and vocabulary insertion (reply to Andrew Carstairs-McCarthy) Message-ID: Hello, > When a Vocabulary item is being chosen for > insertion at a given word-structure node, what morphosyntactic > information (if any) *not located at that node* can influence the > choice? One answer, which uses the metaphor of 'discharge', is: > 'Only information that has not yet been discharged through > phonological spell-out'. Another answer is: 'Only information > encoded by material already phonologically present in the wordform, > e.g. by affixes already added'. DM of course prefers the first > answer, but the second is implied in Shelly Lieber's approach and > also, I guess, in Kiparskyan level-ordered morphology. Both have a > ring of plausibility about them. Which is correct? Or are both in > some degree correct -- which may suggest that what is really going on > is something different from either? I thought a paper of mine in MITWPL 1999 would be relevant to mention (it's cited by Bobaljik in at least some versions of his Itelmen paper) that argues for something very much like the "other answer" you gave. Essentially, when vocabulary insertion operates on a bundle of features the features can either be (1) spelled out phonologically, (2) fissioned off, or (3) deleted. In the case of either (1) or (2) the features are still available for morphological operations. In the case of (3), of course, they are not. Now, I must say that the paper was on Irish agreement and pro-drop, a notoriously controversial topic, so perhaps not the most solid evidence for this position one would hope for ... > My article 'Grammatically conditioned allomorphy, paradigmatic > structure, and the Ancestry Constraint' (Transactions of the > Philological Society 99 (2001), 223-45) addresses this directly, with > discussion of Jonathan's Itelmen paper. I propose a constraint on > grammatically conditioned allomorphy which, though different from > Jonathan's, is consistent with the spirit of DM, I think. I look forward to reading it. Best, Julie ------------------------ Julie Anne Legate MIT Linguistics From mcginnis at ucalgary.ca Wed May 22 15:00:48 2002 From: mcginnis at ucalgary.ca (Martha McGinnis) Date: Wed, 22 May 2002 09:00:48 -0600 Subject: Carson Schütze: gender copying and vocabulary insertion Message-ID: Hi everyone, I have a question about the treatment of gender concord in DM, which either wasn't answered in the earlier discussion of gender in DM more generally, or else I'm not able to extract the answer from the many larger issues in that discussion. Here's the context. I've been reading Roland Pfau's dissertation on analyzing speech errors in DM (which I highly recommend, by the way, especially for those interested in "architecture" issues). He reports a striking finding about noun substitution slips in German and their effect on gender marking in the containing DP (which I'll omit for brevity). He then proposes an explanation in DM that rests on the following claim that he attributes, I think correctly, as a principle endowed upon DM by its Creators: In the MS/SpellOut component, all operations on terminal nodes, including fusion, feature insertion (if that's how you do subject-verb agreement), merger, feature copying, etc., must precede Vocab Insertion. This has the consequence that if, just at the point of Vocab Insertion, the target noun root gets erroneously replaced with a different root of a different gender, it could not happen that the gender inflections elsewhere in that DP would be adjusted to match the gender of the intruder root. That's because once we begin Vocab Insertion, we aren't allowed to do feature copying anymore, so in particular we aren't allowed to copy the newly-arrived gender feature of the intruder noun onto the determiner & adj next to it. Thus, even if that D and A haven't yet themselves been spelled out (undergone Vocab Insertion), they'll be stuck spelling out as the forms appropriate to the gender of the original target noun, in this case mismatching the (intruder) noun actually uttered. There are all sorts of interesting issues here, but the one I'm focusing on at the moment is this: What was the motivation for prohibiting operations like feature copying after Vocab Insertion has begun? Was this simply an a priori way to limit the power of the system, perhaps conceptually motivated, or did they have in mind particular conceivable empirical phenomena that never occur that this prohibition was designed to rule out? Apologies if this is in the published literature somewhere. thanks, Carson From mcginnis at ucalgary.ca Thu May 23 15:04:47 2002 From: mcginnis at ucalgary.ca (Martha McGinnis) Date: Thu, 23 May 2002 09:04:47 -0600 Subject: Heidi Harley: gender copying and vocabulary insertion (reply to Carson Schutze) Message-ID: Hi all -- this is a v. interesting question, which in general i think has a straightforward answer, but in this specific is tricky. Straightforwardly, if Vocab Insertion discharges features by *replacing* them, then feature copying, fusion, etc, will not be prohibited but just impossible. The intuition would be that the point of Vocab Insertion is to rewrite bundles of morphosyntactic features as peices of phonology. Perhaps feature copying could happen after vocab insertion has begun, assuming VI applies bottom-up successive-cyclically, but it would only apply to and from nodes that have not yet been discharged -- effectively it could only apply to nodes that hadn't been involved in VI yet. On the other hand, though, if Jonathan B. is right (see 1999 Maryland Morphology Mayfest paper on Itel'men), things are a bit trickier particularly with respect to gender/class features: according to him, such features are fundamentally different from other morphosyntactic features. Because they belong to particular roots, he argues they must be, in effect, 'phonological', and do not exist before Vocab Insertion. If that's so, sensitivity to such features, copying 'em and so forth, can ONLY happen *after* VI. (He argues that these are the only cases of countercyclic feature sensitivity, i.e. 'inwards' feature sensitivity, and that this type of sensitivity is explained if class/gender features are phonological.) But seems like that's incompatible with Roland's nifty-sounding explanation for gender mismatches after mistakes in VI. If Jonathan's wrong, though, and all features are truly morphosyntactic, including gender&class, and if VI does discharge features by replacing 'em, then we can understand why feature copying isn't possible after VI has begun. I think. :) hh --------------------------------------------------------------------- Heidi Harley Department of Linguistics Douglass 200E University of Arizona Tucson, AZ 85721 Ph: (520) 626-3554 Fax: (520) 626-9014 hharley at u.arizona.edu From mcginnis at ucalgary.ca Thu May 23 18:48:32 2002 From: mcginnis at ucalgary.ca (Martha McGinnis) Date: Thu, 23 May 2002 12:48:32 -0600 Subject: Carson Schutze: gender copying and vocabulary insertion (reply to Heidi Harley) Message-ID: Thanks for the already very helpful info, I'll read that paper of Jonathan's. But meanwhile I perhaps should have given Roland's entire result, to make the nature of the problem clearer. Noun substitution errors can be of two kinds: either you slip to a semantically-related noun or to a phonologically similar noun. The finding is that if you slip to something semantically related, gender inflection on D, A etc. *does* get adjusted to match it, whereas if you slip to a phonologically similar noun, gender does *not* get adjusted. The intuition, which he works out, is that meaning-based word selection (hence confusion) happens 'early' but form expression (hence confusion) happens 'late', and somewhere between the two is when gender features propagate from the head N to the rest of the DP. He already has to propose a change to the DM architecture to make this work (namely to carry pointers to roots through the whole derivation, i.e. as part of the numeration), and I'm playing with whether we can do without that change, if we can monkey with the timing of the copying of gender features at MS (which he was assuming we can't, following the edict that I referred to in my first message). The intuition would be that the 'early' slips happen at the point of choosing among vocabulary items, after which gender feature propagation happens 'as normal' according to Jonathan's treatment, as I now understand it. The late slips happen after selection of vocab items, in the course of actually phonologically realizing the root, i.e. it's a slip from one string of phonemes to a nearby similar string of phonemes (very crudely speaking). This ph. realization happens after all the feature stuff, including gender feature copying. Does anyone see big problems in principle with this alternative approach? thanks, Carson From mcginnis at ucalgary.ca Fri May 24 03:22:20 2002 From: mcginnis at ucalgary.ca (Martha McGinnis) Date: Thu, 23 May 2002 21:22:20 -0600 Subject: Andrew Carstairs-McCarthy: gender copying and vocabulary insertion (reply to Heidi Harley) Message-ID: Hello Jonathan Bobaljik's 1999 Maryland Morphology Mayfest paper, mentioned by Heidi in connection with gender copying, contains ideas that bear on the general question: When a Vocabulary item is being chosen for insertion at a given word-structure node, what morphosyntactic information (if any) *not located at that node* can influence the choice? One answer, which uses the metaphor of 'discharge', is: 'Only information that has not yet been discharged through phonological spell-out'. Another answer is: 'Only information encoded by material already phonologically present in the wordform, e.g. by affixes already added'. DM of course prefers the first answer, but the second is implied in Shelly Lieber's approach and also, I guess, in Kiparskyan level-ordered morphology. Both have a ring of plausibility about them. Which is correct? Or are both in some degree correct -- which may suggest that what is really going on is something different from either? My article 'Grammatically conditioned allomorphy, paradigmatic structure, and the Ancestry Constraint' (Transactions of the Philological Society 99 (2001), 223-45) addresses this directly, with discussion of Jonathan's Itelmen paper. I propose a constraint on grammatically conditioned allomorphy which, though different from Jonathan's, is consistent with the spirit of DM, I think. Any comments and criticisms will be welcome. Andrew -- Andrew Carstairs-McCarthy Professor and Head of Department Department of Linguistics, University of Canterbury, Private Bag 4800, Christchurch, New Zealand phone (work) +64-3-364 2211; (home) +64-3-355 5108 fax +64-3-364 2969 e-mail a.c-mcc at ling.canterbury.ac.nz http://www.ling.canterbury.ac.nz/adc-m.html From mcginnis at ucalgary.ca Sun May 26 11:14:19 2002 From: mcginnis at ucalgary.ca (Martha McGinnis) Date: Sun, 26 May 2002 05:14:19 -0600 Subject: Julie Legate: gender copying and vocabulary insertion (reply to Andrew Carstairs-McCarthy) Message-ID: Hello, > When a Vocabulary item is being chosen for > insertion at a given word-structure node, what morphosyntactic > information (if any) *not located at that node* can influence the > choice? One answer, which uses the metaphor of 'discharge', is: > 'Only information that has not yet been discharged through > phonological spell-out'. Another answer is: 'Only information > encoded by material already phonologically present in the wordform, > e.g. by affixes already added'. DM of course prefers the first > answer, but the second is implied in Shelly Lieber's approach and > also, I guess, in Kiparskyan level-ordered morphology. Both have a > ring of plausibility about them. Which is correct? Or are both in > some degree correct -- which may suggest that what is really going on > is something different from either? I thought a paper of mine in MITWPL 1999 would be relevant to mention (it's cited by Bobaljik in at least some versions of his Itelmen paper) that argues for something very much like the "other answer" you gave. Essentially, when vocabulary insertion operates on a bundle of features the features can either be (1) spelled out phonologically, (2) fissioned off, or (3) deleted. In the case of either (1) or (2) the features are still available for morphological operations. In the case of (3), of course, they are not. Now, I must say that the paper was on Irish agreement and pro-drop, a notoriously controversial topic, so perhaps not the most solid evidence for this position one would hope for ... > My article 'Grammatically conditioned allomorphy, paradigmatic > structure, and the Ancestry Constraint' (Transactions of the > Philological Society 99 (2001), 223-45) addresses this directly, with > discussion of Jonathan's Itelmen paper. I propose a constraint on > grammatically conditioned allomorphy which, though different from > Jonathan's, is consistent with the spirit of DM, I think. I look forward to reading it. Best, Julie ------------------------ Julie Anne Legate MIT Linguistics