tag:blogger.com,1999:blog-5275657281509261156.post6573080592723647864..comments2024-03-28T04:04:55.806-07:00Comments on Faculty of Language: Why Morphology? (part deux)Norberthttp://www.blogger.com/profile/15701059232144474269noreply@blogger.comBlogger11125tag:blogger.com,1999:blog-5275657281509261156.post-62594604035615724742014-11-20T14:21:05.075-08:002014-11-20T14:21:05.075-08:00Yes, they do create it. But I guess I want to know...Yes, they do create it. But I guess I want to know WHY they do. What's the morphology doing for them. Here's one reason, because kids come with UGs that favor morphology. Thus, because they are good at it and UG favors it when they have the chance to construct Gs with overt morphology they do so. Ok, next question: why does UG care about Gs with morphology? Now one answer is that UG doesn't care about morphology at all. Morphology is just what you get when humans form constituents from strings. If that's so, we should find morphology in non-linguistic systems. Does one? I don't know. But one possibility is that UG "likes" Gs with overt morphology because it promotes another important feature of language; large vocal growth. How does it do this? By allowing the natural talents that kids have (find and create morphology) to ornament Gs with these properties so that large vocals will be easily acquirable. Does this make sense? There are two questions: What powers must kids have to allow for a lot of morphology? And Why does UG exploit these properties to favor Gs that have this property? Norberthttps://www.blogger.com/profile/15701059232144474269noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-42941850680440762632014-11-20T14:05:02.506-08:002014-11-20T14:05:02.506-08:00It does, if one agrees that kids being good at mor...It does, if one agrees that kids being good at morphology can explain the emergence of morphology in the cases I mentioned (or potential emergence, for NSL). The idea is that kids are SO good at morphology that in some cases they create it where it didn't exist before, leading to languages with more morphology.<br /><br />I don't think this idea inconsistent with your story either - this seems to me like a question of function/utility (what helpful things does morphology do once it's there) vs. source (why is it there/how did it get there in the first place). Heidi Getzhttps://www.blogger.com/profile/03645269440087217443noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-36843010357818203622014-11-20T10:52:01.792-08:002014-11-20T10:52:01.792-08:00A note from Anne Christophe (whose name I misspell...A note from Anne Christophe (whose name I misspelled in the post but have now corrected) regarding the Chinese stuff that I mentioned. She corrects matters. So for the record from her note to me:<br /><br />"Anne C tells me that all the stuff done on French discussed in the slides replicates in Mandarin" it looks as if all the research presented replicates in Mandarin (including the baby work etc), which is of course not the case at all; only the 'Frequent frames' thing has been replicated in Mandarin and it works quite well thanks to the noun classifiers (and this is not published unfortunately…)Norberthttps://www.blogger.com/profile/15701059232144474269noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-74977582017777282462014-11-20T10:49:58.827-08:002014-11-20T10:49:58.827-08:00So, if I get you correctly, your story is that mor...So, if I get you correctly, your story is that morphology is a by-product of how gets segment things. I am not sure that this is inconsistent with the story I was trying to push. Why morphology? To license vocab acquisition (rather than, say, parsing, or production). For it to do this, of course, kids better be ok at tracking it (and it appears that they are as Newport and Suppala a.o. have argued). But the fact that kids are good at it does not mean that languages should display lots of it, or does it? Norberthttps://www.blogger.com/profile/15701059232144474269noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-13398423863346262182014-11-20T07:03:39.944-08:002014-11-20T07:03:39.944-08:00Morphology may well facilitate vocabulary learning...Morphology may well facilitate vocabulary learning, but I'm not sure that tells us how we got so much morphology in the first place. The process of acquisition might explain it: kids seem to be pretty good at identifying the component morphemes of a complex form, even when those morphemes are presented simultaneously (as in ASL verbs of motion - Elissa Newport and Ted Supalla have work from the 80's showing that kids will actually pull out morphemes individually and then [incorrectly] assemble them sequentially for a while before mastering the grammatical rules for simultaneous combination). They can identify morphemes and end up with a regular morphological system even when their input is messy and irregular (Lissa again, in studies of deaf children learning from non-native signers- Simon's the famous one). Hints of the same effects come through in the work on Nicaraguan Sign Language by Annie Senghas et al where it looks like kids may be inventing morphemes where none existed in their (non-linguistic, gestural) input.<br />So maybe kids really like segmented/componentially organized systems, or they really 'want' language to be organized that way for whatever reason (UG if you like), and as a result we get lots of morphology.Heidi Getzhttps://www.blogger.com/profile/03645269440087217443noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-76505674322765492672014-11-19T07:24:29.833-08:002014-11-19T07:24:29.833-08:00Thx for the info. One of the charms of writing abo...Thx for the info. One of the charms of writing about something you know nothing about is that it is pure speculation. It's nice to know that this can actually be backed up with real work. Thx for the Valian reference. I should have known about this one. <br /><br />I could not agree more heartily with your last remark. IT is fashionable (I do this too) to set up a dichotomy between structure and induction. But one way of looking at what GG has done is to try and identify those principles and domains that will allow induction to fruitfully proceed. Norberthttps://www.blogger.com/profile/15701059232144474269noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-58551928844802062672014-11-19T07:19:28.811-08:002014-11-19T07:19:28.811-08:00You are right. I might no invidious distinction be...You are right. I might no invidious distinction between ASL and, say, English. I should have said "generally expressed in sound…." Thx. Norberthttps://www.blogger.com/profile/15701059232144474269noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-34169979834207587932014-11-19T04:21:33.333-08:002014-11-19T04:21:33.333-08:00This stuff is fascinating, I agree. I think the ge...This stuff is fascinating, I agree. I think the general research strategy -- explaining some quite obvious gross properties of language through the way they facilitate language acquisition -- is great, and may be quite widely applicable as it can explain them without having to be part of UG, and so doesn't aggravate Darwin's problem. <br />Furthermore, as Charles points out, it is amenable to computational investigation. Alex Clarkhttps://www.blogger.com/profile/04634767958690153584noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-41823269179444216992014-11-18T21:00:38.810-08:002014-11-18T21:00:38.810-08:00A minor quibble. You write, "generally phonol...A minor quibble. You write, "generally phonological, but not exclusively so, think ASL". Sign languages have a level of structure roughly comparable to phonological structure in spoken languages, and the consensus in the sign language linguistics community is to call this level phonology. One could argue whether this is a neutral choice of terminology, given that the word has the Greek root for voice in it. But that battle is over, and statements like this risk giving the impression that ASL (and other sign languages) lack phonological structure. If instead you meant the possibility of iconic signs, most (all?) iconic signs have phonological structure, and iconic vocabulary exists for spoken languages as well.Josh Falkhttps://www.blogger.com/profile/13394996800799458021noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-53882119356045783302014-11-18T20:29:21.568-08:002014-11-18T20:29:21.568-08:00Regarding this portion of Norbert's post:
[.....Regarding this portion of Norbert's post:<br /><br /><i>[...] morphology per se is not what we should be looking at. Rather it is morphology plus closed class items.</i><br /><br />It might be worth noting that within a "syntax all-the-way-down" model (as in, say, Distributed Morphology), the disjunction embodied by the 'plus' in this quote would be an illusion. That's because, in such a model, there's little difference between so-called closed class "words" and closed class (derivational) morphemes.<br /><br />Of course, there are nuances here -- in particular, inflectional morphology is not as neatly unified with free-standing closed class "words" as derivational morphology is.<br /><br />But to me, it's interesting to consider that the child really isn't summing over two classes (closed class "words" and closed class morphemes) at all. It only <i>looks</i> that way to adult linguists (and associated laypersons) who have been falsely indoctrinated into a false belief in the existence of "words" :-)<br /><br />NB: I don't think anybody denies that there is such a thing as a <i>phonological word</i>, of course, not even dyed-in-the-wool DMers. But notice that, crucially, closed class items seldom constitute phonological words in their own right. So the existence of phonological words is only useful for ascertaining what the closed class items are insofar as, say, closed class items tend to occur exclusively at the edges of phonological words. (This is something that Anne mentioned in her talk, though I don't remember off hand if it was in the slides.)<br /><br />Omerhttps://www.blogger.com/profile/06157677977442589563noreply@blogger.comtag:blogger.com,1999:blog-5275657281509261156.post-1235827332143320262014-11-18T19:33:18.583-08:002014-11-18T19:33:18.583-08:00Form is easy, meaning is hard: this should be ensh...Form is easy, meaning is hard: this should be enshrined somewhere.<br /><br />The evidence that closed class words facilitate the acquisition of open class words comes from several places. One type is summarized by Christophe--an idea that Virginia Valian refers to as "anchoring points" (J. Mem. Lg, 1988), high frequency items (e.g., closed class words) can be used as anchors to determine the properties of low frequency items (e.g., open class words). Roger Brown’s classic “sib” experiment started this all. <br /><br />Another strand comes, perhaps surprisingly, from NLP. In a series of insightful papers, Qiuye Zhao, who just finished at Penn, exploited these very ideas to achieve state of the art results for part of speech tagging and Chinese word segmentation in a largely unsupervised setting. The key idea is to use high frequency elements, including words AND morphological endings--Norbert: that’s indeed what morphology is for!—to define the distributional profile for the lower frequency elements. (Email me for a copy of her dissertation or find her papers at venues such as EMNLP, ACL etc.) <br /><br />But it remains unclear how much this very first step of bootstrapping is statistical (or non-linguistic). Page 19 of the slides cites an important study by Shi and Melancon (http://www.tpsycho.uqam.ca/NUN/D_pages_Profs/D_GRL/Publications/shi_melancon_2010.pdf). When presented with "le mige", the latter a non-sense French word, 14 month old infants interpret "mige" as a noun. But when presented with "je mige", they do NOT interpret "mige" as a verb. I have checked the statistical properties of child directed French and more extensively, English, which is similar in this respect. The transitional probability of D-N is actually LOWER than that of Pronoun-V: a statistical learner should be do better at recognizing “mige” the verb than “mige” the noun.<br /><br />We are not quite sure why the French infants behaved the way they did, but a plausible idea is very much a linguistic one: D-N is part of a phase but Pronoun-V straddles phrase boundaries. It is a general principle of language that statistical correlation over structurally ill-formed units, no matter how strong, is simply not attended to. BTW, this is exactly what Carl de Marcken showed in his 1995 ACL paper: statistical correlations frequently mislead, unless they are constrained by structural principles.Charles Yanghttps://www.blogger.com/profile/06041398285400095406noreply@blogger.com