From marthajo at linc.cis.upenn.edu Thu Feb 4 19:29:12 1999 From: marthajo at linc.cis.upenn.edu (Martha McGinnis) Date: Thu, 4 Feb 1999 14:29:12 -0500 Subject: form-class and gender Message-ID: Dear DM-listers, Something has been troubling me recently about the nature of agreement with nouns. Traditionally there are two kinds of more or less arbitrary classes into which a noun can fall: its "form class" and its "gender". It's said that gender is relevant for agreement, but that agreement ignores form class. For example, Spanish nouns have form classes that determine whether they end in -a, -o, etc, but these don't correlate perfectly with gender. So you get "el muchach-o es feo" (the boy is ugly), with masculine agreement, but "la man-o es fea" (the hand is ugly), with feminine agreement. But it's not completely straightforward to say that form class doesn't trigger agreement. For one thing, the form class morphology itself is in a sense "agreeing" with the lexical root, unless the root comes with a separate "form class" feature which is spelled out by the form class morphology. For another thing, I believe that in some cases other morphology on the noun is affected by its form class, e.g. case affixes on nouns in Greek can vary with form class. So potentially we have (at least) two arbitrary properties of the stem, X and Y, which trigger agreement in different domains -- one perhaps very locally, and one less locally, e.g. on adjectival predicates, determiners, etc. Now a question arises -- is this a principled difference, or do we in fact find languages with three or four such properties, all triggering different kinds of agreement, or agreement in different locality domains? Suppose the difference _is_ principled. How should it be captured? Within a syntactic theory where Infl (or Agr, or N-features) plays a syntactic role, we could suppose that potentially nonlocal "gender" agreement is really syntactic agreement, i.e., the pronunciation of agreement features that play a role in the syntax (e.g. attracting NPs, and/or checking/assigning their Case features) -- while "form-class" agreement is purely morphological, referring only to features that play no role in the syntax. However, I don't know of any convincing evidence that agreement plays _any_ role in the syntax (by contrast with abstract Case, for which I think there is some evidence). Can anyone point to evidence of that kind, or does anyone have another suggestion as to how to deal with the split between form-class and gender in DM? Martha McGinnis From marthajo at linc.cis.upenn.edu Tue Feb 9 13:12:57 1999 From: marthajo at linc.cis.upenn.edu (Martha McGinnis) Date: Tue, 9 Feb 1999 08:12:57 -0500 Subject: sound symbolism Message-ID: >From Zylogy at aol.com Tue Feb 9 01:03 EST 1999 From: Zylogy at aol.com Date: Tue, 9 Feb 1999 01:03:10 EST Subject: Distributed Morphology --------------------------------------------- I'm very curious to know if the theory has anything to say about the vast amounts of sound symbolism to be found in ideophone systems and lexical roots in languages from around the world. Personally I'd be quite happy if the theory was open to new views about this. Let me know. Thanks. Sincerely, Jess Tauber zylogy at aol.com From marthajo at linc.cis.upenn.edu Wed Feb 10 01:48:10 1999 From: marthajo at linc.cis.upenn.edu (Martha McGinnis) Date: Tue, 9 Feb 1999 20:48:10 -0500 Subject: sound symbolism Message-ID: To my knowledge, DM has nothing special to say about sound symbolism. However, it does make a distinction that might be useful to people interested in the topic. DM distinguishes between two types of meaning. "Syntactic" (or "semantic") meaning is a property of the elements manipulated by the syntax. For example, the components of meaning that distinguish finite from non-finite clauses are syntactic/semantic. "Encyclopedic" meaning is meaning that plays no role in the syntax. The components of meaning that distinguish "dog" from "cat" (or "sneer" from "snarl") are encyclopedic. In DM, the "pieces" of morphophonology -- Vocabulary items -- are phonological strings (e.g. /-ed/) associated with a morphosyntactic category (e.g. Tense) and syntactic/semantic features (e.g., [+past]). Items of a given category compete for insertion into a morphosyntactic node of the same category, which is a terminal node of a syntactic structure (tree). The winner of the competition is the vocabulary item that best matches the features of the morphosyntactic node. My understanding is that sound symbolism makes reference to phonological strings that aren't Vocabulary items, because they aren't inserted into their own separate morphosyntactic nodes. For example, the sequence sn- (sneeze, snore, snort, sneer, snarl) doesn't correspond to some morphosyntactic node such as Tense (T) or Voice (v). On the contrary, my native-speaker intuition is that sn- is rather loosely associated with some component of meaning like "nose- related." This kind of meaning doesn't play a role in syntax -- it's encyclopedic. It would be interesting to me if it turned out that sound symbolism _always_ iconifies encyclopedic meaning, rather than syntactic meaning. > I'm very curious to know if the theory has anything to say about the vast > amounts of sound symbolism to be found in ideophone systems and lexical roots > in languages from around the world. Personally I'd be quite happy if the > theory was open to new views about this. Let me know. Thanks. > > Sincerely, > Jess Tauber > zylogy at aol.com > From marantz at MIT.EDU Fri Feb 19 22:59:33 1999 From: marantz at MIT.EDU (Alec Marantz) Date: Fri, 19 Feb 1999 17:59:33 -0500 Subject: From Morris Halle, in response to Carstairs-McCarthy Message-ID: Responding to Andrew Carstairs-McCarthy's comments on zero affixes: A crucial aspect of the theory of DM has been the recognition that phonetic zero must be included among the phonetic exponents of a morpheme. To see the necessity for this consider the appearance of the helping verb in yes-no questions such as those in (1). (1) John snore-[d, Past] di-[d, Past] John snore? John lef-[t, Past] di-[d, Past] John leave? John sang-[zero, Past] di-[d, Past] John sing? It is generally recognized that the main verb is blocked from adjoining the Tense morpheme in yes-no questions. Since the Tense morpheme is an affix and must not remain stranded, the helping verb "do" is inserted. Like many other grammatical formatives the Past tense has different phonetic exponents depending on the verb to which the tense affix is adjoined. Specifically, as illustrated in (1) English has three Past tense exponents: /d/ as in snore-d, sol-d, di-d; /t/ as in bough-t, los-t, knel-t, and zero as in sang-0, burst-0, beat-0, and the choice among them is determined -- as is perfectly normal -- by the verb to which the affix attaches. Equally typical is the fact that several distinct morphemes have the same phonetic exponent. For example, in English /s/ is the exponent of 3SgPres of verbs, and of the Plural and Possessive of nouns. The same is true of zero. In addition to being the exponent of the Past, zero is also the exponent of the Perfect Participle of some verbs, and of the Present tense (for Persons other than 3Sg) of all verbs. In the latter case the reality of the zero suffix is vouchsafed by the facts in (2), which precisely parallel (1). (2) they snore-[zero, 3PlPres] do-[zero, 3PlPres] they snore? The preceding facts settle the question of whether English has zero affixes for they show that -- as C-McC demands -- "that zero affixes behave just like overt affixes." C-McC's further requirement that they do so also "in respect to readjustment" is not met, for there is no more reason for this requirement than there is for the requirement that affixes triggering readjustment should share any other phonetic property. Note, in particular, where zero is the exponent of the Present tense as in (2) it triggers no readjustment. As Past exponent, zero sometimes causes phonetic modifications in the stem to which it attaches, sometimes not. In this again zeros parallel other Past tense exponents. A few examples are cited in (3). (3) Zero-Past with modifications: sang-0, struck-0, dug-o /t/-Past with modifications: knel-t, lef-t, bough-t /d/-Past with modifications: sol-d, tol-d di-d, ha-d Zero-Past without modifications: beat-0, shed-0, cast-0, burst-0, hurt-0, rid-0, spread-0 /t/-Past without modifications: burn-t, learn-t, dwel-t, knel-t /d/-Past without modifications: seat-ed, bread-ed, putt-ed, blast-ed, thirst-ed Since /d/ is the default suffix, it is perhaps to be expected that it would trigger stem modification only exceptionally. C-McC is mistaken in believing that there is a difference between /t/ and zero as to the extent to which they cause stem modification. The fact that there are fewer varieties of modification with /t/ than with zero is a result of the different number of stems that take the respective suffixes. There are about 40 stems that take /t/, and about 130 that take zero. (Details in Bloch, 1947, Language 23, 399-418.) Nor should it be overlooked that the zero exponent of the Present tense illustrated in (2) induces no stem modifications whatever. These facts support the DM proposition that the stem modifications are independent of the phonetic character of the suffix. (Incidentally, the zero plural suffix as in fish, quail, deer, moose, etc. appears without stem modification in the preceding, but triggers modifications in geese, mice, lice, men, women.) marantz at mit.edu From marthajo at linc.cis.upenn.edu Tue Feb 23 21:40:14 1999 From: marthajo at linc.cis.upenn.edu (Martha McGinnis) Date: Tue, 23 Feb 1999 16:40:14 -0500 Subject: Andrew Carstairs-McCarthy: zero affixes again Message-ID: Date: Tue, 23 Feb 1999 13:06:22 +1300 From: Andrew Carstairs-McCarthy Subject: zero affixes again Here is a rejoinder to Morris Halle's reply to my posting 'Zero affixes in DM'. In that posting I expressed unease about the positing of a zero affix in the DM analysis of past tense forms such as _cut, sang_ and past participle forms such as _cut, sung, come_. Halle does not succeed in allaying my unease, for three reasons: 1. His reply relies on confusing two senses of the term 'morpheme'. Both of these senses are sanctioned in DM, but to my mind this is unfortunate, because it risks giving rise to just the kind of confusion that undermines Halle's argument (I suggest). 2. He fails to address the peculiar phonological restrictions on those past and past participle verb forms that allegedly carry the zero affix. 3. He offers no evidence that the advantages of rejecting the zero-affix analysis for these verb forms are spurious or illusory. 1. The two senses of 'morpheme' In DM a distinction is drawn, wisely, between morphosyntactic feature bundles and the Vocabulary items that may express them. A Vocabulary item has in principle a phonological shape (even though for some items this may be zero); a morphosyntactic feature, or bundle of features, does not. Also, one bundle of features may be expressed by different Vocabulary items in different contexts. So which, if either, of these kinds of thing is a morpheme, according to DM? The answer is that *both* kinds of thing count as morphemes. As Halle and Marantz put it (1993:114): 'we have chosen to call the terminal elements "morphemes" both before and after Vocabulary insertion, that is, before and after they are supplied with phonological features'. Thus, apparently, we are entitled to say that _waited_ and _given_ display different past participle morphemes because their affixes are different, but also that they display the same morpheme because both these affixes realize [past participle]. Halle and Marantz claim that nothing hinges on this terminology. However, it undermines what Halle sees as the clinching argument for zero affixes in English verbs, namely the argument from do-support. The argument runs like this. In yes-no question contexts, lexical verbs cannot carry Tense-Number marking; this marking must instead be realized on the dummy verb _do_, as in: (1) Doe-s John snore? (2) Di-d John snore? But this _do_ shows up even when the Tense-Number features in question are [non-3rd-singular present]: (3) Do they snore? So, even though [non-3rd-singular present] is never phonologically overt in any context, it must be realized by a zero affix in order to trigger do-support. In Morphological Structure, therefore, (3) is a better represented as (4): (4) Do-0 they snore? This argument fails because it relies on a blurring of the distinction between morpheme-as-feature-bundle and morpheme-as-affix. One may agree that a feature bundle needs a peg on which to get realized, without agreeing that the realization must inevitably be carried out by a morpheme-as-affix. To substantiate the claim that a zero morpheme-as-affix is present in (3) (or (4)), Halle needs to do more than just invoke the phenomenon of do-support. The fact that he does not do more is at least partly due, I suggest, to DM's use of the term 'morpheme' to designate both syntactic terminal elements (feature bundles) and also affixes. Let us grant that the morpheme-as-feature-bundle [non-3rd-singular present] occupies a terminal position in a syntactic tree. It does not follow that there must be an independent Vocabulary item (a morpheme-as-affix) to realize it -- unless of course the theory of DM decrees that such a Vocabulary item *must* exist, thanks to the two-sided character ascribed in the theory to 'morphemes'. But in that case the postulation of a zero affix for the purpose seems no more than a trick to reconcile the theory with recalcitrant data. It is important to stress the difference between the zero morphological realization of [non-3rd-singular present] and the alleged zero affix for [past]. The feature [past] is uncontroversially realized by an affix in most contexts: _snored, sold, felt_ etc. The postulation of a zero affix in _sang, cut_ etc. therefore seems less contrived than in examples (3) and (4). Even so, the facts of do-support in the past tense count *against* the existence of a [past] zero affix, not for it. Remember that what is at issue is not whether features such as [past] need a peg on which to get realized; rather, the issue is whether this realization ever takes the form of a zero affix. Imagine an imaginary English in which the yes-no questions corresponding to (5) and (6) were not (7) and (8) but rather (9) and (10): (5) John left. (6) John sang. (7) Did John leave? (8) Did John sing? (9) Di-t John leave? (10) Di John sing? In this version of English, it is clearly not just the feature [past] that gets supported by _do_, but the actual affix that realizes [past] on the accompanying lexical verb in positive declarative contexts. In this version of English, the past tense of _do_ will usually carry an affix (_-d_ or _-t_), so one could reasonably argue that in (6) and (10) too there is an affix, albeit a phonologically null one. The forms _sang_ and _di_ would therefore be more accurately represented in Morphological Structure as _sang-0_ and _di-0_. But actual English is not like this. The fact that the dummy _do_ shows up as _did_ in both (7) and (8), ignoring the morphology of the lexical verbs in (5) and (6), shows conclusively that do-support operates at the level of morphosyntactic features, not at the level of the affixes that realize them. This further weakens the case for invoking a zero affix in a [non-3rd-singular present] context such as (3). 2. The phonological restrictions on the past (participle) 'zero affixes' Halle points out correctly that, in the DM analysis, both overt affixes and zero affixes may or may not trigger readjustments. As he puts it: "As Past exponent, zero sometimes causes phonetic modifications in the stem to which it attaches, sometimes not. In this again zeros parallel other Past tense exponents". He also provides a list of verbs in which an (alleged) zero Past exponent causes no phonetic modification, namely _beat, shed, cast, burst, hurt, rid, spread_. But he does not comment on the striking fact that all these verbs end in in coronal stops. If one rejects the DM analysis for these forms, plausible avenues for explanation open up: perhaps these forms have an underlying overt coronal suffix, forming a cluster that gets simplified, or perhaps they have no suffix at all (even a zero one) because they already conform to a past-tense template. Under the DM analysis, however, no such explanation seems available. Why should the zero affix without readjustment favor precisely coronal stems, rather than (say) bilabial ones, or disyllabic ones, or ones with high vowels? Halle does not say. Likewise, Halle does not say why, for Past Participle, the zero affix without readjustment is limited precisely to these coronal stems plus also _run_ and _come_ -- a pair of verbs whose vowel complies with the pattern of _sung, struck, dug_ etc. The fact that all this seems to be relegated to mere coincidence in Halle's DM analysis must surely count against it. 3. Positive evidence for the nonaffixal character of stem alternations Does stem alternation play by the same rules as affixal allomorphy, or by different rules, or by no rules at all (as might be expected if it is merely the random residue of phonological change)? Under DM, much stem alternation is handled by readjustment, sometimes triggered by zero affixes. It will therefore count in favor of DM if stem alternation plays by substantially the same rules as affixal allomorphy does: that is, if it is found that the zeroes that one posits to trigger stem modifications obey any constraints or restrictions that affect overt affixes, and that there are no special constraints on the stem modifications themselves. Evidence on these lines, if sufficiently substantial, could even outweigh the doubts I have expressed about zero affixes in English verbforms. In fact, however, such evidence as I am aware of points firmly the other way, i.e. towards the conclusion that stem alternation plays by different rules than affixal allomorphy. I have in mind the following: (a) stem alternation is irrelevant for the purpose of 'paradigm economy' (Carstairs 1987:221-32; 1988) and 'blur avoidance' (Carstairs-McCarthy 1994); (b) patterns of stem alternation can be productive, independently of the phonological shape of the alternants (Maiden 1992; 1996), in line with the view of stems as part of 'morphology by itself' (Aronoff 1994); (c) implicative paradigm structure conditions of the kind proposed in Natural Morphology by Wurzel (1984) and applied to German verbs by Bittner (1985) are too powerful a mechanism to account for affixal 'blur avoidance' (Carstairs-McCarthy 1994:754-7), but seem well suited to describing observed patterns of stem distribution (Carstairs-McCarthy 1991:237-47). I would welcome debate on whether these observations can be reconciled with DM. For the time being, they seem to count against too ready reliance on affixation plus readjustment in order to handle stem allomorphy. Aronoff, M. 1994. Morphology by itself. Cambridge, MA: MIT Press. Bittner, A. 1985. Implikative Hierarchien in der Morphologie: Das 'Stark-schwach-Kontinuum' der neuhochdeutschen Verben. Acta Linguistica Academiae Scientiarum Hungaricae 35:31-42. Carstairs, A. 1987. Allomorphy in inflexion. London: Croom Helm. Carstairs, A. 1988. Nonconcatenative inflection and paradigm economy. Theoretical morphology: approaches in modern linguistics, ed. by M. Hammond and M. Noonan, 71-7. San Diego: Academic Press. Carstairs-McCarthy, A. 1991. Inflection classes: two questions with one answer. Paradigms: the economy of inflection, ed. by F. Plank, 213-53. Berlin: Mouton de Gruyter. Carstairs-McCarthy, A. 1994. Inflection classes, gender and the Principle of Contrast. Language 70:737-88. Maiden, M. 1992. Irregularity as a determinant of morphological change. Journal of Linguistics 28:285-312. Maiden, M. 1996. The Romance gerund and 'system-dependent naturalness' in morphology. Transactions of the Philological Society 94:167-201. Wurzel, W.U. 1984. Flexionsmorphologie und Natuerlichkeit. Berlin: Akademie-Verlag. (English translation (1989): Inflectional morphology and naturalness. Dordrecht: Kluwer.) Andrew Carstairs-McCarthy Associate Professor 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 marthajo at linc.cis.upenn.edu Wed Feb 24 23:29:14 1999 From: marthajo at linc.cis.upenn.edu (Martha McGinnis) Date: Wed, 24 Feb 1999 18:29:14 -0500 Subject: Martha McGinnis: zero affixes Message-ID: I've been following the discussion on zero affixes, and wanted to ask a couple of clarification questions. First, if I understood correctly, Morris Halle (MH) observed that Tense morphology can appear either on the main verb or on an auxiliary, depending on the syntactic environment. He concluded that the element realized by Tense morphology is a syntactic node. Andrew Carstairs-McCarthy (AC) granted this, but contended that there need not be an independent Vocabulary item (a morpheme-as-affix) to realize the Tense node. How are we to represent the fact that the Tense node is pronounced in some cases, and isn't in others? MH says that non-pronunciation has the same status as a given particular pronunciation (i.e., it's a zero Vocabulary item), while AC suggests that non-pronunciation has a different status from particular pronunciations. I'm curious as to how strict AC wants to be about ruling out zero Vocabulary items. As MH pointed out, null realization of syntactic nodes can occur in the absence of stem changes, e.g. pro-arguments in Japanese, say _0 wakar-u_ (understand-past) "I/you/he/she/they understood". I'm not sure whether AC would want to rule out zero Vocabulary items like this too. In any case, AC's point of departure seems to be the No Blur Principle (NBP), so maybe now is a good time to bring up something that puzzles me about it. The robustness of the NBP is interesting in its own right, but I can't figure out how the child would apply it as a learning principle, even without assuming zero affixes. For example, suppose that a child learns that an affix A is the default realization of a class marker on stems of type X. This means that A can appear on several classes of X (e.g. class I, II, III). According to the NBP, any other realization of that class marker, say B, will uniquely identify the class it realizes (e.g., just class IV); this is supposed to be useful (necessary?) for acquiring the system of class markers. I II III IV V VI A A A B -- -- However, in addition to learning class markers A, B, etc., the child must also learn which classes lack (overt) class markers (V, VI). Otherwise, s/he will use the default affix A for these classes. Even supposing that there are no zero affixes, why is it not a problem for the learner that the unaffixed forms don't all belong to one class? Why is this less of a problem than the problem that would arise if forms with affix B didn't all belong to one class? Given whatever stategy the child uses to acquire the relevant information about classes V and VI, why couldn't the same strategy be used to acquire the relevant information about classes V and VI in the following system, ruled out by the NBP? I II III IV V VI A A A -- B B Thanks for any help with these questions. -Martha From marthajo at linc.cis.upenn.edu Thu Feb 25 16:13:43 1999 From: marthajo at linc.cis.upenn.edu (Martha McGinnis) Date: Thu, 25 Feb 1999 11:13:43 -0500 Subject: Andrew Carstairs-McCarthy: zero affixes Message-ID: Here is a response to Martha McGinnis's queries. 1. Non-pronunciation of nodes Alluding to my rejection of a zero morpheme-as-affix realizing [present, non-3rd-singular], MMcG asks: >How are >we to represent the fact that the Tense node is pronounced in some cases, >and isn't in others? MH says that non-pronunciation has the same status as >a given particular pronunciation (i.e., it's a zero Vocabulary item), while >AC suggests that non-pronunciation has a different status from particular >pronunciations.... >As MH pointed out, null realization of syntactic nodes >can occur in the absence of stem changes, e.g. pro-arguments in Japanese, >say _0 wakar-u_ (understand-past) "I/you/he/she/they understood". This seems to me to be a question for the syntactician rather than the morphologist. How are we to represent the fact that traces are not pronounced, or big-PRO, or little-pro (in Principles-and-Parameters terms)? That is not a question that morphological theory is generally expected to answer -- and rightly so. In a DM-style analysis whereby all morphosyntactic features are in principle located in bundles at terminal nodes in syntactic trees (though possibly subject to merger, fusion and fission in Morphological Structure), the variety and incidence of terminal nodes where nothing phonologically overt gets inserted is quite high, relative to some other styles of analysis. It doesn't follow that some of these nodes *must* have an affix asserted at them, despite appearances (i.e. a phonologically empty affix) -- that is, it doesn't follow *unless one makes it follow*, through one's definition of 'morpheme'. But then the presence or absence of an affix ceases to be an empirical issue. As to whether _Wakaru_ 'I understood' involves a feature bundle [1 singular] that is realized by a zero affix, I take no position. I simply question the usefulness of a style of analysis under which such a feature bundle *must* be deemed to be realized by an affix or pronoun or whatever, even if there's nothing there phonologically. >I'm curious as to how strict AC wants to be about ruling out zero >Vocabulary items. I don't want to rule them out in principle. For example, I suspect a good case can be made for a zero affix on a form like Russian _knig_ 'of the books' (genitive plural). This cell in the paradigm of the lexeme KNIGA is highly marked in morphosyntactic terms (by contrast with e.g. nominative singular _knig-a_); all the other cells have an overt suffix; and this cell too had one in the not-too-distant past. All I am arguing is that the case for a zero affix on _cut, sang_ etc. is weak, being based mainly on a terminological fiat. 2. Learning, bare stems, and the No Blur Principle (NBP). MMcG questions the role of the NBP in inflection-class learning, and asks: >Suppose that a child learns >that an affix A is the default realization of a class marker [rather, of a >morphosyntactic cell] on stems of >type X. This means that A can appear on several classes of X (e.g. class I, >II, III). According to the NBP, any other realization of that [cell], say >B, will uniquely identify the class it realizes [rather, the class to >which lexemes carrying it belong], (e.g., just >class IV); this is supposed to be useful (necessary?) for acquiring the >system of class markers [rather, inflectional affixes]. > > I II III IV V VI > A A A B -- -- > [Table 1] >However, in addition to learning class markers A, B, etc., the child must >also learn which classes lack (overt) class markers (V, VI). Otherwise, >s/he will use the default affix A for these classes. Even supposing that >there are no zero affixes, why is it not a problem for the learner that the >unaffixed forms don't all belong to one class? Why is this less of a >problem than the problem that would arise if forms with affix B didn't all >belong to one class? This is a nice question, and I don't think the evidence is yet available to give a complete answer. A research project for someone, then! One needs to distinguish between two tasks: (a) learning what a given affix (A or B in Table 1) means, and (b) learning which class a given lexeme belongs to. >>From the point of view of learning task (a), classes V and VI in Table 1 present no problem, because in the cell in question they carry no affix. (I'm assuming there is no evidence for a 'real' zero affix here, as in _knig_.) In this respect Table 1 contrasts directly with Table 2: I II III IV V VI A A A -- B B [Table 2] That's why Table 2 can't be learned by using the same strategy as for Table 1: Table 2 contains two blurred affixes, A and B, so fulfilling learning task (a) in respect of Table 2 is predicted by the NBP to be impossible or at least difficult. But what about task (b) in respect of Table 1? In particular, which lexemes go in class V and which go in class VI? Let's suppose that, for a different cell (cell 2), class V lexemes carry a class-identifier, E: I II III IV V VI Cell 1 A A A B -- -- Cell 2 C D C C E C [Table 3] Once the child has spotted that C is the class-default affix for cell 2, she 'knows' that affix E must be a class-identifier, i.e. that all lexemes with E are inflected alike (at least affixally), and differently (in at least cell 2) from any lexeme with C. So she 'knows' that classes V and VI are distinct. It is conceivable, though, that V and VI differ in that V has in cell 2 not a class-identifying affix but no affix at all: I II III IV V VI Cell 1 A A A B -- -- Cell 2 C D C C -- C [Table 4] This poses more of a problem for learning task (b). Let's suppose that neither V nor VI has any class-identifier in any other cell (cell 3, 4, etc.). Then there seems to be nothing for the child to cotton on to for the purpose of 'fixing' the membership of V and VI. So it is tempting to hypothesize that such a pattern could never arise, i.e. that every class must have a class-identifying affix in at least one cell. This represents a tightening of the NBP as presented in my _Language_ article, so it will be nice if it is true. Is there evidence to support it? At first sight, German nouns seem to count against. In Table 4 on page 745 of _Language_ 70 (1994), classes II (Ziegel) and IX (Dorn) are represented as both lacking a class-identifier, and differing only in that, in the Nondative Plural, II has a blank while IX has the class-default suffix _-(e)n_. But, as I note on that page, the Plural blank of class II should probably regarded as a phonologically conditioned variant of the _-e_ suffix of class I/III (Tag). If so, this counterexample is taken care of. There is still a problem with English verbs, however, as represented in Table 5 on page 746. If we ignore the minuscule and (for me) obsolete class V (_crow, dive_), _-en_ becomes the class-default Past Participle suffix. But then class III (_give_) and class IV (_show_) exemplify the pattern that concerns us, i.e. no class-identifiers, and differentiation only by a blank versus a class-default affix in one cell (Past _gave--_, PastPrt _give-n_ versus _show-ed, show-n_). If the strengthened version of the NBP is too strong, then, the problem of fulfilling task (b) in respect of patterns like Table 4 above remains. But notice one fact about the English examples: the suffixless Past form _gave_ is by no means unmarked for [past], because its ablauted vowel is peculiar to it. So perhaps it is stem alternation of this kind that renders inflection class systems learnable in circumstances where some class (or, perhaps, more than one class) has no class-identifying affix. The fact that stem alternation as such is not relevant to blur avoidance need not entail that it is irrelevant to determining inflection class membership. (Recall my comment in my previous positing that, at least in German verbs, stem alternation seems to lend itself well to description in terms of Wurzel-style implicative paradigm structure conditions, which involve lexical specification.) This possibility raises the whole fascinating but complex issue of the relationship between affixal and nonaffixal inflection. All the more important, then, that the distinction between the two kinds of inflection should not muddied through a proliferation of zero affixes! Andrew Carstairs-McCarthy Associate Professor 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 marthajo at linc.cis.upenn.edu Thu Feb 4 19:29:12 1999 From: marthajo at linc.cis.upenn.edu (Martha McGinnis) Date: Thu, 4 Feb 1999 14:29:12 -0500 Subject: form-class and gender Message-ID: Dear DM-listers, Something has been troubling me recently about the nature of agreement with nouns. Traditionally there are two kinds of more or less arbitrary classes into which a noun can fall: its "form class" and its "gender". It's said that gender is relevant for agreement, but that agreement ignores form class. For example, Spanish nouns have form classes that determine whether they end in -a, -o, etc, but these don't correlate perfectly with gender. So you get "el muchach-o es feo" (the boy is ugly), with masculine agreement, but "la man-o es fea" (the hand is ugly), with feminine agreement. But it's not completely straightforward to say that form class doesn't trigger agreement. For one thing, the form class morphology itself is in a sense "agreeing" with the lexical root, unless the root comes with a separate "form class" feature which is spelled out by the form class morphology. For another thing, I believe that in some cases other morphology on the noun is affected by its form class, e.g. case affixes on nouns in Greek can vary with form class. So potentially we have (at least) two arbitrary properties of the stem, X and Y, which trigger agreement in different domains -- one perhaps very locally, and one less locally, e.g. on adjectival predicates, determiners, etc. Now a question arises -- is this a principled difference, or do we in fact find languages with three or four such properties, all triggering different kinds of agreement, or agreement in different locality domains? Suppose the difference _is_ principled. How should it be captured? Within a syntactic theory where Infl (or Agr, or N-features) plays a syntactic role, we could suppose that potentially nonlocal "gender" agreement is really syntactic agreement, i.e., the pronunciation of agreement features that play a role in the syntax (e.g. attracting NPs, and/or checking/assigning their Case features) -- while "form-class" agreement is purely morphological, referring only to features that play no role in the syntax. However, I don't know of any convincing evidence that agreement plays _any_ role in the syntax (by contrast with abstract Case, for which I think there is some evidence). Can anyone point to evidence of that kind, or does anyone have another suggestion as to how to deal with the split between form-class and gender in DM? Martha McGinnis From marthajo at linc.cis.upenn.edu Tue Feb 9 13:12:57 1999 From: marthajo at linc.cis.upenn.edu (Martha McGinnis) Date: Tue, 9 Feb 1999 08:12:57 -0500 Subject: sound symbolism Message-ID: >From Zylogy at aol.com Tue Feb 9 01:03 EST 1999 From: Zylogy at aol.com Date: Tue, 9 Feb 1999 01:03:10 EST Subject: Distributed Morphology --------------------------------------------- I'm very curious to know if the theory has anything to say about the vast amounts of sound symbolism to be found in ideophone systems and lexical roots in languages from around the world. Personally I'd be quite happy if the theory was open to new views about this. Let me know. Thanks. Sincerely, Jess Tauber zylogy at aol.com From marthajo at linc.cis.upenn.edu Wed Feb 10 01:48:10 1999 From: marthajo at linc.cis.upenn.edu (Martha McGinnis) Date: Tue, 9 Feb 1999 20:48:10 -0500 Subject: sound symbolism Message-ID: To my knowledge, DM has nothing special to say about sound symbolism. However, it does make a distinction that might be useful to people interested in the topic. DM distinguishes between two types of meaning. "Syntactic" (or "semantic") meaning is a property of the elements manipulated by the syntax. For example, the components of meaning that distinguish finite from non-finite clauses are syntactic/semantic. "Encyclopedic" meaning is meaning that plays no role in the syntax. The components of meaning that distinguish "dog" from "cat" (or "sneer" from "snarl") are encyclopedic. In DM, the "pieces" of morphophonology -- Vocabulary items -- are phonological strings (e.g. /-ed/) associated with a morphosyntactic category (e.g. Tense) and syntactic/semantic features (e.g., [+past]). Items of a given category compete for insertion into a morphosyntactic node of the same category, which is a terminal node of a syntactic structure (tree). The winner of the competition is the vocabulary item that best matches the features of the morphosyntactic node. My understanding is that sound symbolism makes reference to phonological strings that aren't Vocabulary items, because they aren't inserted into their own separate morphosyntactic nodes. For example, the sequence sn- (sneeze, snore, snort, sneer, snarl) doesn't correspond to some morphosyntactic node such as Tense (T) or Voice (v). On the contrary, my native-speaker intuition is that sn- is rather loosely associated with some component of meaning like "nose- related." This kind of meaning doesn't play a role in syntax -- it's encyclopedic. It would be interesting to me if it turned out that sound symbolism _always_ iconifies encyclopedic meaning, rather than syntactic meaning. > I'm very curious to know if the theory has anything to say about the vast > amounts of sound symbolism to be found in ideophone systems and lexical roots > in languages from around the world. Personally I'd be quite happy if the > theory was open to new views about this. Let me know. Thanks. > > Sincerely, > Jess Tauber > zylogy at aol.com > From bogus@does.not.exist.com Thu Feb 11 12:21:09 1999 From: bogus@does.not.exist.com () Date: Thu, 11 Feb 1999 12:21:09 -0000 Subject: No subject Message-ID: ========================================================================= Date: Thu, 11 Feb 1999 12:21:09 -0500 Reply-To: The Distributed Morphology List To: The Distributed Morphology List From: Martha McGinnis Subject: zero affixes in DM Comments: To: dm-list at listserv.linguistlist.org Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" I would be interested in discussion about the role of zero affixes in DM. In _Language_ 70 (1994), 760-1, I gave reasons for unease about DM's readiness to recognize certain zero affixes in English verbforms; but no one active in DM has since tried to assuage my unease. Would some one on this list like to have a go? Or do people think my unease is justified? Because of the relatively close relationship in DM between syntactic terminal nodes and morphological 'pieces' (stems and affixes), DM needs to be generous in recognizing phonologically null, or zero, affixes. For example, according to DM, the past tense forms _waited_ and _sang_ in English are both got by inserting appropriate Vocabulary items into the structure [V [+past]]. To get _waited_, that's straightforward: we insert /wait/ and /ed/. But to get _sang_, it seems we have to insert /sing/ and zero, and then posit a readjustment rule converting /sing + zero/ into _sang_. As I explained in _Language_, such zeros create problems for my No Blur Principle, which I proposed there as a constraint on inflection class organization. How big a worry that is depends on what you think about the No Blur Principle. My own view is that it is empirically rich, is well supported by the evidence, and is naturally interpretable as a special case of the Principle of Contrast that governs vocabulary acquisition. What's more, a version of No Blur has (I think) further empirically rich implications for inflectional allomorphy outside the domain of inflection class organization -- implications which are well supported by evidence from Latin, Hungarian and Turkish (see my 'How lexical semantics constrains inflectional allomorphy', in _Yearbook of Morphology 1997, 1-24). Quite apart from that, the analysis of e.g. _sang_ as /sing + zero/ (with readjustment) leads to difficulties. In order to support their view that the supposed English zero affixes really are affixes, just like uncontroversial overt affixes, DM protagonists need to show that these zero affixes behave just like overt affixes in respect of readjustment. In other words, they need to show that: (a) substantially the same readjustments that operate where zero affixes are posited also operate where there are overt affixes, and: (b) zero affixes can appear without any readjustment just as freely as overt affixes do. But there is actually strong evidence against both (a) and (b). In respect of (a), it is easy to check that the vowel alternations that occur in verbs with overtly suffixed pasts like _sold_, _felt_ and _brought_ have little overlap with those that occur in verbs with 'zero-suffixed' pasts such as _sang_, _dug_ and _fell_, as Halle and Marantz themselves admit (_View from Building 20_ page 129: 'readjustment rules triggered by the /-n/ past participle and the zero past suffixes are considerably more complex than those triggered by /-d/ or /-t/'). In respect of (b), it turns out that the verbs for which Halle and Marantz posit a zero past suffix or a zero past participle suffix without any readjustment fall into narrow classes -- classes whose narrowness is unexplained in the DM framework. Verbs with a zero past suffix and no readjustment, such as _beat, cut, hit, shed_ and _spread_ all end in coronal stops. Verbs with a zero past participle suffix and no readjustment include all those just mentioned, plus _run_ and _come_. With an overt-suffix or nonsuffixal analysis, both biases seem explicable: perhaps one of the coronal suffixes /-t/ or /-d/ gets added to _beat_ etc. and undergoes cluster simplification, and perhaps _run_ and _come_ already look like 'good' past participle forms on the strength of their vocalic resemblance to _sung, struck, dug_ etc. But with the zero-suffix analysis of Halle and Marantz, these biases have to be regarded as purely accidental. It may be that an alternative to the zero-suffix analysis can be found within DM. If not, then the shortcomings of this analysis are an embarrassment for DM, it seems to me. Do people agree? Andrew Andrew Carstairs-McCarthy Associate Professor 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 marantz at MIT.EDU Fri Feb 19 22:59:33 1999 From: marantz at MIT.EDU (Alec Marantz) Date: Fri, 19 Feb 1999 17:59:33 -0500 Subject: From Morris Halle, in response to Carstairs-McCarthy Message-ID: Responding to Andrew Carstairs-McCarthy's comments on zero affixes: A crucial aspect of the theory of DM has been the recognition that phonetic zero must be included among the phonetic exponents of a morpheme. To see the necessity for this consider the appearance of the helping verb in yes-no questions such as those in (1). (1) John snore-[d, Past] di-[d, Past] John snore? John lef-[t, Past] di-[d, Past] John leave? John sang-[zero, Past] di-[d, Past] John sing? It is generally recognized that the main verb is blocked from adjoining the Tense morpheme in yes-no questions. Since the Tense morpheme is an affix and must not remain stranded, the helping verb "do" is inserted. Like many other grammatical formatives the Past tense has different phonetic exponents depending on the verb to which the tense affix is adjoined. Specifically, as illustrated in (1) English has three Past tense exponents: /d/ as in snore-d, sol-d, di-d; /t/ as in bough-t, los-t, knel-t, and zero as in sang-0, burst-0, beat-0, and the choice among them is determined -- as is perfectly normal -- by the verb to which the affix attaches. Equally typical is the fact that several distinct morphemes have the same phonetic exponent. For example, in English /s/ is the exponent of 3SgPres of verbs, and of the Plural and Possessive of nouns. The same is true of zero. In addition to being the exponent of the Past, zero is also the exponent of the Perfect Participle of some verbs, and of the Present tense (for Persons other than 3Sg) of all verbs. In the latter case the reality of the zero suffix is vouchsafed by the facts in (2), which precisely parallel (1). (2) they snore-[zero, 3PlPres] do-[zero, 3PlPres] they snore? The preceding facts settle the question of whether English has zero affixes for they show that -- as C-McC demands -- "that zero affixes behave just like overt affixes." C-McC's further requirement that they do so also "in respect to readjustment" is not met, for there is no more reason for this requirement than there is for the requirement that affixes triggering readjustment should share any other phonetic property. Note, in particular, where zero is the exponent of the Present tense as in (2) it triggers no readjustment. As Past exponent, zero sometimes causes phonetic modifications in the stem to which it attaches, sometimes not. In this again zeros parallel other Past tense exponents. A few examples are cited in (3). (3) Zero-Past with modifications: sang-0, struck-0, dug-o /t/-Past with modifications: knel-t, lef-t, bough-t /d/-Past with modifications: sol-d, tol-d di-d, ha-d Zero-Past without modifications: beat-0, shed-0, cast-0, burst-0, hurt-0, rid-0, spread-0 /t/-Past without modifications: burn-t, learn-t, dwel-t, knel-t /d/-Past without modifications: seat-ed, bread-ed, putt-ed, blast-ed, thirst-ed Since /d/ is the default suffix, it is perhaps to be expected that it would trigger stem modification only exceptionally. C-McC is mistaken in believing that there is a difference between /t/ and zero as to the extent to which they cause stem modification. The fact that there are fewer varieties of modification with /t/ than with zero is a result of the different number of stems that take the respective suffixes. There are about 40 stems that take /t/, and about 130 that take zero. (Details in Bloch, 1947, Language 23, 399-418.) Nor should it be overlooked that the zero exponent of the Present tense illustrated in (2) induces no stem modifications whatever. These facts support the DM proposition that the stem modifications are independent of the phonetic character of the suffix. (Incidentally, the zero plural suffix as in fish, quail, deer, moose, etc. appears without stem modification in the preceding, but triggers modifications in geese, mice, lice, men, women.) marantz at mit.edu From marthajo at linc.cis.upenn.edu Tue Feb 23 21:40:14 1999 From: marthajo at linc.cis.upenn.edu (Martha McGinnis) Date: Tue, 23 Feb 1999 16:40:14 -0500 Subject: Andrew Carstairs-McCarthy: zero affixes again Message-ID: Date: Tue, 23 Feb 1999 13:06:22 +1300 From: Andrew Carstairs-McCarthy Subject: zero affixes again Here is a rejoinder to Morris Halle's reply to my posting 'Zero affixes in DM'. In that posting I expressed unease about the positing of a zero affix in the DM analysis of past tense forms such as _cut, sang_ and past participle forms such as _cut, sung, come_. Halle does not succeed in allaying my unease, for three reasons: 1. His reply relies on confusing two senses of the term 'morpheme'. Both of these senses are sanctioned in DM, but to my mind this is unfortunate, because it risks giving rise to just the kind of confusion that undermines Halle's argument (I suggest). 2. He fails to address the peculiar phonological restrictions on those past and past participle verb forms that allegedly carry the zero affix. 3. He offers no evidence that the advantages of rejecting the zero-affix analysis for these verb forms are spurious or illusory. 1. The two senses of 'morpheme' In DM a distinction is drawn, wisely, between morphosyntactic feature bundles and the Vocabulary items that may express them. A Vocabulary item has in principle a phonological shape (even though for some items this may be zero); a morphosyntactic feature, or bundle of features, does not. Also, one bundle of features may be expressed by different Vocabulary items in different contexts. So which, if either, of these kinds of thing is a morpheme, according to DM? The answer is that *both* kinds of thing count as morphemes. As Halle and Marantz put it (1993:114): 'we have chosen to call the terminal elements "morphemes" both before and after Vocabulary insertion, that is, before and after they are supplied with phonological features'. Thus, apparently, we are entitled to say that _waited_ and _given_ display different past participle morphemes because their affixes are different, but also that they display the same morpheme because both these affixes realize [past participle]. Halle and Marantz claim that nothing hinges on this terminology. However, it undermines what Halle sees as the clinching argument for zero affixes in English verbs, namely the argument from do-support. The argument runs like this. In yes-no question contexts, lexical verbs cannot carry Tense-Number marking; this marking must instead be realized on the dummy verb _do_, as in: (1) Doe-s John snore? (2) Di-d John snore? But this _do_ shows up even when the Tense-Number features in question are [non-3rd-singular present]: (3) Do they snore? So, even though [non-3rd-singular present] is never phonologically overt in any context, it must be realized by a zero affix in order to trigger do-support. In Morphological Structure, therefore, (3) is a better represented as (4): (4) Do-0 they snore? This argument fails because it relies on a blurring of the distinction between morpheme-as-feature-bundle and morpheme-as-affix. One may agree that a feature bundle needs a peg on which to get realized, without agreeing that the realization must inevitably be carried out by a morpheme-as-affix. To substantiate the claim that a zero morpheme-as-affix is present in (3) (or (4)), Halle needs to do more than just invoke the phenomenon of do-support. The fact that he does not do more is at least partly due, I suggest, to DM's use of the term 'morpheme' to designate both syntactic terminal elements (feature bundles) and also affixes. Let us grant that the morpheme-as-feature-bundle [non-3rd-singular present] occupies a terminal position in a syntactic tree. It does not follow that there must be an independent Vocabulary item (a morpheme-as-affix) to realize it -- unless of course the theory of DM decrees that such a Vocabulary item *must* exist, thanks to the two-sided character ascribed in the theory to 'morphemes'. But in that case the postulation of a zero affix for the purpose seems no more than a trick to reconcile the theory with recalcitrant data. It is important to stress the difference between the zero morphological realization of [non-3rd-singular present] and the alleged zero affix for [past]. The feature [past] is uncontroversially realized by an affix in most contexts: _snored, sold, felt_ etc. The postulation of a zero affix in _sang, cut_ etc. therefore seems less contrived than in examples (3) and (4). Even so, the facts of do-support in the past tense count *against* the existence of a [past] zero affix, not for it. Remember that what is at issue is not whether features such as [past] need a peg on which to get realized; rather, the issue is whether this realization ever takes the form of a zero affix. Imagine an imaginary English in which the yes-no questions corresponding to (5) and (6) were not (7) and (8) but rather (9) and (10): (5) John left. (6) John sang. (7) Did John leave? (8) Did John sing? (9) Di-t John leave? (10) Di John sing? In this version of English, it is clearly not just the feature [past] that gets supported by _do_, but the actual affix that realizes [past] on the accompanying lexical verb in positive declarative contexts. In this version of English, the past tense of _do_ will usually carry an affix (_-d_ or _-t_), so one could reasonably argue that in (6) and (10) too there is an affix, albeit a phonologically null one. The forms _sang_ and _di_ would therefore be more accurately represented in Morphological Structure as _sang-0_ and _di-0_. But actual English is not like this. The fact that the dummy _do_ shows up as _did_ in both (7) and (8), ignoring the morphology of the lexical verbs in (5) and (6), shows conclusively that do-support operates at the level of morphosyntactic features, not at the level of the affixes that realize them. This further weakens the case for invoking a zero affix in a [non-3rd-singular present] context such as (3). 2. The phonological restrictions on the past (participle) 'zero affixes' Halle points out correctly that, in the DM analysis, both overt affixes and zero affixes may or may not trigger readjustments. As he puts it: "As Past exponent, zero sometimes causes phonetic modifications in the stem to which it attaches, sometimes not. In this again zeros parallel other Past tense exponents". He also provides a list of verbs in which an (alleged) zero Past exponent causes no phonetic modification, namely _beat, shed, cast, burst, hurt, rid, spread_. But he does not comment on the striking fact that all these verbs end in in coronal stops. If one rejects the DM analysis for these forms, plausible avenues for explanation open up: perhaps these forms have an underlying overt coronal suffix, forming a cluster that gets simplified, or perhaps they have no suffix at all (even a zero one) because they already conform to a past-tense template. Under the DM analysis, however, no such explanation seems available. Why should the zero affix without readjustment favor precisely coronal stems, rather than (say) bilabial ones, or disyllabic ones, or ones with high vowels? Halle does not say. Likewise, Halle does not say why, for Past Participle, the zero affix without readjustment is limited precisely to these coronal stems plus also _run_ and _come_ -- a pair of verbs whose vowel complies with the pattern of _sung, struck, dug_ etc. The fact that all this seems to be relegated to mere coincidence in Halle's DM analysis must surely count against it. 3. Positive evidence for the nonaffixal character of stem alternations Does stem alternation play by the same rules as affixal allomorphy, or by different rules, or by no rules at all (as might be expected if it is merely the random residue of phonological change)? Under DM, much stem alternation is handled by readjustment, sometimes triggered by zero affixes. It will therefore count in favor of DM if stem alternation plays by substantially the same rules as affixal allomorphy does: that is, if it is found that the zeroes that one posits to trigger stem modifications obey any constraints or restrictions that affect overt affixes, and that there are no special constraints on the stem modifications themselves. Evidence on these lines, if sufficiently substantial, could even outweigh the doubts I have expressed about zero affixes in English verbforms. In fact, however, such evidence as I am aware of points firmly the other way, i.e. towards the conclusion that stem alternation plays by different rules than affixal allomorphy. I have in mind the following: (a) stem alternation is irrelevant for the purpose of 'paradigm economy' (Carstairs 1987:221-32; 1988) and 'blur avoidance' (Carstairs-McCarthy 1994); (b) patterns of stem alternation can be productive, independently of the phonological shape of the alternants (Maiden 1992; 1996), in line with the view of stems as part of 'morphology by itself' (Aronoff 1994); (c) implicative paradigm structure conditions of the kind proposed in Natural Morphology by Wurzel (1984) and applied to German verbs by Bittner (1985) are too powerful a mechanism to account for affixal 'blur avoidance' (Carstairs-McCarthy 1994:754-7), but seem well suited to describing observed patterns of stem distribution (Carstairs-McCarthy 1991:237-47). I would welcome debate on whether these observations can be reconciled with DM. For the time being, they seem to count against too ready reliance on affixation plus readjustment in order to handle stem allomorphy. Aronoff, M. 1994. Morphology by itself. Cambridge, MA: MIT Press. Bittner, A. 1985. Implikative Hierarchien in der Morphologie: Das 'Stark-schwach-Kontinuum' der neuhochdeutschen Verben. Acta Linguistica Academiae Scientiarum Hungaricae 35:31-42. Carstairs, A. 1987. Allomorphy in inflexion. London: Croom Helm. Carstairs, A. 1988. Nonconcatenative inflection and paradigm economy. Theoretical morphology: approaches in modern linguistics, ed. by M. Hammond and M. Noonan, 71-7. San Diego: Academic Press. Carstairs-McCarthy, A. 1991. Inflection classes: two questions with one answer. Paradigms: the economy of inflection, ed. by F. Plank, 213-53. Berlin: Mouton de Gruyter. Carstairs-McCarthy, A. 1994. Inflection classes, gender and the Principle of Contrast. Language 70:737-88. Maiden, M. 1992. Irregularity as a determinant of morphological change. Journal of Linguistics 28:285-312. Maiden, M. 1996. The Romance gerund and 'system-dependent naturalness' in morphology. Transactions of the Philological Society 94:167-201. Wurzel, W.U. 1984. Flexionsmorphologie und Natuerlichkeit. Berlin: Akademie-Verlag. (English translation (1989): Inflectional morphology and naturalness. Dordrecht: Kluwer.) Andrew Carstairs-McCarthy Associate Professor 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 marthajo at linc.cis.upenn.edu Wed Feb 24 23:29:14 1999 From: marthajo at linc.cis.upenn.edu (Martha McGinnis) Date: Wed, 24 Feb 1999 18:29:14 -0500 Subject: Martha McGinnis: zero affixes Message-ID: I've been following the discussion on zero affixes, and wanted to ask a couple of clarification questions. First, if I understood correctly, Morris Halle (MH) observed that Tense morphology can appear either on the main verb or on an auxiliary, depending on the syntactic environment. He concluded that the element realized by Tense morphology is a syntactic node. Andrew Carstairs-McCarthy (AC) granted this, but contended that there need not be an independent Vocabulary item (a morpheme-as-affix) to realize the Tense node. How are we to represent the fact that the Tense node is pronounced in some cases, and isn't in others? MH says that non-pronunciation has the same status as a given particular pronunciation (i.e., it's a zero Vocabulary item), while AC suggests that non-pronunciation has a different status from particular pronunciations. I'm curious as to how strict AC wants to be about ruling out zero Vocabulary items. As MH pointed out, null realization of syntactic nodes can occur in the absence of stem changes, e.g. pro-arguments in Japanese, say _0 wakar-u_ (understand-past) "I/you/he/she/they understood". I'm not sure whether AC would want to rule out zero Vocabulary items like this too. In any case, AC's point of departure seems to be the No Blur Principle (NBP), so maybe now is a good time to bring up something that puzzles me about it. The robustness of the NBP is interesting in its own right, but I can't figure out how the child would apply it as a learning principle, even without assuming zero affixes. For example, suppose that a child learns that an affix A is the default realization of a class marker on stems of type X. This means that A can appear on several classes of X (e.g. class I, II, III). According to the NBP, any other realization of that class marker, say B, will uniquely identify the class it realizes (e.g., just class IV); this is supposed to be useful (necessary?) for acquiring the system of class markers. I II III IV V VI A A A B -- -- However, in addition to learning class markers A, B, etc., the child must also learn which classes lack (overt) class markers (V, VI). Otherwise, s/he will use the default affix A for these classes. Even supposing that there are no zero affixes, why is it not a problem for the learner that the unaffixed forms don't all belong to one class? Why is this less of a problem than the problem that would arise if forms with affix B didn't all belong to one class? Given whatever stategy the child uses to acquire the relevant information about classes V and VI, why couldn't the same strategy be used to acquire the relevant information about classes V and VI in the following system, ruled out by the NBP? I II III IV V VI A A A -- B B Thanks for any help with these questions. -Martha From marthajo at linc.cis.upenn.edu Thu Feb 25 16:13:43 1999 From: marthajo at linc.cis.upenn.edu (Martha McGinnis) Date: Thu, 25 Feb 1999 11:13:43 -0500 Subject: Andrew Carstairs-McCarthy: zero affixes Message-ID: Here is a response to Martha McGinnis's queries. 1. Non-pronunciation of nodes Alluding to my rejection of a zero morpheme-as-affix realizing [present, non-3rd-singular], MMcG asks: >How are >we to represent the fact that the Tense node is pronounced in some cases, >and isn't in others? MH says that non-pronunciation has the same status as >a given particular pronunciation (i.e., it's a zero Vocabulary item), while >AC suggests that non-pronunciation has a different status from particular >pronunciations.... >As MH pointed out, null realization of syntactic nodes >can occur in the absence of stem changes, e.g. pro-arguments in Japanese, >say _0 wakar-u_ (understand-past) "I/you/he/she/they understood". This seems to me to be a question for the syntactician rather than the morphologist. How are we to represent the fact that traces are not pronounced, or big-PRO, or little-pro (in Principles-and-Parameters terms)? That is not a question that morphological theory is generally expected to answer -- and rightly so. In a DM-style analysis whereby all morphosyntactic features are in principle located in bundles at terminal nodes in syntactic trees (though possibly subject to merger, fusion and fission in Morphological Structure), the variety and incidence of terminal nodes where nothing phonologically overt gets inserted is quite high, relative to some other styles of analysis. It doesn't follow that some of these nodes *must* have an affix asserted at them, despite appearances (i.e. a phonologically empty affix) -- that is, it doesn't follow *unless one makes it follow*, through one's definition of 'morpheme'. But then the presence or absence of an affix ceases to be an empirical issue. As to whether _Wakaru_ 'I understood' involves a feature bundle [1 singular] that is realized by a zero affix, I take no position. I simply question the usefulness of a style of analysis under which such a feature bundle *must* be deemed to be realized by an affix or pronoun or whatever, even if there's nothing there phonologically. >I'm curious as to how strict AC wants to be about ruling out zero >Vocabulary items. I don't want to rule them out in principle. For example, I suspect a good case can be made for a zero affix on a form like Russian _knig_ 'of the books' (genitive plural). This cell in the paradigm of the lexeme KNIGA is highly marked in morphosyntactic terms (by contrast with e.g. nominative singular _knig-a_); all the other cells have an overt suffix; and this cell too had one in the not-too-distant past. All I am arguing is that the case for a zero affix on _cut, sang_ etc. is weak, being based mainly on a terminological fiat. 2. Learning, bare stems, and the No Blur Principle (NBP). MMcG questions the role of the NBP in inflection-class learning, and asks: >Suppose that a child learns >that an affix A is the default realization of a class marker [rather, of a >morphosyntactic cell] on stems of >type X. This means that A can appear on several classes of X (e.g. class I, >II, III). According to the NBP, any other realization of that [cell], say >B, will uniquely identify the class it realizes [rather, the class to >which lexemes carrying it belong], (e.g., just >class IV); this is supposed to be useful (necessary?) for acquiring the >system of class markers [rather, inflectional affixes]. > > I II III IV V VI > A A A B -- -- > [Table 1] >However, in addition to learning class markers A, B, etc., the child must >also learn which classes lack (overt) class markers (V, VI). Otherwise, >s/he will use the default affix A for these classes. Even supposing that >there are no zero affixes, why is it not a problem for the learner that the >unaffixed forms don't all belong to one class? Why is this less of a >problem than the problem that would arise if forms with affix B didn't all >belong to one class? This is a nice question, and I don't think the evidence is yet available to give a complete answer. A research project for someone, then! One needs to distinguish between two tasks: (a) learning what a given affix (A or B in Table 1) means, and (b) learning which class a given lexeme belongs to. >>From the point of view of learning task (a), classes V and VI in Table 1 present no problem, because in the cell in question they carry no affix. (I'm assuming there is no evidence for a 'real' zero affix here, as in _knig_.) In this respect Table 1 contrasts directly with Table 2: I II III IV V VI A A A -- B B [Table 2] That's why Table 2 can't be learned by using the same strategy as for Table 1: Table 2 contains two blurred affixes, A and B, so fulfilling learning task (a) in respect of Table 2 is predicted by the NBP to be impossible or at least difficult. But what about task (b) in respect of Table 1? In particular, which lexemes go in class V and which go in class VI? Let's suppose that, for a different cell (cell 2), class V lexemes carry a class-identifier, E: I II III IV V VI Cell 1 A A A B -- -- Cell 2 C D C C E C [Table 3] Once the child has spotted that C is the class-default affix for cell 2, she 'knows' that affix E must be a class-identifier, i.e. that all lexemes with E are inflected alike (at least affixally), and differently (in at least cell 2) from any lexeme with C. So she 'knows' that classes V and VI are distinct. It is conceivable, though, that V and VI differ in that V has in cell 2 not a class-identifying affix but no affix at all: I II III IV V VI Cell 1 A A A B -- -- Cell 2 C D C C -- C [Table 4] This poses more of a problem for learning task (b). Let's suppose that neither V nor VI has any class-identifier in any other cell (cell 3, 4, etc.). Then there seems to be nothing for the child to cotton on to for the purpose of 'fixing' the membership of V and VI. So it is tempting to hypothesize that such a pattern could never arise, i.e. that every class must have a class-identifying affix in at least one cell. This represents a tightening of the NBP as presented in my _Language_ article, so it will be nice if it is true. Is there evidence to support it? At first sight, German nouns seem to count against. In Table 4 on page 745 of _Language_ 70 (1994), classes II (Ziegel) and IX (Dorn) are represented as both lacking a class-identifier, and differing only in that, in the Nondative Plural, II has a blank while IX has the class-default suffix _-(e)n_. But, as I note on that page, the Plural blank of class II should probably regarded as a phonologically conditioned variant of the _-e_ suffix of class I/III (Tag). If so, this counterexample is taken care of. There is still a problem with English verbs, however, as represented in Table 5 on page 746. If we ignore the minuscule and (for me) obsolete class V (_crow, dive_), _-en_ becomes the class-default Past Participle suffix. But then class III (_give_) and class IV (_show_) exemplify the pattern that concerns us, i.e. no class-identifiers, and differentiation only by a blank versus a class-default affix in one cell (Past _gave--_, PastPrt _give-n_ versus _show-ed, show-n_). If the strengthened version of the NBP is too strong, then, the problem of fulfilling task (b) in respect of patterns like Table 4 above remains. But notice one fact about the English examples: the suffixless Past form _gave_ is by no means unmarked for [past], because its ablauted vowel is peculiar to it. So perhaps it is stem alternation of this kind that renders inflection class systems learnable in circumstances where some class (or, perhaps, more than one class) has no class-identifying affix. The fact that stem alternation as such is not relevant to blur avoidance need not entail that it is irrelevant to determining inflection class membership. (Recall my comment in my previous positing that, at least in German verbs, stem alternation seems to lend itself well to description in terms of Wurzel-style implicative paradigm structure conditions, which involve lexical specification.) This possibility raises the whole fascinating but complex issue of the relationship between affixal and nonaffixal inflection. All the more important, then, that the distinction between the two kinds of inflection should not muddied through a proliferation of zero affixes! Andrew Carstairs-McCarthy Associate Professor 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