Journal of Germanic Linguistics 17.4 (2005):225–264
Paradigm Uniformity Effects in German Phonology Tracy Alan Hall Indiana University German words like Siedl-ung [zid.l] ‘settlement’, which contain a sequence of obstruent plus sonorant consonant between vowels, pose two problems assuming the heterosyllabic parse: (a) the first syllable is superheavy, but superheavy syllables are otherwise restricted in their occurrence to word-final position; and (b) the /d/ does not devoice, contrary to what one would expect, since Final Devoicing operates without exception in coda position. The problem in assuming that consonant clusters like [dl] are tautosyllabic is that they are otherwise non-occurring in word-initial position. In the present study, it is argued that the heterosyllabic parse for words like Siedl-ung is correct, and that (a) and (b) derive an explanation by appealing to paradigm uniformity. It is shown in an optimality-theoretic analysis that the canonical patterns capturing the distribution of superheavy syllables and Final Devoicing require the ranking “markedness » input-output faithfulness.” The paradigm uniformity effects, that is, (a) and (b), require a specific output-output faithfulness constraint to dominate the markedness constraint. The proposed paradigm uniformity approach is compared to earlier treatments in which the notion of paradigm does not play a role. It is also shown that the tautosyllabic parse in examples like Siedl-ung is problematic.
1. Introduction. In this article I discuss a large class of German words, which can be illustrated with the two representative examples in 1. (1) a. Atm-ung b. Siedl-ung
[at.m] [zid.l]
‘breathing’ ‘settlement’
This paper was presented at Indiana University in December 2004 and at the Old World Conference in Phonology 2 (OCP-2) at the University of Tromsø in January 2005. Thanks are due to the respective audiences for their comments and suggestions. I would also like to thank two anonymous JGL reviewers for their remarks on an earlier written version. All disclaimers apply. © Society for Germanic Linguistics
226
Hall
These words have in common that they contain a sequence of obstruent plus sonorant consonant between vowels. Canonical syllabification seems to require that these sequences be heterosyllabified, since obstruent plus sonorant sequences like these are either non-occurring or extremely rare in onset position, for example, in word-initial position. However, the heterosyllabic parsing of the examples in 1 is problematic for two reasons. First, the long vowel plus consonant rhymes (that is, [at], [id]) are examples of superheavy syllables, but superheavy syllables are otherwise restricted in their occurrence to word-final position (Hall 2002a,b). Second, the /d/ in the example in 1b does not devoice, which is unexpected because German Final Devoicing operates in coda position and it is assumed not to have idiosyncratic exceptions. In the present study I show that the two problems described with respect to the data in 1 evaporate when one considers PARADIGM UNIFORMITY (henceforth PU). The basic idea behind PU is that otherwise pervasive surface-based markedness constraints can be violated in order to make the relevant forms similar to morphologically related ones. In terms of Optimality Theory (OT; Prince and Smolensky 1993), the canonical phonological pattern is typically captured with the ranking of markedness over input-output (henceforth I-O) faithfulness, as in 2a. With respect to the examples in 1, the canonical phonology in 2a would say that obstruent plus sonorant consonant clusters are heterosyllabified. The PU effects referred to above require that the markedness constraint in 2a be dominated by an output-output (henceforth O-O) faithfulness constraint, as in 2b.1 (2) a. Canonical pattern: markedness » I-O faithfulness b. PU effects: O-O faithfulness » markedness » I-O faithfulness What I ultimately argue is that the two problems referred to above with respect to 1 are examples of PU effects and that they therefore require a specific instantiation of the general ranking in 2b. In the course of my analysis I briefly compare my PU approach to the data in 1 with the earlier treatments of these examples, which employ either cycles, lexical 1
For recent studies on PU effects in languages other than German the reader is referred to Benua 1997, and to the contributions in Downing, Hall, and Raffelsiefen 2005. For a study in which PU effects are discussed in other areas of German phonology, see Raffelsiefen 1995.
Paradigm Uniformity Effects
227
levels, or language-specific resyllabification rules (for example, Rubach 1990, Hall 1992) or an OT-style analysis without O-O constraints (for example, Noske 1999). I also show that these earlier approaches in which the obstruent plus sonorant consonant sequences as in 1 are analyzed as onsets are problematic (see, for example, Vennemann 1972 and Hall 1992). This article is structured as follows. In section 2 I discuss two generalizations pertaining to the canonical phonology illustrating the general ranking in 2a, namely the distribution of superheavy syllables (section 2.1) and the syllabification of intervocalic obstruent plus sonorant clusters (section 2.2). In section 2.3 I turn to examples like those in 1 that go contrary to the canonical distribution of superheavy syllables, and argue that these words illustrate PU effects with respect to moraic structure. Here I show that the problematic examples require an O-O constraint referring to the moraic structure of vowels to outrank the markedness constraint capturing the distribution of superheavy syllables, as in 2b. In section 3 PU effects with respect to Final Devoicing are discussed. In section 3.1 I concentrate on the canonical pattern and show, following the majority of researchers who have addressed this problem, that the process operates in the syllable coda. In section 3.2 I turn to examples like the one in 1b and argue that these words require a highranked O-O constraint that refers specifically to the feature involved in Final Devoicing. It is demonstrated that this O-O constraint must be locally conjoined with the markedness constraint SYLLABLE CONTACT LAW (Murray and Vennemann 1983, Davis 1998) in order to prevent the alternations motivating Final Devoicing in the canonical phonology from exhibiting PU effects. In section 4 I discuss and refute two alternative treatments for Final Devoicing not involving PU effects, and in section 5 I present the conclusion. 2. Paradigm Uniformity Effects in Phonotactics. In this section I discuss two areas of German canonical phonology, namely the distribution of superheavy syllables (section 2.1) and the syllabification of intervocalic obstruent plus sonorant clusters (section 2.2). In section 2.3 I discuss the PU effects and propose an OT treatment.
228
Hall
2.1. The Distribution of Superheavy Syllables: The Canonical Pattern. Superheavy syllables are defined here as syllables ending in a short vowel plus two consonants (see 3a), a long vowel plus one or more consonants (see 3b), or a diphthong plus one or more consonants (see 3c).2 (3) Superheavy syllables a. short vowel + two or more consonants Kalb Amt
[kalp] [amt]
‘calf’ ‘office’
b. long vowel + one or more consonant viel zahm
[fil] [tsam]
‘many’ ‘tame’
c. diphthong + one or more consonant Bein Baum euch
[ban] [bam] [ç]
‘leg’ ‘tree’ ‘you (pl.)’
As in English and Dutch, superheavy syllables in German are restricted in their distribution, as I have illustrated in 4.3 In 4, we can see that superheavy syllables occur in word-final position, as in 4a, before the first part of a compound, as in 4b, and before a consonant-initial suffix, as in 4c.4 The important restriction for our purposes is illustrated in 4d, that is, monomorphemes in which a superheavy syllable occurs internally are lacking. In 4d, I present a hypothetical monomorpheme of this structure.
2
All examples in the present article have been taken from Drosdowski et al. 1990. 3 See Hall 2002b and Raffelsiefen 2004 for more discussion and detailed analyses. 4 Before vowel-initial suffixes potential superheavy syllables do not surface because the final consonant of the stem surfaces in the following onset, for example, viel-e [fi.l] ‘many (adj. ending)’.
Paradigm Uniformity Effects
229
(4) Distribution of superheavy syllables a. Superheavy syllables in word-final position Werk Zeit Buch
[vk] [tsat] [bux]
‘work’ ‘time’ ‘book’
b. Superheavy syllables before a compound boundary Werk-statt Zeit-geist Buch-weizen
[vk.tat] [tsat.gast] [bux.va.tsn]
‘workshop’ ‘zeitgeist’ ‘buckwheat’
c. Superheavy syllables before a CV(C) suffix leb-los Ein-heit lieb-lich
[lep.los] [an.hat] [lip.lç]
‘lifeless’ ‘unit’ ‘dearly’
d. No superheavy syllables morpheme-internally *[a.el.na]
(see Arena [a.e.na] ‘arena’)
Following Hall 2002b and Raffelsiefen 2004, I argue that German restricts the distribution of superheavy syllables to the right edge of a phonological word, as stated in the descriptive generalization in 5.5 (5) A superheavy syllable occurs only at the right edge of a phonological word. With respect to the data in 4, there is consensus in the literature on German that each part of a compound (as in 4b) forms its own phonological word and that consonant-initial suffixes (as in 4c) are noncohering in the sense that they lie outside of the phonological word of the stem to which they attach (see Booij 1985, Wiese 1996, Hall 2002b, and Raffelsiefen 2004). These authors also agree that vowel-initial suffixes 5
Systematic counterexamples to 5 include proper names like Leipzig [lap.tsç]. In section 3.2 I show that proper names do not conform to a second phenomenon in the canonical phonology of German as well. For this reason, I do not consider proper names in subsequent data sets.
230
Hall
are cohering, that is, they form one phonological word with the stem to which they attach. The number of truly idiosyncratic exceptions to 5 is very small. Three examples of morpheme-internal superheavy syllables are Skulptur [sklp.tu] ‘sculpture’, arktisch [ak.t] ‘arctic’, and Auktion [ak.tsjon] ‘auction’. Note that these examples have in common that the superheavy syllable ends in either a short vowel plus two consonants or a diphthong plus one consonant. Counterexamples to 5, which involve superheavy syllables with a long vowel, seem impressionistically to be much rarer, for example, Börse [bø.z] ‘stock exchange’.6 The generalization in 5 is captured formally with an alignment constraint, given in 6a, saying that the right edge of a superheavy syllable (SH) aligns with the right edge of a phonological word. I am not committing myself here to how superheavy syllables are represented phonologically, since this issue is peripheral to the issues I discuss below.7 It is not enough to simply state the constraint in 6a captures the generalization in 5 because one needs to make a statement about how German would treat a potential foreign borrowing in which a superheavy syllable could potentially surface within a phonological word. The fact that German lacks examples like the one in 4d falls out from ALIGN-SH and I-O faithfulness constraints like those in 6b and 6c, which penalize the deletion of a mora and a segment, respectively. The ranking ALIGN-SH » MAX-μ in 6d is a specific instantiation of the general ranking in 2b for the canonical phonological pattern if alignment constraints are interpreted to be a kind of markedness constraint. The ranking MAX-I-O » ALIGN-SH is justified below.
6
Not surprisingly, some speakers prefer the pronunciation with the corresponding short, lax vowel, that is, [bœ.z]. 7
A reviewer notes correctly that alignment constraints like the one in 6a are assumed to refer to constituents, but that a superheavy syllable is not a constituent. One could solve this problem by representing the final consonant in a superheavy syllable as structurally different than the final consonant in heavy syllables. For example, the final consonant in the underlined sequences in 4 could be extrasyllabic, although this proposal is not in line with the analysis of Hall 2002a,b. I leave this issue open.
Paradigm Uniformity Effects
231
(6) Constraints and ranking for canonical pattern of phonotactics a. ALIGN-SH: Align the right edge of a superheavy syllable (SH) with the right edge of a phonological word. b. MAX-μ: An input mora must have an output correspondent. c. MAX-I-O: An input segment must have an output correspondent. d. MAX-I-O » ALIGN-SH » MAX-μ In the tableau in 7, I show that if the hypothetical word in 4d were to enter German as a loanword, then the ranking in 6d tells us that it would surface in such a way that the long vowel shortens.8 (7) A hypothetical word that could potentially violate ALIGN-SH a. b. c.
/aelna/ [a.l.na] [a.el.na] [a.e.na]
MAX-I-O
ALIGN-SH
MAX-μ
* *! *!
My assumption here and below is that all vowels are underlyingly moraic; hence, short vowels are prelinked to a single mora and long vowels to two. In 7, we can observe that the faithful candidate in 7b violates ALIGN-SH, and that is therefore less harmonic than the vowel shortening candidate in 7a. The additional form in 7c is important because it shows that mora deletion is preferred as a repair strategy to the deletion of an entire segment. An interesting question is if and how German repairs potential violations to ALIGN-SH in loanwords that have no long vowel, for example, a hypothetical word like [a.lm.na]. Recall that there are a few real words of this structure, for example, Skulptur [sklp.tu] ‘sculpture’, arktisch [ak.t] ‘arctic’, and Auktion [ak.tsjon] ‘auction’. Significantly, the ranking MAX-I-O » ALIGN-SH in 7 correctly selects the 8
Hall and Hamann (2003) concede that historical examples illustrating vowel shortening as in 7 are difficult to find. As noted in that article (p. 71), one possible historical example illustrating this kind of vowel shortening is the word Rektor [k.to] ‘rector’, because the [] was historically the long (tense) vowel [e] in the donor language Latin.
232
Hall
respective faithful forms. (There is no tendency to insert a vowel as a repair strategy; hence, the constraint penalizing output forms with epenthesis (DEP-I-O) is ranked ahead of ALIGN-SH.) A set of morphologically complex words with surface [VC] superheavy syllables in violation of ALIGN-SH is discussed in section 2.3. These examples pose an apparent problem for my analysis in 7, because the word-internal [VC] syllable is not repaired by vowel shortening (or in some other way). I argue ultimately that these data require that ALIGN-SH be dominated by an O-O constraint because I show that they involve a PU effect. Before I discuss these data, however, I consider briefly in section 2.2 canonical syllabification patterns because I need to clarify how my analysis captures well-formed versus ill-formed onset clusters. 2.2. Canonical Syllabification. The data in 8a consist of words containing two intervocalic consonants, the first of which is an obstruent and the second of which is a sonorant consonant (abbreviated henceforth as OS). In these words, the OS sequence is a permissible onset, as indicated in 8b, in which an example is provided of a word beginning with the respective OS sequence. The S in these examples can be / l n/.9 (8) VOSV is parsed V.OSV if and only if OS is a permissible onset a. Metro Afrika Kobra Iglu Problem Magnet
9
[me.to] [a.fi.ka] [ko.ba] [i.glu] [p.blem] [ma.gnet]
‘metro’ ‘Africa’ ‘cobra’ ‘igloo’ ‘problem’ ‘magnet’
There are many German words in which the S in OS is /m/, but unlike the examples in 8, these clusters are non-occurring word-initially (see 10 and 26b below). The only other sonorant consonant is //, which cannot occur as the second member of a consonant cluster.
Paradigm Uniformity Effects
b. Tritt frei bringen Glanz Probe Gnom
[tt] [fa] [bn] [glants] [pob] [gnom]
233
‘kick’ ‘free’ ‘bring’ ‘brightness’ ‘sample’ ‘gnome’
It can be observed from the phonetic forms in 8a that the OS sequence is syllabified as an onset—a parsing considered uncontroversial in the literature. There are two arguments for this parsing as opposed to a heterosyllabic one. First, if the OS sequence were to be heterosyllabified, then examples such as the first four would constitute violations to ALIGN-SH. Second, a heterosyllabic parse would be problematic for the final four examples in 8a, because Final Devoicing, which affects an obstruent in coda position, would incorrectly be predicted to apply to the O in these OS sequences.10 The examples in 9a are structurally similar to the ones in 8a because they also contain an intervocalic OS, which surfaces as an onset. The difference between the words in 8a and 9a is that the vowel following OS in the latter words begins a suffix. The examples in 9b show that the intervocalic OS sequences in 9a are well-formed onset clusters because they surface word initially. (9) VOS-V is parsed V.OS-V if and only if OS is a permissible onset
10
a. fiebr-ig ekl-ig nebl-ig regn-en
[fi.bç] [e.klç] [ne.blç] [e.gnn]
‘feverish’ ‘disgusting’ ‘foggy’ ‘rain’
b. Brot Klang Blei Gnom
[bot] [kla] [bla] [gnom]
‘bread’ ‘sound’ ‘lead’ ‘gnome’
See section 3.1 below for data and justification that Final Devoicing is syllable-based.
234
Hall
In 10, we can see words that are structurally similar to the ones in 8a, the crucial difference being that the OS sequence in the former words is not a permissible onset in the sense that there are no words beginning with these sequences. The heterosyllabic parsing here is uncontroversial. I give here only examples in which the O is voiceless. (10) VOSV is parsed VO.SV only if OS is not a permissible onset Atlas Athlet Technik ethnisch Schrapnell Rhythmus Atmosphäre
[at.las] [at.le t] [tç.ni k] [t.n] [ap.nl] [t.ms] [at.mo.sf .]
‘atlas’ ‘athlete’ ‘technology’ ‘ethnic’ ‘shrapnel’ ‘rhythm’ ‘atmosphere’
Examples like those in 10, in which the O is voiced, are discussed in section 3, which concerns itself with Final Devoicing. Words such as those in 10, in which the OS is followed by a vowel that begins a suffix, are discussed in section 2.3 below because they are the ones that are argued to exhibit PU effects. Let us now consider the two markedness constraints necessary to account for well-formed syllable-initial clusters. The first is SON (see 11a), posited independently by Raffelsiefen (1995) and Hall (2002a) to account for sonority sequencing in German. As stated below, this constraint is undominated in German because sequences that violate it (for example, *[ld, *[nd) are unpronounceable. Since sequences like these in other languages are repaired if they enter German as loanwords (see note 11 below), SON clearly must outrank the faithfulness constraints MAX-I-O, and DEP -I-O. The second markedness constraint necessary is OWF in 11b, which is simply shorthand for the constraints that rule out OS onsets satisfying SON, for example, *[tl, *[dl, and *[tm. Illicit clusters like *[tm are significant because they show that OWF cannot simply be reduced to some constraint specifying that the two consonants are homorganic. As pointed out by an anonymous referee, clusters like these are pronounceable (in contrast to the ones violating SON), and, in fact, they are attested in a few marginal examples in word-initial position; for example, Tmesis [tme zs] ‘tmesis (the separation of the parts of a compound word)’ and Tlingit [tlgt] ‘Tlingit
Paradigm Uniformity Effects
235
(language name)’. To account for the fact that clusters violating OWF are not repaired in some way, OWF must be outranked by faithfulness (that is, MAX-I-O and DEP-I-O). The ranking of SON, OWF, and faithfulness is presented in 11c.11 (11) Constraints and ranking for syllable-initial clusters a. SON: A sonorant consonant in the syllable onset may be preceded only by obstruents (for example, *[ld and * [nd). b. ONSET WELLFORMEDNESS (OWF): Certain OS sequences are ungrammatical in the onset (for example, *[tl, *[dl, and *[tm). c. SON » FAITHFULNESS (MAX-I-O, DEP-I-O) » OWF The syllabification patterns in 8–10 require OWF as well as the two markedness constraints in 12a,b, which are familiar from the literature (see, for example, Prince and Smolensky 1993). (12) Constraints and ranking for canonical syllabification patterns 8–10 a. NOCODA: The syllable is open. b. NOCOMP ONSET:At most one segment occurs syllable-initially. c. OWF » NOCODA » NOCOMP ONSET The ranking in 12c captures the syllabification patterns in 8–10, as illustrated in the two tableaux in 13:
11
The constraint MAX-I-O must outrank DEP-I-O because clusters violating SON are repaired by epenthesis and not deletion. For example, the Bantu language Ndao [ndao] would be pronounced in German [ndao] (with epenthesis, that is, sonorant syllabification), and not [dao] or [nao] (with deletion). In 11c, I simply have both faithfulness occupying the same niche in the constraint hierarchy because D EP-I-O plays no role in the following sections.
236
Hall
(13) Tableaux illustrating canonical syllabification OWF
a. b.
/pblem/ [p.blem] [pb.lem]
OWF
c. d.
/atlas/ [a.tlas] [at.las]
NOCODA
NOCOMPONSET
* *! NOCODA
NOCOMPONSET
*!
* *
In 13, we can observe that the candidate in 13b with the heterosyllabic parse is not optimal because it incurs a fatal violation of NOCODA, so 13a wins out. By contrast, in the structurally similar word /atlas/, the form with the heterosyllabic parse in 13d is selected over the one in 13c because the latter candidate violates OWF fatally. Since NOCODA and NOCOMP ONSET do not play a role in the following analysis, I ignore them in all subsequent tableaux. In the tableau in 14, I have presented a hypothetical (and phonotactically ill-formed) word, which is akin to the hypothetical example in 7. The difference between the hypothetical example in 14 and the real example in 13d is that the first vowel is short in the latter word, whereas it is long in the former one. (14) A hypothetical word that could potentially violate ALIGN-SH if it surfaced faithfully a. b. c. d.
/atlas/ [a.tlas] [at.las] [a.las] [at.las]
MAX-I-O
OWF
ALIGN-SH
MAX-μ
*! *! *!
* *
The tableau in 14 shows that if a word like /atlas/ were the input, then the expected outcome would be the form in 14d with vowel shortening. Candidate 14a cannot be selected because it violates OWF, candidate 14b because it violates ALIGN-SH, and candidate 14c because it violates MAX-I-O. Note that my analysis predicts that clusters that could potentially violate OWF if they were to surface in the onset are always
Paradigm Uniformity Effects
237
heterosyllabified if these clusters are word internal, for example, the /tm/ in Rhythmus in 10. By contrast, word-initial clusters violating OWF surface faithfully; for example, the marginal examples Tmesis and Tlingit mentioned above. One question my analysis does not answer is why there are so few examples of words like Tmesis and Tlingit. This is a question I leave open. 2.3. The Distribution of Superheavy Syllables: PU Effects. In this section, I show that ALIGN-SH has systematic counterexamples that, I argue, make sense if they are analyzed as a PU effect. From a formal point of view, ALIGN-SH is shown to be outranked by a specific O-O constraint referring to the moraic structure of vowels. The problem discussed below has not been discussed in the literature on German to my knowledge. The examples in 15a are like the ones in 10 because they contain an intervocalic OS, in which OS is not a permissible onset. Note that, in contrast to the words in 10, the items in 15a have a morpheme boundary between the OS sequence (which occurs at the end of the stem) and the following vowel (which is the first segment of the suffix). It should also be noted that the vowel before OS in 15a is long. There is no generalization concerning the kind of suffix that can be present in these words, since it can be either derivational or inflectional. The underlying form of the stems in 15a contains no schwa, as I have shown in 15c. Thus, I adopt the majority view in German phonology that the schwa in stems of this structure is epenthetic (Giegerich 1987, Wiese 1988, Hall 1992, Noske 1993, Wiese 1996, and Noske 1999). The O in the OS sequences of 15a is either voiceless (in the first two examples) or voiced (in the final four). That Final Devoicing does not apply here is an issue discussed in section 3.2 below.
238
Hall
(15) Words containing VOS-V; OS is not a permissible onset (see 15a) a. Atm-ung atm-e Siedl-ung Siedl-er Basl-er ein-ebn-en
[at.m] [at.m] [zid.l] [zid.l] [baz.l] [an.eb.nn]
‘breathing’ ‘breathe, 1sg.’ ‘settlement’ ‘settler’ ‘one from Basel’ ‘level off’
b. Atem siedel-n Basel eben
[a.tm] [zi.dln] [ba.zl] [e.bn]
‘breath’ ‘settle’ ‘Basel’ ‘even’
c. /atm/ /zidl/ /bazl/ /ebn/
The words in 15a are significant because they are not in line with ALIGN-SH. The reason there are word-internal superheavy syllables in these examples is that the canonical phonology (that is, the ranking in 13) demands that the OS sequence be heterosyllabified, and since the preceding vowel is long, the result is a word-internal superheavy syllable. The claim that the OS clusters in 15a are tautosyllabified is often made in the literature, but, as shown below, this view is problematic. I argue that the violations of ALIGN-SH occur in 15a because of pressure to maintain the length of the stem vowel in all words containing the same stem, as can be observed in 15b. An analysis of the words in 15a with the ranking in 14 incorrectly predicts that the long vowel in words like Atmung is shortened. The problem is made clear in the tableau in 16 for the word Atmung— although 16a is the intended winner, 16c is incorrectly selected (indicated here and below as ). (16) The incorrect winner is selected given the ranking: [MAX-I-O » OWF » ALIGN-SH » MAX-μ] a. b. c. d. e.
MAX-I-O /atm/ at.m a.tm at.m a.tm a.m *!
OWF
ALIGN-SH
MAX-μ
*! *! *!
* *
Paradigm Uniformity Effects
239
The tableau in 16 shows that 16b and 16d cannot be selected as optimal because they are not in line with OWF, and that 16e likewise incurs a fatal violation of MAX-I-O. The intended winner in 16a loses out to the incorrect winner in 16c because the former form violates ALIGN-SH. The problem in 16 is straightforwardly resolved if the analysis refers to paradigms like those in 17 for the pair {Atmung, Atem}. (17) Six paradigms for {Atmung, Atem}: A correct, B–F incorrect A at.m a.tm
B at.m a.tm
C at.m a.tm
D at.m a.tm
E a.m a.tm
F a.tm a.tm
I offer no formal definition of paradigm here (see Downing, Hall, and Raffelsiefen 2005 for various proposals). For the purposes of this article paradigms can be thought of as a set of morphologically related words, which are evaluated together in a tableau. In 17, we can observe six possible paradigms for the pair {Atmung, Atem}. 17a is correct, while 17(b–f) are not. The paradigms in 17 have been simplified for the present analysis because they only contain two members. A complete paradigm for the stem Atem would also include all of the inflected forms, for example, [at.m] ‘breathe (1sg.)’. The simplified paradigms do not affect the analysis presented below. The present treatment requires the two additional constraints in 18a,b. The former constraint, O-O-IDENT-μ-V, is a specific example of an O-O constraint because it refers to the relationship between the two output forms in a paradigm. O-O-IDENT-μ-V states that the moraic structure of a vowel in some stem is identical with the moraic structure of the same vowel in derived words containing the same stem. Constraint 18b is a markedness constraint, which derives motivation from other languages (see Murray and Vennemann 1983 and Ham 1998). It needs to be stressed here that the present analysis requires not only reference to the output forms in paradigms, but also to the underlying representation (that is, the input). This point is made clear below.
240
Hall
(18) Constraints and rankings required for the PU effect a. O-O-IDENT-μ-V: The moraic structure of a vowel in “stem A” is identical with the moraic structure of the same vowel in derived words containing “stem A.” b. STRESSED SYLLABLE LAW (SSL): Stressed syllables are heavy. c. O-O-IDENT-μ-V, SSL » MAX-I-O » OWL » ALIGN-SH » MAX-μ The ranking in 18c accounts for the fact that for the form Atmung in 17a there is a violation of ALIGN-SH. Consider now the tableau in 19. The six candidates correspond to the six paradigms in 17. Note that the ranking O-O-IDENT -μ-V » ALIGN-SH is a specific instantiation of the general ranking in 2b for PU effects. (19)
Correct winner is selected, given the ranking in 18c /atm-/
a. b. c. d. e. f.
A B C D E F
O-O-IDENT-μ-V
SSL
*! *!
*
MAX-I-O
OWF
ALIGN-SH MAX-μ
* *
*! *! *!
* * ** ** **
Consider first paradigms B and C. We can observe in 19 that both of these paradigms violate the O-O constraint O-O-IDENT-μ-V. The reason is that for paradigm B the first vowel in Atmung is long (see 17), but the same vowel is short in Atem. For C, the first vowel in Atmung is short and the same vowel in Atem is long. Since the two vowels in paradigms B and C differ in terms of moraic structure, they violate O-O-IDENT-μ-V. Consider now candidates A and D. Both paradigms are in line with O-OIDENT-μ-V, because the vowels in the respective stems have the same number of moras. However, in D the word Atem violates the SSL, because the first syllable is stressed and yet it is monomoraic. An examination of candidate A reveals that neither of the two words violates the SSL, and for that reason this paradigm is more harmonic. A comparison of the winner in A with E reveals that the latter one cannot
Paradigm Uniformity Effects
241
be selected as optimal because it is not in line with MAX-I-O.12 Paradigm E is therefore important, since it reveals that the present analysis must also refer to the underlying representation (that is, /atm-/) and not only to the output forms in paradigms. Paradigm F is less harmonic than A, because the former one is not in line with OWF (which is violated in its first member [a.tm], see 17). What the tableau in 19 therefore shows is that an otherwise pervasive markedness constraint—in this case it is the constraint ALIGN-SH—is violated in certain words in order to ensure that a PU effect for vowel length is maintained. 3. PU Effects in Final Devoicing. In section 3, I show that the process of Final Devoicing has a systematic set of counterexamples that I argue are accounted for by appealing to PU. I show first in section 3.1 that the canonical pattern of Final Devoicing operates in the syllable coda, and that it requires a specific instantiation of the ranking in 2a, that is, markedness » I-O faithfulness. In section 3.2, it is demonstrated that this markedness constraint is dominated by a specific O-O constraint referring to the feature [voice]. This O-O constraint must be locally conjoined with the markedness constraint SYLLABLE CONTACT LAW (Murray and Vennemann 1983, Davis 1998)
12
Recall that one of the defining characteristics of the words in 15a is the length of the stem vowel; hence, in 19 a word was evaluated in which the stem vowel is long. Consider now the examples below: i. eitl-es [at.ls] ‘vain (adj. ending)’ ii. Ordn-ung [d.n] ‘order’ In these words, we can observe that ALIGN-SH violations involve either a diphthong plus consonant (in i) or short vowel plus two consonants (in ii). That the present treatment correctly predicts these forms to be the optimal ones falls out from the ranking MAX-I-O » ALIGN-SH in 19. To see this, consider the correct paradigm for Ordnung and the related word orden-t-lich [dntlç] ‘orderly’: {[d.n], [.dnt.lç]} with its closest competitor {[.n], [.dnt.lç]}. The latter paradigm is less harmonic than the former one because the first word ([.n]) is not in line with MAX-I-O. Recall from section 2.1 that the ranking MAX-I-O » ALIGN-SH is required to account for the fact that word-internal superheavy syllables in the three rare examples Skulptur, arktisch, and Auktion surface faithfully.
242
Hall
in order to prevent the alternations motivating Final Devoicing in the canonical phonology from exhibiting PU effects. 3.1. Final Devoicing: The Canonical Pattern. The examples in 20a illustrate that German has alternations between voiced and voiceless obstruents. Words like the ones in 20b show that the alternations in 20a cannot be analyzed as a process involving the voicing of underlying voiceless obstruents in prevocalic position. That Final Devoicing sets in at the end of the stem in the second column of 20c, but not in the fifth column, can be explained by analyzing the domain of Final Devoicing as (a) the syllable coda, or (b) the right edge of a phonological word. Recall from section 2.1 that consonant-initial suffixes like -sam lie outside of the phonological word of the stem to which they attach. I follow the majority view that (a) is correct (see Vennemann 1972, Wiese 1988, Rubach 1990, Hall 1992, Wiese 1996, Noske 1999, Lombardi 1999). The problem with the (b) alternative is discussed below. (20) Alternations between voiced and voiceless obstruents a. Dieb Rad Tag Nerv Haus orange
[dip] [at] [tak] [nf] [has] [oa]
‘thief’ ‘wheel’ ‘day’ ‘nerve’ ‘house’ ‘orange’
Dieb-e Rad-es Tag-e nerv-ös Haus-es orang-e
[di.b] [a.ds] [ta.g] [n.vøs] [ha.zs] [oa. ]
b. bunt krank
[bnt] [kak]
‘colorful’ ‘sick’
bunt-e krank-e
[bn.t] ‘colorful (adj.)’ [ka.k] ‘sick (adj.)’
c. streb-sam Bünd-nis bieg-sam les-bar
[tep.zam] [b nt.ns] [bik.zam] [les.ba ]
‘ambitious’ ‘alliance’ ‘bendable’ ‘readable’
streb-e Bund-es bieg-en les-en
[te.b] [bn.ds] [bi.gn] [le.zn]
‘thieves’ ‘wheels’ ‘days’ ‘nervous’ ‘houses’ ‘orange (adj.)’
‘strive, 1sg.’ ‘alliance, gen.sg.’ ‘bend, 1pl.’ ‘read,1pl.’
Paradigm Uniformity Effects
243
Some recent work has argued that the alternations in 20 are not accounted for with a rule altering the feature [voice], but instead [spread glottis]; for example, Iverson and Salmons (1995), and Jessen and Ringen (2002). In the present analysis, I am assuming that [voice] is the correct feature. An alternative treatment of the canonical “devoicing” pattern and the morphologically complex examples in section 3.2 in terms of [spread glottis] would be compatible with my analysis provided that there is some equivalent O-O constraint for that feature (as opposed to the feature [voice]). This is an alternative I leave open for further study. The additional examples in 21 are similar structurally to the ones presented earlier in 10. Both sets of examples are monomorphemic loanwords that contain an intervocalic CC cluster in which CC is not a well-formed onset. In 21a and 10 the consonant cluster is OS, and in 21b it is OO. Note that in all of the examples in 21a,b the syllable-final obstruent is voiced in the donor language but devoiced in German.13 In the first two examples in 21b, the second O is voiced; hence, the devoicing of the first O cannot be attributed to regressive assimilation. The devoicing pattern in 21 makes sense if the OS and OO sequences are analyzed as heterosyllabic.
13
A remark in an unpublished manuscript by Steriade (1997:40, note 18) suggests that she does not believe devoicing before nasals, as in 21a, to be the canonical pattern. Her evidence is that no Final Devoicing applies before nasals in proper names (for example, Abner, Bodmer) and in the obscure nautical term Bodmerei ‘bottomry’. As stated in note 5, there are good reasons for ignoring proper names. Drosdowski et al. (1990) transcribe Bodmerei with a [d], but I do not consider a single example of a word no one has heard of to be enough to invalidate my claim that the devoicing pattern in 21 is canonical. More generally, I do not discuss the licensing-by-cue approach to Final Devoicing as proposed in Steriade 1997, because it cannot account for all of the data in this section, in particular the examples illustrating PU effects in section 3.2.
244
Hall
(21) VCCV parsed VC.CV, only if CC not permissible onset (as in 10) a. Admiral Badminton Kidney(bohne) Kadmium Charisma
[at.mi.al] [bt.mn.tn] [kt.ni] [kat.mi.m] [ka.s.ma]
‘admiral’ ‘badminton’ ‘kidney bean’ ‘cadmium’ ‘charisma’
b. Rugby absurd Wodka
[ak.bi] [ap.zt] [vt.ka]
‘rugby’ ‘absurd’ ‘vodka’
Since the examples in 21, in contrast to the ones in 20, do not have alternants in German with a voiced obstruent, I assume that the underlying representations have been restructured with voiceless obstruents, for example, /akbi/ and not /agbi/ for Rugby. Note that this restructuring took place in coda position and not at the right edge of a phonological word. For this reason the examples in 21 provide evidence against the phonological word environment in 21b above, but they make sense given the coda context in 21a.14 I account for the Final Devoicing data in 20 (and the restructuring of the underlying forms in 21) with the constraints in 22a,b. Constraint 22a is a markedness constraint penalizing voiced obstruents in coda position, and the second one is an I-O faithfulness constraint penalizing the change from [+voice] to [–voice].15
14
Jessen and Ringen (2002:212) argue that their equivalent of Final Devoicing operates at the right edge of a phonological word: “A prosodic word-final obstruent is [spread glottis].” However, it is unclear how Jessen and Ringen’s treatment accounts for the examples in 21. Steriade (1997:58) considers devoicing in the examples in the second column of 20c to be PU effects. I follow Raffelsiefen (2005), who examines similar data sets from other languages and concludes that they exhibit “boundary-effects” as opposed to PU effects. (In the examples in 20c, the relevant boundary is the right edge of a syllable.)
15
Lombardi’s (1991, 1999) treatment of German relies on a markedness constraint stating that laryngeal features are licensed by a following tautosyllabic sonorant. The difference between my markedness constraint and Lombardi’s is not relevant for the following analysis.
Paradigm Uniformity Effects
245
(22) Constraints/rankings for canonical pattern of alternations in 23 a. *[–son, +vc]]: No voiced obstruent in coda position. b. I-O IDENT-[vc]: The specification for the feature [voice] of an input segment must be preserved in its output correspondent. c. *[–son, +vc]] » I-O-IDENT-[vc] The ranking in 22c, a specific version of 2a, is illustrated in the tableau in 23 for the word Dieb, representative of the data in 20. (23) Dieb [dip] (from /dib/) (illustrating ranking in 22c) a. b.
/dib/ [dip] [dib]
I-O-IDENT -[vc] *
* [–son, +vc]] *!
In this tableau, we can observe that the faithful candidate in 23b incurs a fatal violation of the high ranked markedness constraint and therefore 23a is selected. The tableau in 24 for the word Admiral is representative of examples like those in 21. The underlying form /admial/ is understood here to be the perceptual input from the donor language. (24) Admiral [at.mi.al] a. b. c.
/admial/ [ad.mi.al] [a.dmi.al] [at.mi.al]
OWF
*[–son, +vc]]
I-O-IDENT-[vc]
*! *! *
It is shown in this tableau that the ranking *[–son, +vc]] » I-O-IDENT-[vc] and OWF tell us that the voiced obstruent devoices. 3.2. Final Devoicing: PU Effects. With the analysis of the canonical phonology for voicing alternations in mind, we can now consider the words in 25a. These examples have a similar structure to the ones in 15a, the one significant difference being that in 25a I have only included examples in which the O is voiced. In
246
Hall
25a, the OS sequence is preceded by either a long vowel or a short vowel plus consonant. As noted above, the examples below have been taken from Drosdowski et al. 1990; they also form the basis of the various treatments proposed in the theoretical literature discussed below. The underlying form of the stems in 25c does not contain schwa, in conformance with the majority view in German phonology (Giegerich 1987, Wiese 1988, Hall 1992, Noske 1993, Wiese 1996, and Noske 1999). In section 4, I consider and reject Lombardi’s (1991) treatment of German, which relies on underlying forms with schwa for the first stem morpheme in 25a, for example, /zi dl/. There is no reason to distinguish inflectional from derivational suffixes in 25a, since the ALIGN-SH violations occur before both. (25) a. Siedl-ung Siedl-er Handl-ung handl-e Pendl-er Paddl-er Basl-er Ordn-ung ordn-en ein-ebn-en
[zi d.l] [zi d.l] [hand.l] [hand.l] [pnd.l] [pad.l] [ba z.l] [d.n] [d.nn] [an.e b.nn]
b. siedel-n [zi .dln] handel-n [han.dln] pendel-n [pn.dln] paddel-n [pa.dln] Basel [ba .zl] orden-t-lich [.dnt.lç] Ebene [e .b.n]
‘settlement’ ‘settler’ ‘act’ ‘act (1sg.)’ ‘commuter’ ‘paddler’ ‘one from Basel ‘order’ ‘put in order’ ‘level off’ ‘settle’ ‘act’ ‘commute’ ‘paddle’ ‘Basel’ ‘orderly’ ‘level’
c. /zi dl/ /handl/ /pndl/ /padl/ /ba zl/ /dn/ /e:bn/
An examination of the phonetic transcriptions in 25a reveals that the O is analyzed as syllable-final, meaning that the examples in 25a constitute systematic exceptions to Final Devoicing.16 My treatment of the OS 16
That the [d] in words like Handlung in 25a is phonetically voiced has been demonstrated by an acoustic experiment conducted by Jessen and Ringen (2002: 207ff.). All of the words in 25a, with the exception of [pad.l], also violate
Paradigm Uniformity Effects
247
sequences in 25a as heterosyllabic is occasionally found in the literature (for example, Rubach 1990), but as a whole the majority of researchers treat these sequences as tautosyllabic. I account for the problematic examples in 25a by analyzing them as a PU effect, that is, the reason the syllable-final obstruent in the stems in 25a is voiced is that there is a morphologically-related word in which the same obstruent is predicted by the canonical phonology to be voiced. My surface-based PU analysis of the data in 25 is not shared by the majority of researchers, who analyze the words in 25a in some other way, for example, in terms of cycles (Rubach 1990), a language-specific rule of resyllabification (Hall 1992), or by analyzing German as a language “not in harmony with the Law of Initials” (Vennemann 1972). While it is not my intention to discuss each of these alternative proposals, I make passing reference to the problems a non-PU analysis of the data in 25a poses. In section 4, I discuss two alternative non-PU approaches to the German data that were brought to my attention by an anonymous referee and show their drawbacks. That the non-application of Final Devoicing in 25a is attributed to PU derives support from two factors. First, all of these forms are heteromorphemic and have some morphologically related word containing the same stem in which the obstruent is predicted to be voiced by the canonical phonology (see 25b). Second, monomorphemes illustrating the same problem with respect to Final Devoicing are extremely rare (see below for discussion). Since PU effects by definition involve words that are related morphologically, then the existence of morphologically related forms and the absence of monomorphemes illustrating the same problem should come as no surprise. The observation that monomorphemes of the same phonological structure as 25a are very rare is a point that needs elaboration because it goes contrary to the assumption of many writers. The one monomorpheme commonly discussed in the literature, which is akin to the data in 25a, is presented in 26a. Assuming the heterosyllabic parse in this ALIGN-SH. For words like Siedlung, in which the stem contains a long vowel, the A LIGN-SH violations can be captured formally with the ranking in 19 above. For words like [d.n], the surface violation of A LIGN-SH follows from the ranking MAX-I-O » ALIGN-SH. I have no explanation for why there are so few examples like [pad.l].
248
Hall
word (which falls out from my analysis in section 2), we can observe here that the first syllable ends in a voiced obstruent. To my knowledge, this is the only true idiosyncratic exception to my analysis of Final Devoicing. Note that this word is also one of the rare idiosyncratic exceptions to the generalization in 4, which describes the distribution of superheavy syllables. There exists a sizable number of monomorphemic words in 26b that are unlike Adler because they are systematic exceptions to my analysis. The reason these words are systematic exceptions, as opposed to true exceptions, is that they all involve the same cluster, namely [gm] (and no other one). (26) Idiosyncratic and systematic exceptions to the present analysis a. Idiosyncratic exception Adler [a dl]
‘eagle’
b. Systematic exceptions (that is, only one cluster, namely [gm]) Magma [magma] ‘magma’ Segment [zgmnt] ‘segment’ Pigment [pgmnt] ‘pigment’ Sigma [zgma] ‘sigma’ Fragment [fagmnt] ‘fragment’ Stigma [stgma] ‘stigma’ Phlegma [flgma] ‘phlegm’ Dogma [dgma] ‘dogma’ Syntagma [zntagma] ‘syntagma’ Diaphragma [diafagma] ‘diaphragm’ Paradigma [paadgma] ‘paradigm’ Änigma [ngma] ‘enigma’ Pragmatik [pagmatk] ‘pragmatics’ Pygmäe [pgm ] ‘pygmy’ Stalagmit [talagmi t] ‘stalagmite’ Raffelsiefen (2004) points out that there are good reasons for considering words like Magma in 26b to be different from other OS sequences not in line with OWF (see also Kumashiro 2000, which comes to a similar
Paradigm Uniformity Effects
249
conclusion).17 First, and most obviously, the [g] in [gm] does not devoice, contrary to the canonical pattern in 21 for monomorphemic words like Admiral. Examples like Admiral are important because they show that Final Devoicing affects an obstruent before both [m] and [n] of the ON sequence if it is not a well-formed onset. Second, the vowel preceding [gm] onsets seems to be consistently short, whereas it is typically long before obstruent plus liquid sequences in monomorphemic words (recall 8a). While I can offer no explanation for the special behavior of [gm] clusters (other than the one proposed in Raffelsiefen 2004), it suffices to say that the correct treatment requires some mechanism not related to PU.18 The advantage of the present treatment is that the violations to Final Devoicing in 25a are predicted to occur exclusively in morphologically complex words. Likewise, the disadvantage of earlier approaches that analyze the OS sequences in 25a as tautosyllabic is that they miss the generalization that the overwhelming number of examples like these are 17
I summarize here Raffelsiefen’s treatment and another one she does not consider, but I leave open which of the two should be adopted in my own analysis. Raffelsiefen (2004) proposes that constraints are evaluated at two levels, that is, lexically and postlexically. The lexical level selects the heterosyllabic parse for words not in line with OWF, for example, [mag.ma], but postlexically these heterosyllabified [gm] clusters are tautosyllabified; that is, [ma.gma] is the correct output form. At the lexical level, OWF and constraints stating that syllables cannot end in a short vowel are undominated, but they lose their undominated status in the postlexical component. In a similar approach, one could argue—given my own analysis of syllabification in section 2—that [gm] is tautosyllabified in words like Magma (that is, [ma.gma]) by ranking the constraint *[gm lower than the structurally similar constraints that were grouped together under the cover term OWF. This analysis would then see the lack of word-initial [gm] as accidental. Note that this approach works technically, but it cannot explain why [gm] is the only OS cluster not occurring word-initially, that is tautosyllabified.
18
Several authors also include proper names in their discussion of German syllabification, for example, Leibniz [labnts]. See Giegerich 1992, Hall 1992, and Yu 1992. Recall from note 5 that I do not consider proper names in my analysis because they are not representative of canonical phonological patterns. From a formal point of view one might want to say that proper names have their own constraint ranking. This is an issue I leave open.
250
Hall
heteromorphemic, and that the tautomorphemic words like Adler in 26a are extremely rare, not to mention the fact that [gm] deserves special treatment (Vennemann 1972, Hall 1992, Yu 1992, Giegerich 1992, and Noske 1999).19 Consider now the three paradigms in 27 for the pair {Siedlung, siedeln}, which is representative of all of the examples in 25a. In the Standard German dialect described here, A is correct and B and C are not. In the Northern dialects discussed in Vennemann 1972 and Brockhaus 1995:40, Final Devoicing affects the O, so that B is correct and A and C are not. I consider first Standard German and then I return to the Northern dialects. (27) Three paradigms for the pair {Siedlung, siedeln}: A is correct A zid.l zi.dln
B zit.l zi.dln
C zi.dl zi.dln
Paradigms for {Siedlung, siedeln} with shortened vowels as in 17 (for example, {[zd.l], [zi.dln]}) will not be dealt with here because the analysis posited in section 2.3 always treats the paradigms in A and B in 27 as more harmonic than paradigms with shortened vowels. Another candidate I do not consider here is one in which the stem is identical in both forms, that is, {zi.dl, zi.dln}. This paradigm is clearly more uniform than the correct one in A, but it loses out to the latter one because the first word in it (that is, ['zi.d.l]) does not satisfy the metrical constraint guaranteeing that the foot is trochaic (see Féry 1998). By contrast, both members of the correct paradigm in A above satisfy the constraint TROCHEE. Crucial for my treatment is the O-O constraint O-O-IDENT-[vc] in 28a, which says that the value of the feature [voice] is the same for all of the members in a paradigm. Recall from above that some linguists have argued that [spread glottis] and not [voice] is distinctive for German 19
The criticism that the “exceptions” to Final Devoicing involve heteromorphemic examples does not apply to Rubach 1990, in which the cyclic treatment correctly predicts this to be the case. In contrast to Rubach 1990, my analysis requires no cycles.
Paradigm Uniformity Effects
251
obstruents. If this is the case, then the O-O constraint in 28a would refer to [spread glottis]. (28) a. O-O-IDENT-[vc]: The value of [voice] for an obstruent in “stem A” is identical with the value of the feature [voice] of the corresponding obstruent in derived words containing “stem A.” b. O-O-IDENT-[vc], OWF » *[–son, +vc] ] » I-O-IDENT-[vc] (ranking to be modified below) The ranking of O-O-IDENT-[vc] over the markedness constraint *[–son, +vc]] in 28b is a specific version of the general ranking in 2b for PU effects. This is the ranking discussed below, although later on in this section I demonstrate that it needs to be modified slightly. An examination of the three paradigms in 27 reveals that paradigm A satisfies O-O-IDENT-[vc] because the [d] is present in both examples, but that in paradigm B, O-O-IDENT-[vc] is violated because the [t] in the first word corresponds to a [d] in the second word. Paradigm C is not optimal because its first member (that is, [zi.dl]) violates OWF. The analysis is illustrated in the tableau in 29. (29) Evaluation for the paradigm {Siedlung, siedeln} /zidl-/ a. b. c.
O-O-IDENT-[vc]
OWF
A B C
*[–son, +vc]]
I-O-IDENT-[vc]
* *! *!
* *
The treatment of PU effects in this tableau is revised in 34 below. Some data from the Northern dialects mentioned above have been presented in 30. Observe that the syllable-final obstruent devoices in 30a (Vennemann 1972).
252
Hall
(30) Syllable final obstruent devoicing in Northern German dialects a. Siedl-ung Handl-ung
[zit.l] [hant.l]
‘settlement’ ‘act’
b. siedel-n handel-n
[zi.dln] ‘settle’ [han.dln] ‘act’
c. /zidl/ /handl/
The ranking for the Northern dialects is illustrated in the tableau in 31. (31) Evaluation for the paradigm {Siedlung, siedeln} in N. German /zidl-/ a. A b. B c. C
*[–son, +vc]]
OWF
O-O-IDENT-[vc]
*!
I-O-IDENT-[vc]
* * *!
This tableau shows that the B candidate can be selected in the Northern dialects by demoting O-O-IDENT -[vc], so that it is lower in the hierarchy than the markedness constraint *[–son, +vc]], which bans voiced obstruents in the coda. Put differently, in Standard German there are PU effects involving Final Devoicing, but in the Northern dialects it is the canonical phonology that wins and reference to PU is therefore not necessary.20 Consider once again the PU analysis for Standard German in tableau 29. Although the ranking in that tableau correctly captures the fact that the syllable-final obstruent in the words in 25a does not devoice, it is problematic because it incorrectly predicts that the alternations in 20a,c should show PU effects as well. This problem is made clear in the following tableau for the paradigm {Dieb, Diebe} ‘thief, thieves’ from 20a. I consider here three paradigms, namely the correct (but nonuniform) one, {dip, di.b}, as well as two that seem to be more harmonic than the correct one because they exhibit consistent PU effects with respect to the feature [voice]: {dib, di.b} and {dip, di.p}. In tableau 32, the intended winner in 32a incurs a fatal violation of the PU constraint O-O-IDENT -[vc] (since the [p] in [dip] differs from the 20
Note that the heterosyllabic parse of the OS in a word like Siedlung for candidate B (that is, [zit.l]) falls out from the ranking OWF » NOCODA » NOCOMPONS for the canonical phonology in 12–13.
Paradigm Uniformity Effects
253
[b] in [di.b]). The markedness constraint *[–son, +vc]] is responsible for ruling out the 32b form, so that the unintended winner in 32c is incorrectly selected as optimal. /dib/
(32) a. b. c.
dip d.b dib dib dip di.p
O-O-IDENT-[vc]
*[–son, +vc]]
*!
I-O-IDENT-[vc]
* *! **
Consider once again the phonetic representation of the data in 25a. These items all have a phonological similarity in the sense that the first syllable ends in an obstruent and the second one begins with a sonorant consonant. By contrast, none of the examples in 20 are of this structure. Thus, from the point of view of markedness all of the forms in 25a violate the SYLLABLE CONTACT LAW (SCL) in 33a.21 Raffelsiefen (1995) argues that the SCL plays an important role in her treatment of German schwa. Since there are many SCL violations in German (for example, words like Atlas [at.las] in 10), we can conclude that SCL must be dominated by faithfulness constraints (for example, MAX-I-O and DEP-I-O). This ranking MAX-I-O, DEP -I-O » SCL is necessary because the SCL violation in words like Atlas is not repaired in some way (for example, [a.t.las] or [a.tas]), and instead these words surface faithfully. It cannot be determined how SCL is ranked with respect to the other constraints discussed above. (33) a. SYLLABLE CONTACT LAW (SCL): In . the sonority of is greater than the sonority of . b. {O-O-IDENT-[vc] & SCL}, OWF » * [–son, +vc]] » I-O-IDENT-[vc]
21
For pre-OT work on the SCL, see Murray and Vennemann 1983 and Vennemann 1988. The status of the SCL in the OT framework is discussed in Bat El 1996, Green 1997, Davis 1998, Davis and Shin 1999, and Green 2003.
254
Hall
In the revised analysis below, I argue that the examples in 25a (and not the ones in 20a,c) can be analyzed as PU effects if the PU constraint, that is, O-O-IDENT-[vc], is locally conjoined with SCL and if this local conjunction dominates the markedness constraint *[–son, +vc]], as in 33b.22 Consider now the following tableau for the paradigm {Siedlung, siedeln}, in which the ranking in 33b is illustrated. (34) PU effects for 25a given the ranking in 33b a. b. c.
/zidl/ zid.l zi.dln zit.l zi.dln zi.dl zi.dln
O-O-IDENT-[vc] & SCL
OWF
*!
* [–son, +vc]]
I-O-IDENT-[vc]
* *!
Paradigm 34c cannot be optimal because its first member does not satisfy OWF. Paradigm 34b is not in line with the constraint conjunction because this pair of words violates both O-O-IDENT -[vc] (by virtue of the fact that there is a [t] in the first word in the paradigm but a [d] in the other one) and the SCL (that is, in [zit.l]). By contrast, paradigm 34a satisfies the conjunction, since this pair of words does not violate the O-O constraint, O-O-IDENT-[vc]. The tableau in 35 is an evaluation of the paradigm {Dieb, Diebe}. Three paradigms have been included in the tableau; the one with the nonuniform paradigm in 35a is correct, and the ones with uniform paradigms in 35b,c are not.
22
For literature on local conjunction in OT, see ubowicz 2002, Alderete 2004, and Kirchner 2004. See also McCarthy 2003 for a useful survey of the literature, and Hall 2005, which shows that the SCL is locally conjoined with an onsetspecific constraint to account for the fact that the regular process of Glide Formation in German is systematically blocked in certain phonologically defined contexts.
Paradigm Uniformity Effects
255
(35) No PU effects for 20 given the ranking in 33b a. b. c.
/dib/ dip di.b dib di.b dip di.p
O-O-IDENT-[vc] & SCL
*[–son, +vc]]
I-O-IDENT-[vc]
* *! **!
An examination of the three paradigms in 35 reveals that none of them violates the constraint conjunction {O-O-IDENT-[vc] & SCL}, because all six words are in line with the SCL. What this means is that the evaluation of the paradigms in 35 is put off until the second round, in which 35b incurs a fatal violation of the markedness constraint *[–son, +vc]]. That 35a is correctly selected over 35c can only be ascertained in the final column, in which it is shown that the latter paradigm incurs two violations of I-O-IDENT -[vc] and 35a only one.23 Note that the modified analysis for Standard German in 34–35 with the constraint conjunction does not require a reanalysis for Northern German (recall the data in 30 and the analysis in 31). Thus, to obtain the correct output forms (that is, 34b) the constraint conjunction must be lower ranked than *[–son, +vc]]. However, this reranking does not affect the outcome in 35 so that 35a is selected for Northern German as well. One question discussed in the literature on constraint conjunctions is what kinds of constraints can and cannot be locally conjoined (see, for example, ubowicz 2002 and McCarthy 2003 and references cited therein). Note that my proposed conjunction {O-O-IDENT-[vc] & SCL} conjoins an O-O constraint with a markedness constraint. To my know23
The analysis in 34 holds for words of the 25a type, that is, words like Siedl-ung containing prevocalic OS, in which the O is voiced and in which OS is not a permissible onset. Note that the same analysis is predicted for words like wappnen [vap.nn] ‘to arm’, in which the O is voiceless (see Wappen [vapn] ‘coat of arms’, /vapn/). Thus, the ranking in 34 predicts the heterosyllabic parse in [vap.nn] because the candidate with a tautosyllabic parse (that is, [va.pnn]) is not in line with OWF.
256
Hall
ledge none of the constraint conjunctions proposed in the literature involves constraints from these two families. I consider this gap to be purely accidental and attribute it to the general lack of studies dealing with PU effects in natural languages. It should be noted that constraints from the markedness family and the O-O constraint family have in common that they both refer to the output forms on the surface (as opposed to the input). I conclude that the burden of proof now lies on the shoulders of linguists who contend that an analysis of German is possible without my proposed conjunction.24 I conclude this section by considering once again the morphologically complex words like nebl-ig presented earlier in 9a (repeated in 36a), since my implicit claim is that these examples, contrary to the structurally similar ones in 25a, do not show PU effects. An examination of the phonetic forms in 9a reveals that Final Devoicing does not set in, but in contrast to words like Siedl-ung in 25a, the syllable boundary is assumed to precede the OS. Note that each of the stems in 36a occurs without a suffix in 36b and that the intervocalic O is predicted by the canonical phonology to be voiced.
24
It is interesting to note that one of the constraints in the proposed conjunction, namely the SCL, has been argued to be present in another conjunction to account for the fact that Glide Formation is blocked in one context (see Hall 2005). One could speculate that constraint conjunctions between constraints belonging to different families are only allowed if one or more of these constraints is present in another conjunction in the same language. Concerning the OT literature on constraint conjunctions, McCarthy (2003:26) criticizes ubowicz’s (2002) proposal that markedness constraints should be conjoined with I-O faithfulness constraints in order to account for non-derived environment blocking effects. The primary reason why he criticizes this type of constraint conjunction is that it does not capture Kiparsky’s (1973) original notion of a “derived environment.” Note that McCarthy’s criticisms cannot be extended to the conjunction in 34, which involves a markedness constraint and an O-O faithfulness constraint, because (a) this conjunction involves an O-O constraint, and (b) it is not intended to capture a derived environment. See Hall “in press” for criticisms of McCarthy’s (2003) theory of comparative markedness.
Paradigm Uniformity Effects
257
(36) VOS-V is parsed V.OS-V if and only if OS is a permissible onset a. nebl-ig fiebr-ig regn-en
[ne.blç] [fi.bç] [e.gnn]
‘foggy’ ‘feverish’ ‘rain’
b. Nebel Fieber Regen
[ne.bl] [fi.b] [e.gn]
‘fog’ ‘fever’ ‘rain’
One might argue that there is no difference between 36a and 25a, and that the latter ones are therefore syllabified like the former ones (for example, [neb.lç]), implying that both sets of words are captured in terms of PU effects. (See Kumashiro 2000:367–368, who proposes a PU account for words like the ones in 36a, which is cast in a very different theoretical framework.) While the alternative analysis of the nebl-ig words in 36a in terms of PU might work technically given the ranking in 35, I assume that this is not the case and they instead indicate the canonical pattern because of the following asymmetry: For VOSV in which OS is a well-formed onset there are many examples of (a) monomorphemes (see 8a) and (b) heteromorphemic examples (see 36a). However, for VOSV in which OS is not a well-formed onset there are many heteromorphemic examples (see 25a), but not monomorphemes (see 26a). This asymmetry makes sense if words like Siedlung in 25a are analyzed in terms of PU but not examples like neblig in 36a.25
25
Several writers have pointed out that the medial obstruent in words like Siedlung in 25a (as well as ones like neblig in 36a) can optionally undergo Final Devoicing in Standard German in more colloquial registers (Wiese 1988, Hall 1992, Yu 1992, Brockhaus 1995:186, and Wiese 1996). I analyze this free variation for words like neblig as a permutation of the ranking of NOCOMPLEXONSET and NOCODA. Thus, in 12–13 the ranking of NOCODA over NOCOMPLEXONSET holds for the style of speech captured in the citation forms in the pronouncing dictionaries (for example, Drosdowski et al. 1990), but the reverse ranking describes a more colloquial speech register. For words like Siedlung, the devoiced /d/ implies that in colloquial speech the highest two constraints are switched in the ranking.
258
Hall
4. Alternative Analyses. An anonymous reviewer suggests two alternative analyses for German Final Devoicing that do not involve O-O constraints. In this section I summarize these alternatives and show that they suffer serious drawbacks that do not hold for the present analysis. According to the first approach, Final Devoicing operates syllablefinally, but must be limited to morphologically composite environments, that is, the segment that devoices must be syllable-final and morphemefinal. Thus, on this view the process sets in word-finally (as in Dieb [dip] ‘thief’, see 20a), and morpheme-finally followed by a consonantinitial suffix (as in Bünd-nis [bnt.ns] ‘alliance’, see 20b). By contrast, Final Devoicing does not set in before vowel-initial suffixes. In a word like Dieb-e [di.b] ‘thieves’ (see 20a), the [b] stays voiced even though it is morpheme-final because it is syllable-initial. In an example like nebl-ig [ne.blç] ‘foggy’ (see 26a), the syllable-initial [b] remains voiced because it is not final in a morpheme. The explanation for the lack of devoicing in words like Handl-ung [hand.l] ‘act’ (see 25a), is similar: Final Devoicing does not set in because the [d] is not morpheme-final. According to this reviewer’s suggestion even the [d] in Adler [ad.l] in 26a is correctly predicted not to undergo Final Devoicing, because it is not final in a morpheme. Thus, from this perspective, the alternative just described is superior to the one I propose because it has no exceptions.26 This alternative analysis suffers from two weaknesses. First, and most seriously, it cannot explain why the syllable-final obstruent in words like Admiral [at.mi.al] (recall 21) is devoiced when these words enter German. (The reason the syllable-final /d/ in the donor language is expected to surface in German as [d] is that this sound is not morphemefinal). Although the literature on German Final Devoicing has traditionally concentrated exclusively on words that alternate (e.g. Die[p] vs. Die[b]e), a truly complete analysis needs to take monomorphemic loan words like Admiral into consideration as well. The second drawback with the alternative sketched above is that it does not explain why the “exceptional” words like Handlung in 26a are heteromorphemic. Thus, the reviewer’s suggested analysis does not capture the true 26
According to the reviewer a constituent-final environment counts as a “derived environment.” For incorporation of the Derived Environment Con-straint into OT, see Kim 2002 and Iverson 2004.
Paradigm Uniformity Effects
259
generalization: The exceptional forms are bimorphemic and therefore show PU effects. According to the second alternative (proposed in Lombardi 1991:67– 68 and surveyed in Iverson 1997), the absence of expected Final Devoicing in one form correlates with the presence of an additional vowel in a related form. Thus, in the case of Sied.lung, where Final Devoicing does not take place, the related word is sie.deln, with schwa preceding [l]. Lombardi suggested that the underlying form of the stem contains an “empty” vowel, which is filled in sie.deln (or just manifested as syllabicity in the [l]) but remains empty in Sied.lung, that is, Sie.dV.lung. This places the [d] structurally into an onset at an abstract stage in the derivation and so handles the paradigm relatedness matter rather directly. That is, the underlying stem /zidVl/ is realized as [zidl] in Siedlung, but as [zidl] in siedeln. On this view, the claim is that the [d] in Siedlung escapes Final Devoicing because it is prosodically syllable-initial at an abstract stage, as it is in related siedeln, and hence is not a candidate for neutralization with [t]. The problem with this approach is that it requires stems ending in two consonants that rise in sonority from left to right, such as /zidVl/, to have an underlying “schwa” (that is, an empty V) that is subject to deletion in words like Handlung. As I noted above in section 2.3, there is consensus that schwa in stems of this structure is epenthetic (see Giegerich 1987, Wiese 1988, Hall 1992, Noske 1993, Wiese 1996, and Noske 1999). The reason these authors (and others) see stems like siedlas being schwa-less (i.e. /zidl/) is that the schwa is predictable as a function of syllable structure. Put differently, schwa is epenthesized in stems like /zidl/ (or alternatively the /l/ is made syllabic) to make the stem pronounceable. To argue that the underlying forms of all “exceptional” forms contain an underlying schwa accompanied with a derivation in which schwa deletion is ordered before Final Devoicing, misses this generalization. 5. Summary and Conclusion. In this article, I have shown that German has systematic counterexamples to two generalizations for the canonical phonology. First, there exist words containing superheavy syllables of the form [VC] within a phonological word, although such syllables are normally restricted to word-final position. Second, there are many words containing syllable-
260
Hall
final obstruents that do not devoice, contrary to the canonical pattern of Final Devoicing in the syllable coda. It was argued above that both problems are examples of PU effects that can be captured formally by positing that two specific O-O constraints dominate the markedness constraints responsible for the canonical patterns. One of these O-O constraints refers to the moraic structure of vowels and the other to the feature [voice]. The latter constraint was argued to be locally conjoined with the markedness constraint SYLLABLE CONTACT LAW. The present analysis is very different from earlier treatments on German phonology, which make no reference to PU. There are at least two theoretical questions my analysis raises. One is whether or not there are constraints on the kinds of features that can be referred to in an O-O constraint. My preliminary answer is negative; one could easily imagine some language in which specific O-O constraints refer to distinctive features other than [voice] or to the moraic structure of consonants (as opposed to the moraic structure of vowels). If languages are not attested with such specific O-O constraints, then I would argue that this is simply an accident due to the relative paucity of PU-style analyses for other languages. A second question is why German would choose to make paradigms uniform along certain dimensions (that is, the moraic structure of vowels, the feature [voice]), but not along others. One very obvious example of a segmental phenomenon typically not subject to the kind of PU analysis proposed above involves the umlauted vowels in the German nominal and verbal systems. For example, in the singular, indicative conjugation of the verb tragen ‘carry’, we have trag-e (1sg.), träg-st (2sg.), träg-t (3sg.), trag-en (1pl.), trag-t (2pl.), trag-en (3pl.), but this is a nonuniform paradigm due to the allomorphs trag- and träg-. One could easily rank the relevant O-O constraint (O-O-IDENT -[back]) lower down in the constraint hierarchy from a formal point of view, but this approach begs the question of why certain O-O constraints are highly ranked whereas others are not. Although there might not be a purely formal difference between features like [voice] versus [back], one needs to consider the morphological function of these features as well. Many linguists (for example, Wurzel 1984) have pointed out that umlaut in examples such as the one mentioned above has a morphological (as opposed to a phonological function). For example, in the form träg-t, the morphological categories “third person, singular, indicative” are
Paradigm Uniformity Effects
261
expressed not only with the suffix -t, but also with the change in backness of the stem vowel. By contrast, the feature [voice] has a purely phonological function in German. What this discussion therefore suggests is that the only kinds of features that cannot be present in a high-ranking O-O constraint are those features that have some kind of morphological function. By contrast, features that have a purely phonological function are the only ones that can potentially be present in a high-ranking O-O constraint. Whether or not this idea can be shown to be correct for other dialects of German or for other languages is a question I leave open for further study. REFERENCES Alderete, John. 2004. Dissimilation as local conjunction. Optimality theory in phonology. A reader, edited by John J. McCarthy, 394–406. Oxford: Basil Blackwell. [Excerpt from Alderete 1997. Dissimilation as local conjunction. North Eastern Linguistic Society (NELS) 27.17–32.] Bat El, Outi. 1996. Selecting the best of the worst: The grammar of Hebrew blends. Phonology 13.283–328. Benua, Laura. 1997. Transderivational identity: Phonological relations between words. Doctoral dissertation, University of Massachusetts, Amherst. Booij, Geert. 1985. Coordination reduction in complex words: A case for prosodic phonology. Advances in nonlinear phonology, ed. by Harry van der Hulst and Norval Smith, 143–160. Dordrecht: Foris. Brockhaus, Wiebke J. 1995. Final devoicing in the phonology of German. Tübingen: Max Niemeyer. Davis, Stuart. 1998. Syllable contact in optimality theory. Korean Journal of Linguistics 23.181–211. Davis, Stuart, and Seung-Hoon Shin. 1999. The syllable contact constraint in Korean: An optimality-theoretic analysis. Journal of East Asian Linguistics 8.285–312. Downing, Laura, Tracy Alan Hall, and Renate Raffelsiefen (eds.). 2005. Paradigms in phonological theory. Oxford: Oxford University Press. Drosdowski, Günther [with Peter Eisenberg]. 1990. Duden. Das Aussprachewörterbuch. Wörterbuch der deutschen Standardaussprache. Mannheim: Dudenverlag. Féry, Caroline. 1998. German word stress in optimality theory. Journal of Comparative Germanic Linguistics 2.101–142.
262
Hall
Giegerich, Heinz J. 1987. Zur Schwa-Epenthese im Standarddeutschen. Linguistische Berichte 112.449–469. Giegerich, Heinz J. 1992. Onset maximization in German: The case against resyllabification rules. Silbenphonologie des Deutschen, ed. by Peter Eisenberg, Karl Heinz Ramers, and Heinz Vater, 134–171. Tübingen: Gunter Narr. Green, Antony Dubach. 1997. The prosodic structure of Irish, Scots Gaelic, and Manx. Doctoral dissertation, Cornell University. Green, Antony Dubach. 2003. Extrasyllabic consonants and onset wellformedness. The syllable in optimality theory, ed. by Caroline Féry and Ruben van de Vijver, 238–253. Cambridge: Cambridge University Press. Hall, Tracy Alan. 1992. Syllable structure and syllable related processes in German. Tübingen: Max Niemeyer. Hall, Tracy Alan. 2002a. Against extrasyllabic consonants in German and English. Phonology 19.33–75. Hall, Tracy Alan. 2002b. The distribution of superheavy syllables in Standard German. The Linguistic Review 19.377–420. Hall, Tracy Alan. 2005. German glide formation and its theoretical implications. Manuscript, Department of Germanic Studies, Indiana University. Hall, Tracy Alan. In press. Derived environment blocking effects in optimality theory. Natural Language and Linguistic Theory. Hall, Tracy Alan, and Silke Hamann. 2003. Loanword nativization in German. Zeitschrift für Sprachwissenschaft 22.1.56–85. Ham, William H. 1998. A new approach to an old problem: Gemination and constraint reranking in West Germanic. Journal of Comparative Germanic Linguistics 1.225–262. Iverson, Gregory K. 1997. Final devoicing in the phonology of German by Wiebke J. Brockhaus (review article). American Journal of Germanic Languages and Literatures 9.255–264. Iverson, Gregory K. 2004. Deriving the derived environment constraint in nonderivational phonology. Studies in Phonetics, Phonology and Morphology 11.1–33. (Special issue. Optimality Theory: Applications and Challenges). Iverson, Gregory K., and Joseph Salmons. 1995. Aspiration and laryngeal representation in Germanic. Phonology 12.369–396. Jessen, Michael, and Catherine Ringen. 2002. Laryngeal features in German. Phonology 19.189–218. Kim, Jin-Hyung. 2002. Lexicon optimization reconsidered. Language Research 38.31–50. Kiparsky, Paul. 1973. Phonological representations. Three dimensions of linguistic theory, ed. by Osamu Fujimura, 3–136. Tokyo: TEC.
Paradigm Uniformity Effects
263
Kirchner, Robert. 2004. Synchronic chain shifts in optimality theory. Optimality theory in phonology. A reader, ed. by John J. McCarthy, 407–416. Oxford: Basil Blackwell. [Excerpt from Kirchner 1996. Synchronic chain shifts in Optimality Theory. Linguisic Inquiry 27.341–350.] Kumashiro, Fumiko. 2000. Loan word phonology: A case for a non-reductionist approach to grammar. Berkeley Linguistics Society 25.359–370. Lombardi, Linda. 1991. Laryngeal features and laryngeal neutralization. Doctoral dissertation, University of Massachusetts, Amherst. Lombardi, Linda. 1999. Positional faithfulness and voicing assimilation in optimality theory. Natural Language and Linguistic Theory 17.267–302. ubowicz, Ania. 2002. Derived environment effects in optimality theory. Lingua 112.243–280. McCarthy, John J. 2003. Comparative markedness. Theoretical Linguistics 29.1–51. Murray, Robert W., and Theo Vennemann. 1983. Sound change and syllable structure in Germanic phonology. Language 59.514–528. Noske, Manuela. 1999. Deriving cyclicity: Syllabification and final devoicing in German. The Linguistic Review 16.227–252. Noske, Roland. 1993. A theory of syllabification and segmental alternation. With studies on the phonology of French, German, Tonkawa, and Yawelmani. Tübingen: Max Niemeyer. Prince, Alan, and Paul Smolensky. 1993. Optimality theory. Manuscript, Rutgers University and University of Colorado. Raffelsiefen, Renate. 1995. Conditions for stability: The case of schwa in German. Manuscript, Heinrich-Heine Universität, Düsseldorf. Raffelsiefen, Renate. 2004. Phonological effects in word formation. Habilitationsschrift, Freie Universität, Berlin. Raffelsiefen, Renate. 2005. Paradigm uniformity effects versus boundary effects. Downing, Hall, and Raffelsiefen 2005, 210–262. Rubach, Jerzy. 1990. Final devoicing and cyclic syllabification in German. Linguistic Inquiry 21.79–94. Steriade, Donca. 1997. Phonetics in phonology: The case of laryngeal neutralization. Manuscript, University of California, Los Angeles. Vennemann, Theo. 1972. On the theory of syllabic phonology. Linguistische Berichte 18.1–18. Vennemann, Theo. 1988. Preference laws for syllable structure and the explanation of sound change. Berlin: Mouton de Gruyter. Wiese, Richard. 1988. Silbische und Lexikalische Phonologie: Studien zum Chinesischen und Deutschen. Tübingen: Max Niemeyer. Wiese, Richard. 1996. The phonology of German. Oxford: Clarendon Press. Wurzel, Wolfgang U. 1984. Flexionsmorphologie und Natürlichkeit. Berlin: Akademie Verlag.
264
Hall
Yu, Si-Taek. 1992. Silbeninitiale Cluster und Silbifizierung im Deutschen. Silbenphonologie des Deutschen, ed. by Peter Eisenberg, Karl Heinz Ramers, and Heinz Vater, 172–207. Tübingen: Gunter Narr.
Tracy Alan Hall Department of Germanic Studies Ballantine Hall 644 Indiana University 1020 Kirkwood Avenue Bloomington, Indiana 47405-7103 USA [
[email protected]]
Journal of Germanic Linguistics 17.4 (2005):265–277
The “Conundrum” of Old Norse i-Umlaut: A Reply to Iverson and Salmons Joseph Voyles University of Washington Both Iverson and Salmons 2004 and Voyles 1982 develop viable theories of the Old Norse i-umlaut. It is argued, however, that while Iverson and Salmons’ theory posits implausible changes, the changes assumed by Voyles are plausible and well attested.
1. Introduction. The crux of the Old Norse (ON) i-umlaut problem resides in the masculine i-class nouns. While the long-stem nouns, such as Old Icelandic (OI) gestr ‘guest’ (< +gast+iR), generally show i-umlaut, the short-stem nouns such as OI star ‘place’ (< +sta+iR ) generally do not. In what follows, I first summarize and critique the Iverson and Salmons 2004 account (the I&S theory) of these developments. I then present an alternative account (first proposed in Voyles 1982, hereafter the V theory), and discuss Iverson and Salmons’ critique of this theory. 2. Iverson and Salmons’ Theory and Its Evaluation. 2.1. Summary. Under the I&S theory, three stages of ON i-umlaut are posited. In the first stage, both i-umlaut and vowel syncope are active in Old Norse. The former fronts vowels preceding a short or long /i/ or /j/ in the following syllable, and the latter deletes a mora of length after a long syllable. Below in 1 are the derivations for the nominative singular and plural forms of a short- and a long-stem noun: (1)
Short Stem
Long Stem
/sta+iR/ /sta+iiR/ /gast+iR/ /gast+iiR/ ste+iR ste+iiR gest+iR gest+iiR [steiR] [steiiR] [gestR] [gestiR]
‘place’, ‘guest’ i-umlaut Syncope, long stems
© Society for Germanic Linguistics
266
Voyles
According to Iverson and Salmons (2004:93), the second stage proceeds as follows: With umlaut still alive in the phonology (that is to say, vowels were still fronted when a following i was phonetically present, as in gestiR NOM.PL.), we surmise that a new generation of learners, seeking common phonetic ground with the previous generation, inferred that because umlaut could not be predicted from the surface environment in long-stem words like [gestR] NOM.SG., the most coherent account was simply to posit vowels as already front in their underlying representations. That is, the underlying forms /gast+iR/ and /gast+iiR/ were—on the basis of the occurrence of the surface form [gestR]—restructured to /gest+iR/ and /gest+iiR/, respectively. Hence, at this stage, the derivation was as in 2. (2) /sta+iR/ /sta+iiR/ ste+iR ste+iiR [steiR] [steiiR]
/gest+iR/ /gest+iiR/ gest+iR gest+iiR [gestR] [gestiR]
i-umlaut Syncope (long stem)
Finally, in the third stage, i-umlaut “died,” that is, it was no longer phonologically conditioned.1 In addition, syncope at this point expands its environment to apply not only after long stems, but also after short stems. The relevant derivations of the nominative singular and plural forms are: (3) /sta+iR/ /sta+iiR/ [staR] [staiR]
/gest+iR/ /gest+iiR/ (Loss of i-umlaut) [gestR] [gestiR] Syncope (short/long)
According to this account, the originally umlauted forms such as [steiR] (for which there is absolutely no graphemic evidence) all now revert to
1
As a parallel type of development, Iverson and Salmons (2004:101) cite Dresher’s (2000) analysis of Middle English open syllable lengthening, which, like the purported ON i-umlaut case, resulted in an “opaque” derivation. I reject this comparison in section 2.2.
Old Norse Umlaut
267
[staR]. In an attempt to cite a parallel case of reversion, Iverson and Salmons (2004:97–8) state: Interestingly, a close parallel to this specific reversion exists elsewhere in Germanic: The Old High German Rückumlaut verbs must once have evidenced umlaut, which came to be lost by analogy, as argued persuasively by Robinson (1980). For this class of forms, with umlautless preterite optatives like branti ‘would have burned’ (compare brennen ‘burn’), Robinson has laid out a solid structural case for reversion from previously umlauted forms, namely, avoidance of ambiguity in the key verbal categories of present vs. preterite optative. In fact, as I argue in 2.2, the view that the Old High German (OHG) case can provide a parallel to the purported ON development rests on a misinterpretation of the OHG data, which leaves the Iverson and Salmons analysis based on a type of wholesale analogical change otherwise unattested. 2.2. Critique. The belief that OHG umlaut originally applied across the board to all forms is an article of faith rather than one of empirical evidence. In fact, as argued in some detail in Voyles 1991, there is absolutely no evidence or reason to believe that the OHG i-umlaut rule was ever exclusively phonologically conditioned. As Iverson and Salmons (2004:97–98) note, if OHG umlaut had indeed applied to the past subjunctive forms of verbs with stem-final dentals, such verbs would have become practically homophonous with the present indicative forms; for example, in (the admittedly few) cases such as the infinitival forms senten ‘send’ and resten ‘rest’, umlaut would have resulted in the past plural subjunctive forms *sentn or *restn (instead of the attested and non-umlauted sentn and rastn).2 Hence there was a semantic motivation for the avoidance of OHG umlaut in these forms, even at its earliest inception.3
Note that + is used for reconstructed forms, and * for incorrect or unexpected forms. 2
3
See Voyles 1991 for further discussion.
268
Voyles
It should also be added that, contrary to the assumption underlying the I&S theory, many—if not most—phonological changes are at least in part morphosyntactically conditioned, even from their earliest time of occurrence (see Voyles 1991). In fact, numerous instances of partial morphosyntactic conditioning in newly added phonological rules are attested. For example, in the case of African American Vernacular English (AAVE), Labov (1972:190) records the simplification of word-final clusters through the deletion of the final segment.4 However, deletion applies less often (and for some speakers not at all) if the word-final consonant cluster contains the past marker /+d/ or /+t/; for example, reduced bol’ ‘bold’, pas’ ‘past’, but unreduced rolled /rol+d/ and passed /pæs+t/. The reluctance of AAVE speakers to delete the final consonant in the past of weak verbs was, just as the absence of umlaut in the OHG forms discussed above, motivated by semantic considerations. Past forms, such as the non-occurring *roll’ for rolled and *pass’ for passed would have been homophonous with the present tense forms. I conclude: a) that this kind of semantic ambiguity could not possibly have motivated the absence of i-umlaut in the short-stem ON nouns (and Iverson and Salmons do not posit such), and b) that the wholesale nonumlaut analogy in ON—whether by restructuring or reversion—as posited by Iverson and Salmons does not have a precedent in the OHG case, nor does it, as far as I can tell have a precedent anywhere else. I now consider the purported parallel provided by Iverson and Salmons to the change whereby the ON underlying singular form /gast+iR/ and the plural form /gast+iiR/ were restructured to /gest+iR/ and /gest+iiR/, respectively. This change was triggered by the surface form [gestR], in which the occurrence of umlaut could not be predicted from the surface forms of long-stem words. In this sense, then, the derivations of the form had become opaque. The purported analog Iverson and Salmons cite is Dresher’s (2000) analysis of Middle English (ME) open syllable lengthening. Since the ME case is intended as a crucial parallel to the I&S interpretation of the ON change, I turn to Dresher’s (2000) account in some detail. In this interpretation, there were two relevant Middle English phonological changes: open syllable lengthening (OSL) and trisyllabic shortening (TSS) (presumably occurring in that order), which 4
For detailed discussion, see Labov 1994, especially chapter 20.
Old Norse Umlaut
269
are reflected in alternations of nominative singular and plural forms, such as those found in 4. (4) Original Short Vowel Forms a. tale, tales ‘tale, tales’ b. god, godes ‘god, gods’ c. hol, holes ‘hole, holes’ d. bever, bev eres ‘beaver, beavers’ Original Long Vowel Forms e. devel, dev eles ‘devil, devils’
(OSL, singular and plural) (OSL, plural) (OSL, plural) (OSL, singular; TSS, plural) (TSS, plural)
At the next stage, the vowels in inflectional endings are syncopated, yielding the forms in 5. (5) Syncopated Forms a. tal , tals b. god, gods c. hol, hols d. bever, bevers e. devel, devels Dresher (2000:58–59) comments on this situation as follows: The loss of the inflectional vowel leaves the vowel length situation in a confused state. In some words, there is lengthening in the plural, but no change in syllable structure, while in other words, but not all, the plural is associated with shortening. There are no prospects for salvaging a phonological rule from this situation. But even a morphological rule appears to be unavailable: we cannot, for example, associate length with any particular morphological category. Thus, there is no reasonable way to reconstruct a rule or set of rules that could lead to the observed alternations. In such circumstances, paradigmatic leveling is liable to step in. On this account, language learners despair of a rule, and opt instead to choose a consistent vowel quantity on a word-by-word basis [emphasis, JV].
270
Voyles
Therefore the Modern English reflexes have various and random Middle English antecedents; tale < ME tal; god < ME god; hole < ME hol; beaver < ME bever, and devil < ME dev el. Whether or not one agrees with Dresher’s account is not at issue here. The point is that Iverson and Salmons cannot cite Dresher’s analysis as a parallel to their account of ON i-umlaut. If it were, one would expect a restructuring on a word-by-word basis, and hence a more or less equal occurrence in Old Norse of umlauted and non-umlauted long stems (for example, gestr ‘guest’ alongside *brastr ‘lack’, instead of the actually occurring brestr), or at least some more long-stem nonumlauted forms, as opposed to the overwhelming preponderance of ON i-umlaut in the long stems. What Iverson and Salmons actually suggest is that the opacity that they posit results not in word-by-word changes à la Dresher, but rather in a new generalization, namely, umlaut in all long stems. That is, in Iverson and Salmons’ terms, the long-stem forms have been “restructured,” but the proviso for the restructuring, namely “long stems,” does, in fact constitute a phonological condition. Accordingly, this analysis would seem to involve “salvaging a phonological rule from the situation,” and as such stands in direct contrast to the Dresher analysis—whether right or wrong—under which this is impossible. 3. The V Theory. 3.1. Old Norse i-Umlaut. As described in Voyles 1982 and 1992:115–118, the V theory assumes that the development of ON i-umlaut occurred as in 6, with the environment expanding in at least three successive stages: (6) [V(V), +stressed] [-back] 5 Stage 1: ... C0i(:) Stage 2: ... C0i(:) or j Stage 3: ... C0i(:) or j; or ... k’ or g’ or R Consider, for instance, the following derivations: 5
V(V) represents a short vowel, a long vowel or a diphthong; k’ and g’ are the palatalized allophones of /k/ and /g/ when they occur before front vowels; and R, which was probably [z] at this time, develops from Proto-Indo-European /s/ by Verner’s Law and eventually results in OI /r/.
Old Norse Umlaut
(7) Pre-umlaut
a. b. c. d. e. f. g. h. i. j.
ainaR katilaR katle natja sokijan taljan dagaR dag’e tak’enaR diuRa
Stage 1
a. b. c. d. e. f. g. h. i. j.
einaR ketilaR katle natja søkijan taljan dagaR dag’e tak’enaR diuRa
‘one’ ‘kettle’ NOM.SG. ‘kettle’ DAT.SG. ‘net’ NOM.SG. ‘seek’ INF. ‘tell’ INF. ‘day’ NOM.SG. ‘day’ DAT .SG. ‘taken’ PERF.PART . ‘animal’ NOM.SG.
(no change) (no change) (no change) (no change) (no change) (no change) (no change)
Stage 2: The forms are identical to those in Stage 1, except d. netja f. teljan Stage 3: The forms are identical to those in Stage 2, except h. deg’e i. tek’enaR j. diyRa Old Icelandic
a. b. c. d. e. f. g. h. i. j.
einn ketill katle net søkja telja dagr dege tekenn dr
271
272
Voyles
3.2. The ON Reflex of Sievers’ Law. In the ON grammar, a process of vowel insertion—probably the ON reflex of the Indo-European rule of Sievers’ Law, or perhaps, a restructured version of it—was interacting with ON i-umlaut. In any case, it represents a commonly occurring type of change, which can be represented as in 8. (8) Ø i / (a) VV (b) VC (c) VCV
C1 ... j
According to 8, /i/ is inserted before /j/ after one of the following: a long vowel or diphthong plus at least one consonant; a short vowel plus at least two consonants; two syllables (with the second one ending in at least one consonant), that is, after a closed long foot.6 Examples of the application of 8 are the pre-ON i-umlaut infinitives +sokjan ‘seek’ and +taljan ‘tell’ > +sokijan and +taljan (Sievers’ Law). 3.3. An Account of the Old Norse Developments. With i-umlaut and Sievers’ Law both active in the ON grammar, the occurrence of long-stem i-class umlaut versus its absence in short stems can be explained as follows. Even by very early ON times, the i-class was in the process of dissolution: virtually all the feminine i-class nouns had by this time already passed into the o-class, for example, +aksl+iR ‘shoulder’ (early replacement of the i-class ending +-iR by the o-class ending +-u) > +akslu > Old Icelandic xl. The same process was, albeit at a slower pace, also under way in the i-class masculine nouns. The stems of such nouns (consisting of the root plus the i-class marker /i/) received the endings that generally belonged to the a-class (with a few from the u-class). Later, the i-class marker was dropped, after which the a-class endings were the only ones retained. In view of this, the derivation of most long- and short-stem nouns proceeded as follows:
6 Note that in all three of these environments, at least one consonant, C1, must precede /j/. I return to this point immediately below.
Old Norse Umlaut
(9) stai+R stai+aR stajaR stajaR stajaR staaR star
gasti+R gasti+aR gastjaR gastijaR gestijaR gestaR gestr
273
‘place’, ‘guest’ (NOM.SG.) Addition of a-class ending Unstressed /i/ > /j/ when followed by a vowel Sievers’ Law ON i-umlaut, stage 1 Loss of the i-class marker, -(i)j-, in most words i-umlaut, stage 2 Old Icelandic
Of course, the transfer of i-class nouns into other classes was wavelike within the lexicon. Therefore, there were a few nouns that retained the i-class marker /-(i)j-/ until stage 2 of ON i-umlaut. An example of this is the short stem, but nonetheless umlauted, OI bylr ‘breeze’ < +bul +jaR. Conversely, there are also a few nouns that lost the /-(i)j-/ marker early and hence did not undergo ON i-umlaut, for example, burr ‘birth’ < +bur+ijaR. Iverson and Salmons (2004:91) attempt to discern a regularity in these irregularities: The exceptional masculine long-stem forms without umlaut inflect like regular short stems (which are also without umlaut), while the exceptionally umlauted short-stem words inflect like regular (umlauted) long-stem nouns. However, such cases are peripheral, and the Iverson and Salmons account is not totally accurate: the long stem sultr ‘hunger’ inflects like both gestr, as well as star. And the short stem bylr ‘gust of wind’ also inflects like star, as well as gestr. The V theory also accounts for the occurrence of ON i-umlaut in other classes. For example, in the j-class weak verbs, the ON i-umlaut occurs in long-stem forms throughout the paradigm since the i-umlaut occurred in these verbs earlier; hence, OI erfa, erfa, erfr ‘inherit’ < +arjan, etc. However, in the short-stem j-class weak verbs, in which the i-umlaut occurred later, the umlaut did not extend to the past indicative, but still occurred in the short-stem present, which retained the /j/; hence, OI telja, tala, talr ‘tell’. Note also in this connection the long-stem j-class weak verbs, which nonetheless inflect like telja above, such as OI lja ‘mash’, lúa, lúr < +lujan, etc. Although such verbs are long stems, they nonetheless inflect like a short-stem verb because Sievers’ Law
274
Voyles
could not apply here to produce *luijan. Therefore, the i-umlaut was not triggered by an /i/, but rather later by the /j/, as in the short-stem verbs. Also, one can note in this connection the occurrence of i-umlaut throughout the paradigms of ja-class nouns, both short and long stems, since—unlike the i-class nouns—the ja-class nouns were not in the process of changing their declension: OI kyn < +kunja, as well as lkner ‘doctor’ < +laknijaR. 3.4. Iverson and Salmons’ Critique. Iverson and Salmons (2004:89–90, note 6) offer the following three objections to the V theory. “First, the philological base of the analysis requires reading the inscription on the Stentoften Stone as the ‘guest’ word, which […] no longer appears tenable.” That is, what was once read as i-umlauted gestumR ‘guests’ is now read as non-umlauted hagestumR ‘stallions’. However, whether or not this is the case, the objection is immaterial. The inscription probably dates from ca. 675 C.E.; even if ON i-umlaut began later than 675 C.E., this has no bearing on the V theory. That is, the V theory maintains only that ON i-umlaut was operative sometime before the attestation of classical Old Icelandic. How long or shortly before this time is immaterial for the theory. Iverson and Salmons continue: “Second, it seems odd for Sievers’ Law still to have been active at this presumably late stage of umlaut development (see Kim 2001, for example).” Two points should be made with respect to this remark. First, it is by no means clear that Kim 2001 demonstrates that Sievers’ Law—or perhaps some altered version of it— had died out by ON times. In fact, Kim does not address the question of the reflex of Sievers’ Law in Old Icelandic, but only its reflex in Gothic. Further, Kim (2001:100) leaves open the possibility that Sievers’ Law was more active in certain “conservative daughter languages,” among which may well have been Old Icelandic. Second, and on the contrary, ON inscriptions teem with instances of the application of Sievers’ Law; for example, raunijaR ‘tester’ (ca. 200 C.E.) and irbijaR (proper name, ca. 450 C.E.; Antonsen 1975:29, 48); auja ‘luck’ (ca. 500 C.E.), where Sievers’ Law has predictably not applied, versus arbija ‘inheritance’ (ca. 400 C.E.), where it has (Johannesson 1923:105, 115); and in talija ‘tool’ (ca. 250 C.E.; Noreen 1970:392). I take these reflexes to indicate that Sievers’ Law was an active phonologically conditioned rule in the ON area at this time.
Old Norse Umlaut
275
Further, Krause (1971:94), who also cites numerous attestations of Sievers’ Law, remarks in this connection: “Eine ganze Gruppe solcher ‘inkorrekten’ Schreibungen sind die Bildungen mit -warijaR ‘Beschützer’.”7 Such forms are not incorrect; rather, they indicate that Sievers’ Law was still alive and productive in ON. As such, Sievers’ Law could be triggered by the environment (c) of the rule 8 above to ensure that the rule could still apply to newly formed compounds, such as the name stainawarijaR, whenever such newly formed compounds were reanalyzed as single words (stainawar+jaR, and not staina#war+jaR). The third objection to the V theory is as follows (Iverson and Salmons 2004:90, note 6): Tacking an extra set of endings onto the existing suffixes in a paradigm is itself peculiar in that double affix marking is typically only sporadic and idiosyncratic, as reflected in the hapax legomenon of the lone doubly marked plural in English, child-er-en. That is, analogy based on categories rather than individuals tends to replace endings, not build over them. Iverson and Salmons are objecting to the “double affix marking,” that is, the positing of a-class endings “on top of” the i-class markers, which were retained for a time, but later lost, leaving only the a-class endings (see 9). However, pace Iverson and Salmons, systemic “double affix marking” has been seen to occur elsewhere. One instance of it is the wholesale transfer of the on-class weak verbs into the jan-class in Old English, Old Frisian, and Old Saxon. In these languages, there was a mass migration of all the on-class weak verbs into the weak jan-class. However, in this process, the on -verbs for a while retained their on-class marker, to which they added the jan-class endings. Eventually these verbs dropped the on-class marker, as in early West Germanic:
7 “A class of such ‘incorrect’ attestations [of Sievers’ Law] is formations [of proper names] with -warijaR ‘protector’” [translation, JV].
276
Voyles
(10) mako+n salo+n ‘make’, ‘anoint’ mako+jan salo+jan Addition of jan-class ending = double affix mak+jan saljan Loss of o-class marker Old English makian, sealfian; Old Frisian makia, salvia. This transition must have occurred after the well-known Old English and Old Frisian palatalizations. Otherwise one would find Modern English */meit/ instead of /meik/, and Old Frisian *matsia */matia/ instead of /makia/. Also, it must have occurred after the time of Old English and Old Frisian i-umlaut; otherwise, one would find Old English *mekian and *sielfian, and Old Frisian *mekia and *selvia. Old Saxon is attested at a transitional stage in that it has makon, makojan and makian ‘make’ and salon, salojan, and salian ‘anoint’. The OHG verbs did not undergo this change; hence only machon ‘make’ and salbon ‘anoint’ occur. 4. Conclusion. Both of the theories discussed here give coherent accounts of ON i-umlaut. However, it has been argued that the I&S theory posits implausible and unattested types of change, whereas the V theory does not. ON i-umlaut is a perennially intriguing problem for Germanic linguistics, and Iverson and Salmons have appropriately identified it as a crucial issue. Exchanges continue to be useful, and I personally would like to see if progress could be made from the perspective of Optimality Theory. REFERENCES Antonsen, Elmer H. 1975. A concise grammar of the older runic inscriptions. Tübingen: Max Niemeyer. Campbell, Alistair. 1964. Old English grammar. Oxford: Clarendon Press. Dresher, B. Elan. 2000. Analogical levelling of vowel length in West Germanic. Analogy, levelling, markedness: Principles of change in phonology and morphology, ed. by Aditi Lahiri, 47–71. Berlin and New York: Mouton de Gruyter. Gutenbrunner, Siegfried. 1951. Historische Laut- und Formenlehre des Altisländischen. Heidelberg: Carl Winter. Holthausen, Ferdinand. 1900. Altsächsisches Elementarbuch. Heidelberg: Carl Winter.
Old Norse Umlaut
277
Iverson, Gregory K., and Joseph C. Salmons. 2004. The conundrum of Old Norse umlaut: Sound change versus crisis analogy. Journal of Germanic Linguistics 16.77–110. Johannesson, Alexander. 1923. Grammatik der urnordischen Runeninschriften. Heidelberg: Carl Winter. Kim, Yookang. 2001. Prosody and i/j alternation in Gothic. Journal of Germanic Linguistics 13.97–130. Krause, Wolfgang. 1971. Die Sprache der urnordischen Runeninschriften. Heidelberg: Carl Winter. Labov, William. 1972. Sociolinguistic patterns. Philadelphia: University of Pennsylvania Press. Labov, William. 1994. Principles of linguistic change, vol. 1. Internal factors. Oxford: Basil Blackwell. Noreen, Adolf. 1904. Altschwedische Grammatik. Halle: Max Niemeyer. Noreen, Adolf. 1970. Altnordische Grammatik. Alabama: University of Alabama Press. [Reprint of 4th edition, 1923.] Robinson, Orrin W. 1980. An exception to Old High German umlaut. American Indian and Indo-European studies: Papers in honor of Madison S. Beeler, ed. by K. Klar, M. Langdon, and S. Silver, 449–460. The Hague: Mouton. Steller, Walther. 1928. Abriss der altfriesischen Grammatik. Halle: Max Niemeyer. Voyles, Joseph B. 1982. Old Norse i-umlaut. Linguistics 20.267–285. Voyles, Joseph B. 1991. A history of OHG i-umlaut. Beiträge zur Geschichte der deutschen Sprache und Literatur 113.159–195. Voyles, Joseph B. 1992. Early Germanic grammar: Pre-, proto- and postGermanic languages. San Diego: Academic Press.
University of Washington Department of Germanics Box 353130 Seattle, WA 98195-3130 USA [
[email protected]]
Journal of Germanic Linguistics 17.4 (2005):279–294
REVIEWS
Der Erwerb der Verbstellungsregeln in der Zweitsprache Deutsch durch Kinder mit russischer und türkischer Muttersprache. By Stefanie Haberzettl. (Linguistische Arbeiten, 495). Tübingen: Max Niemeyer, 2005. Pp. v, 181. Paper. 68.00. Reviewed by CARRIE N. JACKSON, The Pennsylvania State University As indicated by its title, this book describes the acquisition of verb placement in German by four elementary-aged children whose first language is either Turkish or Russian. This work presents longitudinal data from an age group often neglected in language acquisition research and, as such, will be of particular interest to linguists and psycholinguists who work in the areas of both first (L1) and second (L2) language acquisition. Throughout the book Haberzettl stresses the descriptive nature of her study, stating: “die empirischen Beobachtungen sind das Fundament für die Genierung von Hypothesen, nicht für eine Hypothesenüberprüfung” (p. 2). This bottom-up approach is important, not only because of the limited amount of previous research conducted with this type of population, but also because it results in a book rich with empirical data from her subjects. Even though there are certain problems with how Haberzettl chooses to frame her work in relation to previous L2 acquisition research, this book represents a valuable contribution to language acquisition research, due to the nature of Haberzettl’s subject population and because of her presentation and discussion of the data. After a brief introductory chapter, Haberzettl devotes chapter 2 to a summary and discussion of three different explanatory models of language acquisition, namely acquisition within the framework of generative syntax and Universal Grammar (henceforth UG) (such as Flynn and Martohardjono 1998), functional approaches to language acquisition (such as Klein and Perdue 1992; Slobin 1973, 1985), and processing-oriented theories of language acquisition (such as Pienemann 1998). Haberzettl provides a good summary of many of the major findings from both L1 and L2 acquisition research, although one could argue that she devotes a disproportionate amount of time to discussing UG-
© Society for Germanic Linguistics
280
Reviews
based approaches, leaving less space for functional and processingoriented accounts. Chapter 3 provides readers with an overview of German word order. Haberzettl offers readers both a surface explanation of German word order, as well as a summary of German sentence structure within a generative framework. She argues, and correctly so, that while she presents her data in a purely descriptive format, because many German L2 acquisition studies have been conducted within a generative framework (for example, Schwartz and Sprouse 1994; Vainikka and YoungScholten 1994), it is important to present an overview of this analysis of German sentence structure as well. Chapter 4 provides a detailed description of the data collected by Haberzettl. The data from her four subjects, two Turkish and two Russian children who moved to Germany shortly before starting elementary school, come from a larger longitudinal study conducted from 1989 to 1992 with immigrant children in Augsburg. The children in the study were between six and eight years old at the beginning of the study and had varying degrees of contact with German native speakers. Haberzettl analyzes subjects’ responses obtained during a free speech activity at one-month intervals over the course of the study, with respect to the development of verb placement rules in German. For each subject, Haberzettl provides numerous tables and speech examples detailing his or her progression. In the first half of chapter 5, Haberzettl discusses the results of the descriptive analyses presented in chapter 4. In spite of individual differences among the four subjects, there are clear trends in the data reflecting the properties of the subjects’ L1. The Turkish children quickly “discover” the right verbal bracket in German, producing SXVfin sentences, and they have little or no trouble with modal and perfect constructions, or with dependent clause constructions as soon as they start including such constructions in their utterances. Through the use of ist ‘is’ as a sort of dummy auxiliary, the Turkish children also quickly learn to produce SVfinX sentences, although it takes them longer to correctly produce true verb-second constructions involving inversion. In contrast, the two Russian children produce SVfinX constructions from the earliest recorded stages, but their acquisition of the right verbal bracket progresses much more gradually. Similar to the Turkish children, it also takes the Russian children a relatively long time to correctly
Journal of Germanic Linguistics 17.4 (2005)
281
produce true verb-second constructions involving inversion. Important to her discussion of these findings compared to previous research, Haberzettl also reports no correlation between the development of verbal inflection and verb-second among her subjects, regardless of their L1. Based on these findings, Haberzettl argues that transfer from the L1, when supported by evidence from L2 German input, can best explain these subjects’ acquisition of German verb placement rules. The two Turkish children acquire the right verbal bracket early on because German verb-final input matches the SOV pattern in their native language. At the same time, the high degree of German verb-second input—especially with regards to the ist-copula, which does not exist in Turkish—highlights clear structural differences between Turkish and German. These differences are salient to the Turkish children early on, thus forcing them to quickly incorporate a left verbal bracket into their German interlanguage. By contrast, the two Russian children only gradually begin to exhibit correct verb placement in modal and perfect constructions, as well as in dependent clause constructions, because they incorrectly assume, based on German main clause input, that German is an SVO language similar to their native Russian. Since this is only a surface similarity between Russian and German, the combination of German L2 input and the resulting transfer of L1 Russian structures leads the Russian children down the garden path, making it more difficult for them to develop the crucial right verbal bracket in German. As previously stated, Haberzettl’s presentation of the data in chapter 4 and her interpretation of the data in the first portion of chapter 5 are the real strengths of this book. In contrast, the weakest section of this book is the second half of chapter 5, in which she discusses her findings in light of several theories of adult L2 acquisition, namely the UG-based Minimal Trees Hypothesis (Vainikka and Young-Scholten 1991, 1994), the UG-based Full Transfer/Full Access Hypothesis (Schwartz and Sprouse 1994), and Processability Theory (Pienemann 1998). Haberzettl argues that the results of her study pose problems for all three of these models. What Haberzettl does not acknowledge in her discussion, however, is that all three of these theories are based on adult L2 acquisition, whereas her subjects are in early elementary school. Therefore, it is entirely possible that her findings are not compatible with these
282
Reviews
explanatory models of L2 acquisition because of fundamental differences in how adults and children learn languages. In conclusion, this work is important both for its subject matter and the empirical data it presents. Haberzettl’s study examining early elementary-aged children learning German as a second language helps fill an important gap in L1 and L2 acquisition research (see Wegener 1992, 1995 for other research on this population). With the exception of Chomsky’s (1969) seminal study tracing the syntactic development of complex structures in American elementary school children, most L1 acquisition research stops at around age five. Similarly, L2 acquisition research focuses largely on child bilingualism, adult L2 acquisition, or classroom immersion learning. Haberzettl faces a difficult challenge when trying to situate her study in relation to previous research, given that there are few studies in the body of previous L1 and L2 acquisition research with which her findings are truly comparable. Therefore, in spite of my criticism of her final chapter, this book is a needed and valuable addition to the research literature on L1 and L2 acquisition. REFERENCES Chomsky, Carol. 1969. The acquisition of syntax in children from 5 to 10. Cambridge, MA: MIT Press. Hoekstra, Teun, and Bonnie Schwartz (eds.). 1994. Language acquisition studies in generative grammar. (Language acquisition and language disorders, 8.) Amsterdam and Philadelphia: John Benjamins. Flynn, Suzanne, and Gita Martohardjono (eds.). 1998. The generative study of second language acquisition. Mahwah, NJ: Lawrence Erlbaum Associates. Klein, Wolfgang, and Clive Perdue. 1992. Utterance structure. (Studies in bilingualism, 5.) Amsterdam and Philadelphia: John Benjamins. Pienemann, Manfred. 1998. Language processing and second language development: Processability theory. (Studies in bilingualism, 15.) Amsterdam and Philadelphia: John Benjamins. Schwartz, Bonnie, and Rex Sprouse. 1994. Word order and nominative case in non-native language acquisition. Hoekstra and Schwartz 1994, 317–368. Slobin, Dan I. 1973. Cognitive prerequisites for the development of grammar. Studies of child language development, ed. by Charles A. Ferguson and Dan I. Slobin, 175–211. New York: Holt, Rinehart and Winston. Slobin, Dan I. (ed.). 1985. The crosslinguistic study of language acquisition, vols. 1–2. Hilldale, NJ: Lawrence Erlbaum Associates.
Journal of Germanic Linguistics 17.4 (2005)
283
Vainikka, Ann, and Martha Young-Scholten. 1994. Direct access to X'-theory: Evidence from Korean and Turkish adults learning German. Hoekstra and Schwartz 1994, 265–316. Wegener, Heide. 1992. Kindlicher Zweitspracherwerb. Untersuchungen zur Morphologie des Deutschen und ihrem Erwerb durch Kinder mit polnischer, russischer und türkischer Erstsprache. Eine Längsschnittuntersuchung. Augsburg: Universität Augsburg Habilitationsschrift. Wegener, Heide. 1995. Die Nominalflexion der Deutschen—verstanden als Lernergegenstand. (Reihe Germanistische Linguistik, 151.) Tübingen: Max Niemeyer.
Department of Germanic and Slavic Languages and Literatures The Pennsylvania State University 311 Burrowes Building University Park, PA 16802 USA [
[email protected]]
Ellipsis in Comparatives. By Winfried Lechner. (Studies in Generative Grammar, 72), Berlin: Mouton de Gruyter, 2004. Pp. ix, 285. Hardcover. 98.00, $137.20 Reviewed by JOHN R. TE VELDE, Oklahoma State University The objective of this study is to make a contribution to the analysis of two construction-specific types of deletion processes, COMPARATIVE DELETION (CD) and COMPARATIVE ELLIPSIS (CE), the latter hypothesized to be involved in relating phrasal comparatives to clausal constructions. In five chapters (Introduction, Comparative Deletion, Comparative Ellipsis, Comparative Coordination, and Conclusion), Lechner outlines an analysis of comparatives that does not require such deletion processes. He defends two claims. First, comparatives do not involve construction-specific types of ellipsis, but rather result from overt AP-raising, and second, all ellipsis processes in comparatives are “reflected in syntax” (p. 2), that is, CD is an instance of syntactic ellipsis. The assumption that a CD-site is associated with an antecedent by overt
284
Reviews
movement, a strictly syntactic operation, entails that CD is not an instance of semantic ellipsis, contra Kennedy 1999, and Lerner and Pinkal 1995. Instead, Lechner proposes that all instances of CD result from the conjunction-reduction operations Gapping, Right Node Raising, and Across-the-Board Movement, which are typically confined to the domain of coordinate structures. In what follows, I present chapter summaries and brief critical comments, followed by an evaluation of the book. Chapter 1 provides a seven-page overview of the book, focusing especially on the complexity of the topic and the many proposals in the literature, and on the main tenets of the proposal presented in the remaining chapters. In chapter 2, Lechner first presents arguments for CD as an instance of syntactic ellipsis, and lays the groundwork for his AP-Raising Hypothesis. He argues against the assumption of semantic ellipsis that the CD-site is not restored until the derivation reaches semantic interpretation, and in favor of the assumption that it is restored in syntax. Support for this position is drawn from constructions in which clausal conjunctions of predicative adjectives may be targeted for deletion in one of two ways: either the adjectival head alone is targeted, as in 1a, or the adjective together with its argument, as in 1b (e = omitted element(s)). (1) a. Mary is proud of John and Bill is e of Sally. b. Mary is proud of John and Bill is e too. Lechner accounts for 1a via VP ellipsis, and for 1b via object-PP movement prior to the application of VP elipsis, which then appears to affect only the adjectival head proud, but actually targets this head and the trace of movement: (2) Mary is proud of John and Bill is [PP of Sally]i eVPE VPE = [VP [AP proud ti ]] Interestingly, the VP elipsis operation proposed for 2 appears to be a required deletion, in contrast to the optionality of conjunction reduction:
Journal of Germanic Linguistics 17.4 (2005)
285
(3) a. Mary wrote a book and Bill e/wrote an article. b. Mary is proud of John and Bill is of Sally e/*proud. c. Mary is proud of John and Bill is e/proud of Sally. The problem with 3b actually lies not in the non-optionality of the deletion, but rather in the fact that the proposed PP-movement has occurred, as indicated by 3c, where in the version without movement the optionality of deletion matches that in 3a. Therefore, it is not the unification of CD with coordinate ellipsis that causes a problem here, but rather the movement approach to CD. The principle C violation Lechner identifies in 4a is evidence, he claims, that the CD-site d-proud of John (where d equals degree) has been restored already during syntactic computation, that is, prior to semantics. However, he does not consider the alternate version in 4b, which undermines this assumption. (4) a. *Mary is prouder of Johni than hei is e (e = d-proud of Johni) b. Mary is prouder of Johni than hei himself is e Section 3 of chapter 2 deals with the syntax of NP-comparatives and lays out the arguments for a movement approach to CD-constructions with than-XP, as in 5a, given the structure in 5b. (5) a. Mary knows younger authors than Peter knows. b. … [DP [DegP [AP [AP younger [NP authors]] [ [ Degcomp than Peter] In the derivation of 5a, Lechner proposes overt raising of the AP from the Spec of DegP of the comparative complement to the Spec of DegP of the matrix clause. Movement is motivated by the need to eliminate a [+comparative] feature on the higher head Deg. Lechner’s account maintains that in NP-comparatives, AP and NP form a constituent to the exclusion of the degree head, and that AP-raising targets only the subconstituent APNP, stranding the determiner. In section 5 of chapter 2, Lechner considers four predictions made by his AP-Raising Hypothesis involving four systematic differences between prenominally and postnominally modified NP-comparatives. In section 6, he argues that the complex distribution of weak and strong
286
Reviews
crossover effects is consistent with the AP-Raising Hypothesis, and that the availability of de dicto readings for the CD-site in Sally wants to buy many books poses a problem for semantic resolution. He considers two semantic theories, Lerner and Pinkal 1995 and Kennedy 1997, 1999, and shows how the AP-Raising Hypothesis solves certain problems arising under these theories. In the final section of chapter 2, Lechner addresses the question of why the empty operator in SpecCP of the comparative complement does not block movement of the AP by Relativized Minimality. His answer is that relativized minimality effects, such as wh-island violations, are detectable only in cyclic derivations; because AP-Raising does not involve cyclic derivations, relativized minimality effects do not show up. Lechner provides arguments for the assumption that AP-raising does not form a chain, and is therefore exempt from Relativized Minimality. Chapter 3 pursues two main empirical objectives. First, it presents arguments for an ellipsis analysis of phrasal comparatives (those having a single remnant), such as John is prouder of his dog than [Mary]. Second, it investigates the syntax of partially reduced comparatives, such as John is prouder of his dog than [Mary] [of her cat]. Lechner argues for a unified analysis based on his AP-Raising Hypothesis and conjunction-reduction operations because it does not rely on a construction-specific reduction operation, such as Bresnan’s (1975) CE analysis. The analysis “entails significant theoretical consequences” (p. 89): verb-second movement occurs in syntax (not in PF), and the Coordinate Structure Constraint must be computed in syntax (not just in semantics). In section 2 of chapter 3, Lechner examines the weak version of his hypothesis, termed the Conjunction Reduction Hypothesis with respect to the assumption that CD should be handled by conjunction-reduction operations. According to the weak version of this hypothesis, “conjunction-reduction operations can target comparatives.” Indeed, the derivation of John spoke more vehemently against Mary than Tom spoke against Jane looks very much like a Gapping construction. To derive the reduced 6a with Right Node Raising, Lechner must assume the nonreduced 6b as a target. (6) a. I organize more than I actually run her life. b. *I organize more her life than I actually run her life.
Journal of Germanic Linguistics 17.4 (2005)
287
Lechner does not point out, however, that 6b is ungrammatical. Required for his analysis is the assumption that comparatives can be assigned parses that are sufficiently similar to base-generated coordinate structures, that is, they are cases of “comparative coordination,” and that stripping (see Merchant 2001) is “radical Gapping.” The bulk of section 2 is devoted to outlining the internal and external conditions that must be met by conjunction-reduction operations in both elliptical coordinate structures and CD. The third section of chapter 3 is devoted to a comparison of the strong version of the Conjunction Reduction Hypothesis with a number of competing theories of CE. The strong version asserts that all deletion in comparatives derives from conjunction reduction. Lechner’s main goals are to show that conjunction reduction derives all the surface syntactic effects traditionally attributed to CE, and to demonstrate that the differences between phrasal comparatives and clausal comparatives do not reflect an idiosyncratic property of the former, but can be expressed in terms of more general disparities between non-reduced and reduced comparatives. The final section of chapter 3 concerns additional restrictions on comparative coordination. Here Lechner, following Lechner 2001, expands on a problem for the Conjunction Reduction Hypothesis in certain types of phrasal comparatives in German. His solution involves a filter called the Comparative Deletion Scope Condition: The comparative has to take scope over (c-command) the CD-site at LF. A second important component of his analysis is the Coordinate Structure Constraint. Lechner follows Johnson 1996 in assuming that Case-driven subject movement out of the first conjunct is exempt from this constraint, and he adopts Williams’ (1978) condition on Across-the-Board Movement. Lechner must also assume that asymmetric comparative extraction does not induce the Coordinate Structure Constraint. In chapter 4, Lechner defends the assumption that at some point in the derivation, comparatives are assigned a parse similar to one commonly associated with coordinate structures. This assumption must be squared with the AP-raising analysis of chapter 2, where the than-XP semantically serves as an internal argument of the comparative degree head in the matrix clause. Lechner comes to the conclusion that
288
Reviews
comparatives are hybrid constructions with properties of both subordination and coordination. A particularly important assumption is addressed in the fourth section in which Lechner claims that comparatives can be assigned coordinate-like parses at some point in their derivation. Central to this claim is the assumption that there are syntactic as well as semantic constraints on conjunction. Because the Coordinate Structure Constraint plays an important role in his account, it must be defended as having a syntactic basis. Lechner therefore argues that Munn’s (1993) claim that this constraint has no syntactic force is not tenable because Munn’s alternative—analyzing coordination as a parasitic gap construction— cannot be maintained. The final chapter is a five-page conclusion in which Lechner summarizes the main points of his AP-Raising Hypothesis, and how it leads to a principled analysis of a variety of phenomena. Turning now to the evaluation of the book, the central question to address is whether elliptical comparative constructions are fundamentally the same as, or at least very similar to, elliptical coordinate structures, sufficiently so to warrant the unification that Lechner attempts. It is this unification on which the analysis is based on, and it represents the most significant innovation of Lechner’s study. Lechner uses the constructions in 7 to illustrate how CE can be unified with conjunction-reduction operations. (7) a. Gestern habeni viele Leute Sam besucht ti yesterday have many people Sam visited und einige Otto eingeladen ti. and some Otto invited ‘Yesterday many people visited Sam and some invited Otto.’ b. Gestern habeni mehr Leute Sam yesterday have more people Sam
besucht ti visited
als e-CD Otto eingeladen ti. than Otto invited ‘Yesterday more people visited Sam than invited Otto.’
Journal of Germanic Linguistics 17.4 (2005)
289
What this analysis cannot account for is the possibility of V-raising in the second clause of 7a as in 8a, leading to the analysis in 8b. (8) a. Gestern habeni viele Leute Sam besucht ti yesterday have many people Sam visited und einige haben Otto eingeladen. and some have Otto invited ‘Yesterday many people invited Sam and some invited Otto.’ b. Gestern habeni viele Leute Sam besucht ti yesterday have many people Sam visited und einige ei Otto eingeladen ti. and some Otto invited ‘Yesterday many people visited Sam and some invited Otto.’ The comparative in 7b is actually structurally more similar to the construction in 9, where the conjunction of two vPs occurs without any trace of auxiliary raising (it is merged in C in a phase-based approach). (9) Gestern haben viele Leute yesterday have many people [vP Sam besucht und [vP Otto eingeladen]]. Sam visited and Otto invited ‘Yesterday many people visited Sam and invited Otto.’ It would be interesting to consider whether CDs provide any independent evidence of the traces that Lechner proposes, or whether a “traceless” approach along the lines of 9 is more suitable. A more general concern with Lechner’s unification of CE with conjunction-reduction operations is the lack of the prosodic requirement in CE that plays a central role in Gapping, shown in 10a, and Right Node Raising, shown in 10b (see Hartmann 2000). Obligatory focus accent is indicated with : (10) a. Bill fixed the sink and Mary fixed the door. b. Bill fixed the door and Mary painted the door.
290
Reviews
For instance, Lechner’s example of CE in 11 (his 26a, and others are similar), requires no focus accent of this sort. (11) More people visited Millhouse on Monday than visited Otto on Friday. Lechner does not mention the role of focus accent in connection with Gapping until chapter 4, where he states that it is an “interfering factor” (p. 211) that must be left to further research. Interesting also is the fact that the construction in 11 is much more similar to the non-elliptical construction in 12 than to the Gapping constructions in 10, in terms of both, structure and prosody. (12) The people visited [v’[DP Millhouse] [PP on Monday] and [v’[DP Otto [PP on Friday]]. Lechner’s construction in 11 differs substantially from a Gapping construction. First, Gapping cannot target a finite verb in the initial position of the matching clause, that is, immediately after the conjunction. Second, Gapping requires focus accent. Evidence for the former is given in 13, where we see that the struck-through verb cannot be assumed to exist underlyingly and be a target of Gapping according to Lechner’s approach because the non-reduced version is ill-formed: (13) Bill fixed the sink and fixed/#fixed the drain. Lechner claims that the same constraints applying to conjunctionreduction operations also apply to CE; one of these is that it is impossible for Gapping to elide a string that includes a finite CP-node (see p. 100). However, as shown in 14, this constraint depends on what occupies C. If a verb occupies C, Gapping is allowed, whereas if a complementizer occupies C, Gapping is ruled out.
Journal of Germanic Linguistics 17.4 (2005)
291
(14) a. Gestern lasi Peter den neuesten Harry Potter-Roman, yesterday read Peter the newest Harry Potter novel und [CP heute ei Paul den ältesten]. and today Paul the oldest ‘Yesterday Peter read the newest Harry Potter novel and today Paul the oldest.’ b. *Ich weiß, dass Peter den neuesten Harry Potter-Roman las I know that Peter the newest Harry Potter novel read und [CP dass Paul den ältesten ei]. and that Paul the oldest So while there is some flexibility in Gapping with respect to the constraints that Lechner discusses, the same flexibility does not seem to exist in CE, leading, along with the other concerns raised earlier, to the question of whether Lechner’s Conjunction Reduction Hypothesis for CE is entirely on the right track. Certainly, unification is a laudable goal, and it may well be possible on some level. However, the syntactic and semantic principles underlying the operations that license conjunction reduction must themselves be unified with one another and with the grammar of simplex sentences at the proper level of abstraction that accounts for the unique properties of coordinate structures. Lechner’s discussion of coordinate ellipsis does not reach this level. Therefore, the unification within the Conjunction Reduction Hypothesis between CD and conjunction reduction is not as sharp as it could be. It sometimes misses the mark, while other times it overreaches. Lechner uses the term isomorphism as an internal condition on the “depth” at which the gap and its antecedent are embedded. He claims that More people came than expected Mary to come is ungrammatical because Gapping has applied to a non-isomorphic context in which the gap is more deeply embedded than its antecedent (p. 126). A more principled account would first of all recognize that the non-elliptical version of this sentence is semantically ill-formed, and secondly, that the gap to come cannot in this context be targeted by any conjunctionreduction operation in an equivalent coordinate structure for reasons that are not related to embeddedness per se. Similar problems can be found in
292
Reviews
Lechner’s discussion of the distribution of prepositions in CD (pp. 127ff.). A better statement of the condition on this type of ellipsis would in my estimation be the Parallelism Requirement on conjunction reduction (see Chomsky 1995:203, and Hornstein and Nunes 2002). The fact that this requirement applies to non-elliptical coordinate structures also is overlooked by Lechner, as is evident in note 130, where he argues that a matching effect in 15 “does not require coordination.” (15) *Hans verdient [einen besseren Job]acc Hans deserves a better job als enom ihm offeriert wurde. than him offered was ‘Hans deserves a better job than was offered to him.’ A close equivalent of 15 involving conjoined relative clauses, that is, coordination, shows the same matching effect: (16) Der Job, deni Hans verdient hat und *ei/der ihm the job that Hans deserves has and ei/that him angeboten wurde, ist sicher. offered was is secure ‘The job that Hans deserves and that was offered to him is secure.’ Both relative clauses cannot share the relative pronoun den because of a matching requirement on Case that is the same as Lechner’s matching effect in 15, but without any comparative. As this example indicates, the Parallelism Requirement is only descriptive. The best unified analysis of coordination and CD would seek to explain this requirement in terms of syntactic principles that license the ellipse, and semantic principles that account for the interpretation. For the best unification, these principles should be fundamentally the same as those in simplex constructions. Lechner uses a number of descriptive generalizations to test his Conjunction Reduction Hypothesis, making conditions out of many, where underlying principles of grammar would be more suitable.
Journal of Germanic Linguistics 17.4 (2005)
293
Lechner’s proposal depends heavily on the syntactic viability of the Coordinate Structure Constraint (a descriptive generalization). It is possible, in my view, that abandoning or modifying the approach under which the Coordinate Structure Constraint is a necessary and properly stated syntactic condition on movement in both coordinate and comparative constructions would allow a stronger, less stipulated account, with fewer unorthodox movements. In fact, a number of problems with this constraint have been cited in the literature, for example, in Fox 2000, and Lechner does not consider solutions to them. The presentation of this book is very methodical, thorough, well organized and rich in data, though consideration of more languages, in addition to English and German, might be quite illuminating. The complexity of the problems must be underscored. However, my own view is that the problems would be less complex if certain assumptions were abandoned and the differences between coordinate structures and comparatives were to be examined in light of some recent developments in syntactic theory. Questions to which no sufficient answers are provided include the following: What is *embeddedness (see pages 133– 137), that is, why should it exist as claimed? What is isomorphism, and why should it block certain types of Gapping? What is extraposition (of which than-XP-raising is a type), and how can this type of movement be made compatible with the other assumptions made about asymmetric phrase structure? Despite these perceived shortcomings, and although the book would have benefited from careful reading by a native speaker and closer copyediting, I recommend it without serious reservations to all linguists interested in the syntax and semantics of comparative and coordinate structures and related topics. It makes a significant contribution to the literature on comparatives. REFERENCES Bresnan, Joan. 1975. Comparative deletion and constraints on transformations. Linguistic Analysis 1.25–74. Chomsky, Noam. 1995. The minimalist program. Cambridge, MA: MIT Press. Fox, Danny. 2000. Economy and semantic interpretation. Cambridge, MA: MIT Press.
294
Reviews
Hartmann, Katharina. 2000. Right node raising and gapping: Interface conditions on prosodic deletion. Amsterdam and Philadelphia: John Benjamins. Hornstein, Norbert, and Jairo Nunes. 2002. On asymmetries between parasitic gap and across-the-board constructions. Syntax 5.26–54. Johnson, Kyle. 1996. In search of the English middle field. Manuscript, University of Massachusetts, Amherst. Kennedy, Christopher. 1997. Projecting the adjective: The syntax and semantics of gradability and comparison. Doctoral dissertation, Department of Linguistics, University of California, Santa Cruz. Kennedy, Christopher. 1999. Projecting the adjective: The syntax and semantics of gradability and comparison. New York: Garland. Kennedy, Christopher, and Jason Merchant. 1997. Attributive comparatives and bound ellipsis. (Linguistics Research Center Report, LRC-97-03.) Santa Cruz: University of California, Santa Cruz. Lechner, Winfried. 2001. Reduced and phrasal comparatives. Natural Language and Linguistic Theory 19.683–735. Lerner, Jan, and Manfred Pinkal. 1995. Comparative ellipsis and variable binding. Proceedings of SALT (Semantics and Linguistics Theory) 5.222–236. Merchant, Jason. 2001. The syntax of silence. Sluicing, island, and the theory of ellipsis. Oxford: Oxford University Press. Munn, Alan. 1993. Topics in the syntax and semantics of coordinate structures. Doctoral dissertation, Department of Linguistics, University of Maryland. Williams, Edwin. 1978. Across-the-board rule application. Linguistic Inquiry 9.31–43.
Oklahoma State University Department of Foreign Languages and Literature 309 GU Stillwater, OK 74078-1054 USA [
[email protected]]