In this book, Stroik and Putnam take on Turing's challenge. They argue that the narrow syntax – the lexicon, the Numeration, and the computational system – must reside, for reasons of conceptual necessity, within the performance systems.
EDITORS: Matthew Baerman, Greville G. Corbett, Dunstan Brown TITLE: Defective Paradigms SUBTITLE: Missing Forms and What They Tell Us SERIES TITLE: Proceedings of the British Academy PUBLISHER: Oxford University Press YEAR: 2010
Kyle Gorman, Department of Linguistics, University of Pennsylvania, kgorman@ling.upenn.edu
SUMMARY Defectivity is surely one of morphology's dirtiest little secrets. Despite speakers' prodigious abilities to apply familiar patterns to unfamiliar data (e.g., wug-tests and the like), there are, as Halle (1973) notes, corners of the grammar where this generalization ability fails. For instance, many English speakers find themselves unable to generate a preterite form of <forgo>: despite <underwent>, both *<forwent> and *<forgoed> are ungrammatical. In the introductory chapter to 'Defective Paradigms', a collection of papers from a 2008 conference of the same name, Matthew Baerman and Greville G. Corbett set the stage, declaring that this type of data has important consequences for morphological theory, because it is widely assumed that ''[m]orphology ought to be the handmaid of grammatical meaning, producing forms where other components require them'' and that ''there will be productive, default mechanisms that permit the generation of a paradigm from any item whatsoever'' (p. 2), assumptions difficult to reconcile with the behavior of <forgo>.
Stephen Anderson's contribution, 'Failing one's obligations: Defectiveness in Rumantsch reflexes of DE:BE:RE', focuses on the paradigm of the verb <dueir> 'should' in a lesser-known Romance language, Surmiran. The author reports that the verb cannot be used in inflections where it would be expected to show a stem alternation (presumably, /do-/); instead, speakers substitute the verb <stueir> 'must'. Anderson proposes that this is an instance of neutralization to the semantically-similar <stueir> paradigm. Anderson's chapter is unique in grammaticalizing the avoidance pattern (cf. Legendre 2009 for a similar approach of syntactic ineffability): that is, the author proposes that descriptively ''defective inputs'' (e.g., the 1sg. present of <dueir>) are neutralized to licit inputs by the grammar (e.g., <stò> 'I must'). In contrast, the other authors appear to assume (albeit implicitly) that ''defective inputs'' crash the derivation.
Gilles Boyé and Patricia Cabredo Hofherr assume that morphological paradigms are defined by lists of suppletive cells and their interactions and use this to derive defectivity in their chapter 'Defectiveness as stem suppletion in French and Spanish verbs'. For Spanish, this means that the verb <dormir> 'to sleep' is constructed from the stems /dorm-, dwerm-, durm-/ (e.g., <duermo> 'I sleep', <durmamos> 'we would sleep'). Thus a defective paradigm cell is one for which the learner knows no stem. The authors then test a competing hypothesis. Albright (2003) has suggested that defectivity in Spanish verbs occurs when learners are faced with competing generalizations supported only by low-frequency data; this uncertainty is the quantity measured by the confidence scores generated by Albright's minimum generalization learner (MGL). Boyé and Cabredo Hofherr find no correlation between MGL confidence and defectivity in French, leading them to reject Albright's hypothesis.
Andra Kalvača and Ilze Lokmane ('Defective paradigms of reflexive nouns and participles in Latvian') present a new case of defectivity: Latvian reflexive nouns lack a plural instrumental, and both singular and plural lack dative and locative forms, and reflexive participles lack all but the accusative and instrumental in the singular, and lack the instrumental, dative, and locative in the plural. Kalvača and Lokmane show that this gap is general to nearly all Latvian reflexives; apparent exceptions are non-standard and/or semantically non-equivalent. Unlike many of the other cases reviewed in this volume, the Latvian data suggest that the source of defectivity need not make any reference to surface form at all.
John Löwenadler's study ('Relative acceptability of missing adjective forms in Swedish') is a summary of a forthcoming article (it has since appeared in the journal Morphology) investigating the description of the well-known case of defective neuter indefinite adjectives in attributive position in Swedish. Löwenadler administers a rating task to speakers and analyzes the results using phonological markedness constraints.
In their chapter 'Defective verbal paradigms in Hungarian -- Description and experimental study', Ágnes Lukaćz, Péter Rebrus, and Miklós Törkenczy administer a judgment task and show a strong correlation between speakers' judgements and their analysis.
Martin Maiden and Paul O'Neill ('On morphomic defectiveness: Evidence from the Romance languages of the Iberian Peninsula') examine the evidence for gaps in Spanish and Portuguese. The authors' analysis has certain similarities with the chapter by Boyé and Cabredo Hofherr in that it relies on the notion that any stem change is suppletion. But whereas Boyé and Cabredo Hofherr claim that Romance defective verbs lack suppletive stems, Maiden and O'Neill argue that defective verbs are those incapable of undergoing suppletion. This has a precedent in Iverson's description of Swedish inflectional gaps (1981: 141): ''one characteristic common to all cases [of defectivity--KG] is that paradigms with gaps show no allomorphy of the root morpheme, but would if the gap were filled.'' They claim that ''defective verbs of learnèd origin never show any kind of [root--KG] allomorphy'' (though they note one exception in Portuguese), and that this inability to show stem alternations is the key to generating defectivity.
Marianne Mithun's chapter ('The search for regularity in irregularity: Defectiveness and its implications for our knowledge of words') reports defectivity in both Yup'ik and Mohawk dual possessives, and argues that this is evidence for whole-word storage.
Milan Rezac's chapter 'Ineffability through modularity: Gaps in the French clitic cluster' finds that, except in one familiar case, constraints on clitic clusters belong in morphology, since any ''repairs'' to clitic clusters are invisible to the syntax. Rezac concludes that this is incompatible with globalist approaches to the morphology/syntax interface.
Lastly, Gregory Stump ('Interactions between defectiveness and syncretism') considers cases where syncretism and defectivity overlap in Sanskrit and daughter languages. Stump considers a hypothetical language in which defectiveness and syncretism may partially overlap. Consider a four-cell paradigm (with cells {A, B, C, D}) and imagine further that evidence from other paradigms in the same language suggest a syncretism between {A, C} and defectivity for {C, D}. How then will the four cells of this paradigm be realized? First, defectiveness could override syncretism, so that C is defective, despite the expectation of syncretism with A. Alternatively, syncretism could override defectiveness, in which case C is well-formed, showing syncretism with A. Finally, it is possible that ''syncretism determines a domain of defectiveness'': that is, C's status as defective renders A defective, and the only cell which can be realized is B. Stump gives evidence for all three patterns, and presents an analysis using Paradigm Function Morphology.
EVALUATION In general, the descriptive content of this volume alone will be of great interest to morphologists. The one place where the description is unclear is in Mithun's chapter. While the editors in their introductory chapter are careful to distinguish between defectivity and neutralizations in the morphology (i.e., syncretism), Mithun does not provide the reader enough information to confirm that this distinction is being respected (though I do not intend to discount the description itself). Asked for the 1sg. possessive form of ''paper'', a Yup'ik informant responds:
''My two. That's hard. It just jumps to plural. There is no dual'' (p. 129).
Mithun takes this comment to indicate defectivity, but this limited description is just as consistent with a partial syncretism between the dual and plural, something which is not uncommon cross-linguistically. Unfortunately, the reader is not given enough information to decide.
The theoretical content of the volume also merits morphologists' attention. I will address issues raised by the analyses of defectivity given in this volume in schematic order, rather than the alphabetic order in which the chapters appear. Among the papers that deal with the cause of defectivity, there are two major themes: missing stems and phonotactic constraints.
Regarding the former, advanced by Boyé and Cabredo Hofherr, and Maiden and O'Neill, I would like to present a case for skepticism. My own view is that the assumption that different forms of stems, even when the alternations are minimal or predictable, are always stored in memory rather than derived online, denigrates the long tradition of morphology, while making few interesting predictions. The one prediction this account makes with regard to inflectional gaps is one that I suspect is false (though I concede that this is an empirical matter): it predicts that defectivity (and full suppletion, as in <go/went>; see Embick & Halle 2005 for discussion) should be just as common as other types of stem change. Chan (2008: chapter 4) finds, with reference to Spanish, that the primary linguistic data to which children are exposed is so sparse that stem storage alone would leave the child with a grammar in which defectivity is practically the norm, not the uncommon phenomenon it is.
Maiden and O'Neill's proposal is difficult to evaluate for lack of explicitness. The authors claim that Spanish defective verb <abolir> 'to abolish' violates the expectation that it should have a stem-final [u] vowel in certain inflections, such as the 3sg. preterite (in fact, this [o ~ u] alternation is only attested in two other /-i-/-conjugation verbs, the non-defective <dormir and <morir>, and derivatives thereof), but they do not spell out why a failure to meet this expectation should result in ungrammaticality, nor do they explain why the 3sg. preterite indicative <abolió> (p. 105), which also violates this expectation (cf. <durmió, murió>), is well-formed.
Three chapters, the introduction and those by Löwenadler and Lukaćz et al., consider the hypothesis that some inflectional gaps are due to inviolable phonotactic constraints. However, prior findings indicate that these gaps must not be entirely phonotactic in nature, and in fact suggest the lexical nature of the pattern of defectivity. Löwenadler assumes that the Swedish defectivity is the result of phonologically-general inviolable constraints, but this is known to be inadequate: Buchanan (2007: 7) and Löfstedt (2010: chapter 5) identify morphological and lexical exceptions, respectively. Löwenadler's results provide some anecdotal support for lexical effects in this case: for instance, two loanwords from French, <disträ> 'absent-minded' (cf. French <distrait>) and <blase> 'blasé', differ drastically in their acceptability as neuter indefinite adjectives, despite their similar shape. This study also departs considerably from the transcription given in prior studies of this case (e.g., Iverson 1981, Buchanan 2007, Löfstedt 2010), without any discussion as to why this choice was made; for instance, Swedish <fatt>, the neuter indefinite attributive form of 'bland' is transcribed as [fat] instead of the [fat:] given by other authors.
The phonotactic analysis given by Lukaćz et al., while sufficient for the data considered in that chapter, does not extend to other superficially-similar defective verbs in Hungarian. Verb stems which end with a final consonant cluster in citation form generally show a harmonic vowel between the two stem-final consonants when a consonant-initial suffix is attached. For instance, the infinitive of 'to collapse' is <omlik> with the vowel-initial /-ik/ suffix, but the subjunctive form, marked with /-jon/ is <omoljon>, with an stem-internal [o] not seen in the infinitive. Some final-cluster verbs may not occur with these suffixes; speakers reject both forms with an intervening harmonic vowel, as well as (phonotactically invalid) forms without an intervening vowel. The authors argue that these gaps result when certain /CC/- final stems, those which are not marked to undergo an lexically-specific rule of epenthesis, bear a consonant-initial suffix. However, Rebrus and Törkenczy (2009) note that certain consonant-initial suffixes may not attach to some consonant-final roots, even if the /CC-C.../ structure created would undergo a general process which deletes one of the three consonants, resulting in a phonotactically valid [C.C] cluster. For instance, the defective subjunctive for the verb 'to conceive' (stem <fogamz->), where it counterfactually licit, would have no phonotactic violations at all: the medial consonant cluster would be [m.z], since /j/, in the subjunctive suffix /-jon/, deletes when it follows a /z/. To maintain the intuition that a phonotactic constraint results in defectivity, the constraint must have the power to ''crash'' the derivation before the process of /j/-deletion, or /j/-deletion must somehow be constrained in a fashion that it does not apply to stems like <fogamz->. But either way, a surface-true inviolable phonotactic constraint is insufficient to generate the full pattern.
Stump's chapter poses a methodological conundrum. The author assumes that an singular accusative/ablative syncretism in the personal pronouns in Vedic should also be in effect for the relative pronouns, which exhibit defectivity. It is trivial, however, to find an example where a personal and relative pronouns do not share a pattern of syncretism. For instance, Latin shows a syncretism in the singular accusative and ablative personal pronouns (1sg. <ME, ME>, 2sg. <TE, TE>, 3sg. <SE, SE>), but accusative and ablative are distinguished in the relative pronouns (masc.sg. <QVEM, QVO>, fem.sg. <QVAM, QVA>, neut.sg. <QVOD, QVO>). Without the assumption that these two classes must share patterns of syncretism, there is no expectation that there would be any syncreticism/defectivity overlap at all. In fact, this alternative provides a simple explanation for a pattern Stump labels ''defectiveness overrides syncretism''. Vedic neuter nominative and accusative singulars personal pronouns are syncretic, but the author claims that the neuter acc.sg. personal pronoun [enat] cannot be used as the nom.sg. All that needs to be said regarding the fact that syncretism fails to ''rescue'' this defective cell is that learners have not extended the nom./acc. syncretism to the defective paradigm.
Given that this collection focuses on paradigmatic approaches to morphology, it is unfortunate that ''paradigm'' is left undefined, as this determines what a paradigmatic theory of defectivity must account for. One important question that is not addressed by this volume is whether derivational morphology, however defined, is paradigmatic. The introductory chapter refers to derivation as a diachronic source for defectiveness, but in the absence of further evidence, it seems more natural to assume that the Russian defective verb <pylesosit'> 'to vacuum' is synchronically, not just diachronically, related to the noun <pylesos>. Another important question that the definition of paradigm would bear on is whether defectivity should extend beyond the word. If the word determines the scope of defectivity, then a large number of English adjectives which are unable to form a synthetic comparative (e.g., <smart-er> but *<intelligent-er>) are defective; alternately, if the paradigmatic domain extends beyond the word, this is competition between a word and a phrase, not a canonical case of defectivity.
The volume unfortunately does not include several of the papers which were presented at the conference (though handouts and slides are available at the website of the Typology of Defectiveness Project, administered by the editors: http://www.defectiveness.surrey.ac.uk/conference.htm). I mention these omitted papers only to direct interested readers to related material, since the volume itself is uniformly intriguing. This collection is an addition to the growing literature which takes defectivity seriously, but it is by no means the last word.
REFERENCES Albright, Adam. 2003. A quantitative study of Spanish paradigm gaps. In G. Garding and M. Tsujimura, eds., Proceedings of the 22th West Coast Conference on Formal Linguistics, 1-14. Cascadilla, Somerville, MA.
Buchanan, Charles. 2007. Deriving asymmetry in Swedish and Icelandic inflexional paradigms. Master's thesis, Universitetet i Tromsø.
Chan, Erwin. 2008. Structures and distributions in morphology learning. Doctoral dissertation, University of Pennsylvania.
Embick, David and Morris Halle. On the status of stems in morphological theory. In T. Geerts, I. van Ginneken, and H. Jacobs, eds., Romance Languages and Linguistic Theory 2003: Selected papers from Going Romance 2003, 37-62. Amsterdam: John Benjamins.
Halle, Morris. 1973. Prolegomena to a theory of word formation. Linguistic Inquiry 4(1): 3-16.
Iverson, Gregory. 1981. Rules, constraints and paradigmatic lacunae. Glossa 15(1): 136-144.
Legendre, Géraldine. 2009. The neutralization approach to ineffability in syntax. In C. Rice and S. Blaho, eds., Modeling ungrammaticality in Optimality Theory, 237-266. Equinox, London.
Löfstedt, Ingvar. 2010. Phonetic effects in Swedish phonology: Allomorphy and paradigms. Doctoral dissertation, University of California, Los Angeles.
Rebrus, Péter and Miklós Törkenczy. 2009. Covert and overt defectivity in paradigms. In C. Rice and S. Blaho, eds., Modeling ungrammaticality in Optimality Theory, 195-236. Equinox, London.
ABOUT THE REVIEWER
ABOUT THE REVIEWER:
Kyle Gorman is a graduate student at the University of Pennsylvania. His
dissertation research deals with the acquisition of morphological
defectivity. He also studies phonology, language variation, and natural
language processing.
