AUTHOR: John Frampton TITLE: Distributed Reduplication SERIES TITLE: Linguistic Inquiry Monographs PUBLISHER: MIT Press YEAR: 2009
Michael Maxwell, Center for Advanced Study of Language, University of Maryland
At its simplest, morphological reduplication seems trivial: duplicate all or part of a word.
Of course, it is not that simple. First, while morphology and phonology seem to be finite state for the most part, unlimited (whole word) reduplication is not.
Second, when the entire base (the input to the rule of reduplication) is not reduplicated, complications arise in specifying how much of the base gets reduplicated. Generally, what is reduplicated is defined in terms of a target, rather than in terms of what part of the base is reduplicated. For example, the target may be a closed syllable, but the phonemes used to create that closed syllable need not have constituted a syllable in the base.
But the greatest complications come from the interaction of phonology with reduplication. Phonological rules which one would expect to apply only to the base or only to the reduplicant apply to both (over-application), or to neither (under-application).
All of this makes for fertile ground for testing linguistic theories, and reduplication has been front and center in the debate between rule-based theories and constraint-based theories. For a time, it appeared that constraint-based theories (chiefly Optimality Theory, OT) had the upper hand; in the words of McCarthy and Prince (1995), ''... Correspondence Theory is superior, empirically and conceptually, to serial derivational approaches. All such theories are incapable of dealing with cases in which B[ase] copies (or, more neutrally, reflects) R[eduplicant].'' However, linguists working in rule-based approaches were not about to accept defeat so willingly, and modifications to rule-based theories have been proposed since then to address the recalcitrant cases. Two of the best known approaches are by Raimy (2000) and Inkelas and Zoll (2005); Frampton's book is now another.
Underlying Frampton's hypothesis is an approach to morphology and phonology in which the derivation of a word consists of a sequence of modifications to an underlying form, the final outcome of which is a surface form. The underlying and surface forms, and each intermediate form, have the same essential nature: a single autosegmental structure representing the word at that point in the derivation. This differs from OT, for example, where the surface form is selected from a large pool of candidate forms. Where Frampton's theory differs from previous derivational approaches is the structure of those intermediate forms, a structure which enables him to account for the cases McCarthy and Prince claimed were outside the domain of derivational theories.
In the next section, I will sketch the outline of the book, while the section after that will assess Frampton's success.
The first chapter introduces the theory of Distributed Reduplication in relatively simple terms. Frampton makes an analogy with the biological process of the transcription of DNA to RNA; like most biological processes, it has turned out to be more complex than it first appeared. The same complexity is true of Frampton's theory; while one can easily follow the derivations in this first chapter, they become more baroque in later chapters. But in its basic form, reduplication happens when the morphology inserts a pair of junctures (boundary markers) into the timing tier of the base, at points determined by reference to the phonemes associated with the nodes in the timing tier. For example, the left-hand juncture might be inserted before the timing tier element associated with the first vowel, and the right-hand juncture might be inserted before the timing tier element associated with the second vowel. The phonology ''responds'' by copying the entire timing tier to the left or right of the base's timing tier, and then removing the junctures; Frampton refers to this process as transcription.
The result of this copying is that reduplicated forms resemble geminates in their structure. But whereas geminates consist of a single phoneme linked to two timing slots, reduplicants consist of a sequence of phonemes each linked to two timing slots. Crucially, this results in crossing association lines, i.e. violations of the No Crossing Constraint (NCC). Frampton assumes that the NCC violations need not be immediately repaired--indeed, they can stay around until the end of the phonological derivation, but they must be resolved before the phonetic interface. That is, the NCC is an interface condition.
Complications arise in reduplication as a result of the fact that some phonological rules apply after the junctures have been introduced, but before transcription, while other phonological rules apply between transcription and the resolution of the NCC condition, or after the NCC violations have been resolved, but before the phonology-phonetics interface.
Chapter two further develops the idea of transcription, and the application of phonological rules to post-transcription, pre-NCC Repair structures. Some of the mysteries of reduplication are said to result from this stage in which reduplicated words resemble geminates, in the sense that there are phonemes which are linked to two timing slots. In particular, under-application arises from a generalization of geminate inalterability: a phonological rule whose structural description is satisfied by the reduplicant but not by the base, or vice versa, fails to apply for the same reason that a phonological rule cannot apply to one half of a geminate. Similarly, over-application is claimed to arise in a rarer situation where the rule is not blocked by the equivalent of geminate inalterability.
Frampton refers to his theory as ''Distributed Reduplication''; the word ''distributed'' refers to Distributed Morphology, a theory originated by Morris Halle, Alec Marantz, and others, which Frampton motivates in chapter three. In particular, he introduces ''readjustment rules,'' i.e. a sort of phonological rule that modifies certain forms in a derivation prior to the application of more ordinary phonological rules, generally to produce allomorphs which are constrained to a limited part of the lexicon. In Distributed Reduplication, the rules which insert junctures into the timing tier are readjustment rules, determining which part of the base is copied.
Chapter four discusses just how reduplicants are copied from the string of timing tier elements between these junctures. The copying process uses an additional pair of junctures, which serve as bookkeeping aids to show which part of the string has already been copied. If this pair of junctures initially contains part of the base, then that stretch will not get copied. By using readjustment rules to position these junctures, junctures can double as a means to truncate the reduplicant to satisfy a prosodic requirement, without affecting the base. The use of transcription junctures inside copying junctures which are in turn inside other transcription junctures can even be used for reduplicants which copy non-adjacent parts of the base. Chapter five continues in this vein, looking at how other sorts of variation in reduplication can be explained by rules which insert transcription junctures, and chapter six shows how the application of phonological rules can alter forms after transcription junctures have been inserted, but before actual transcription. (I will have more to say about this below.)
Chapter seven illustrates how the proposed theory accounts for reduplication in nine languages; the variety of reduplication processes covered is quite impressive. Finally, one appendix argues for the claim that the No Crossing Constraint (NCC) is an interface condition; a second appendix contrasts Frampton's theory with another rule-based theory of reduplication, that of Raimy (2000).
I will first make two general comments, before turning to an evaluation of this work.
First, it is not uncommon in presenting a theory of reduplication to show that the theory is incapable of generating some construct or class of forms that another theory could generate, and claiming that to be an advantage of your theory on the grounds that the construct in question is unattested, and therefore impossible. Frampton makes such claims (e.g. in arguing on pp. 86-88 against Inkelas and Zoll 2000), but he is certainly not the only morphologist to use this argument, and similar arguments have been used in other areas of linguistics. I believe that there are two reasons this kind of argument is a bad one.
First, there is no guarantee that the subset of languages that have thus far been analyzed by linguists exhausts the possible human languages. There are instances of word-level stress systems, for example, which are attested in only a few languages--or depending on the granularity of one's typology, in only one language (Gordon 2002, Bane and Riggle 2008). When a language construct can be that rare, for whatever reason, then the probability that at least some as yet unattested systems actually exist is high.
Second, if one's theory is intended to be a theory of how the human mind ''does'' language, then the lack of a construct in any known language might not mean that the non-existence of the construction is because the human language component cannot deal with it. It might instead mean that there is no evolutionary path to that construction, or that given there are only a finite number of languages in the world, that the probability of a language traversing a path to arrive at that construct is sufficiently small that we have no hope of ever seeing such a language. In other words, the current set of languages does not exhaust the space of ''humanly possible languages.'' This point has of course been made in the context of phonology, most forcefully by Juliette Blevins (2004).
In sum, I do not find arguments from lack of evidence to be persuasive. Indeed, it is quite possible that the kind of reduplication that Frampton argues is impossible in the passage on pp. 86-88 is attested in the Pima language (see Riggle 2006).
My second general comment is that I doubt whether anyone can hope to work though all the derivations in this book with pencil and paper. There are simply too many interacting factors (such as constraints on rules) for this to be done reliably. While in my job as a reviewer, I should have verified the derivations, I gave up after the first few. Verification--ensuring that the results are reproducible--cries out for computational tools. At the same time, having spent a year or two implementing a relatively straightforward morphological parsing engine, I am all too aware that asking for an implementation of an individual theory is nearly hopeless. I note the problem; I do not have a solution.
I now turn to discussion of some particulars.
One controversial claim of Distributed Morphology is the analysis of partly irregular forms like English 'sold' as being due to a readjustment rule that changes 'sell' to 'sol' (or rather, the phonological equivalents of those spellings) in the past tense, followed by the addition of the regular past tense suffix '-d'. An alternative analysis would be that the word 'sold' is simply the suppletive irregular past tense of 'sell.' The evidence from English (and most other languages) is insufficient, but Frampton makes an intriguing argument for readjustment rules from a reduplicative process in Erromangan (Australian Oceanic). He also argues against Inkelas and Zoll's (2000) analysis of reduplication in this language, on the basis that ''massive and transparent regularity in the lexicon'' must be explained by rules, whereas their analysis assumes that the existence of any irregularity precludes explaining semi-regular forms by (synchronic) readjustment rules. While the details are too complex to cover in this review, I will say that I find this second argument less than compelling; semi-regularities arise in diachrony, but need not, in my opinion, find a synchronic explanation. Rather, whether generalizations for which there are exceptions become rules in the mind of the native speaker, or whether they are treated as a memorized set, is something that cannot be proven on the basis of counting the number of forms which do or do not undergo the rule. We will need some other tool to determine whether the mind captures sub-generalizations. (This issue has of course had a long and controversial history in phonology.)
In the context of the discussion of phonological rule application prior to transcription (chapter six), Frampton suggests augmenting phonological rules to allow them to include a derivational constraint which expresses the ''goal'' of the rule. Because the goal applies to the output of the rule, this extension implies a sort of back-tracking in the derivation: a rule is applied, and then the output is tested against the constraint. If the output satisfies the constraint, the derivation continues, else the derivation resumes with the input of the rule, but without applying the rule itself. Alternatively, several different rules might have the same goal of creating a particular syllabic structure, and the different rules are tried until that goal is satisfied, or possibly the rule applies which produces the maximal improvement towards that goal. An extension is the idea of rules that are blocked if their output produces some undesirable structure; this is the opposite of goal driven behavior. I found the discussion of these points unclear, nor was it clear whether all phonological rules are supposed to be goal driven, or only some.
Goal-driven rules appear to be motivated as a means of building the notion of rule ''conspiracies'' into the theory. Personally, I am not convinced that this goal is necessary. The notion of rules which apparently conspire to meet a goal need not, I think, be reflected in how the mind does phonology, any more than the fact that giraffes have long necks need be driven by some kind of goal-seeking behavior in young giraffes. Instead, rules of phonology may have arisen during diachrony with the effect of making surface forms more like other surface forms in the language; but that does not necessarily mean that the goal of conforming to other surface forms need be reflected in synchronic derivations.
Another reviewer has commented that Frampton's theory ''seems to be so powerful that I am left wondering if there is, in fact, anything that it CAN'T do'' (Haugen 2010:956, emphasis in original). Given my comments above that we do not know what sorts of linguistic constructs are impossible, and in particular what kinds of reduplication--if any--might be impossible, I am less troubled by this. However, Haugen's other general criticism is that Distributed Reduplication ''is a theoretical chimera that combines rules and derivations with output constraints ... and evaluations of possible candidates.'' Most theories of reduplication and phonology, it seems to me, have become baroque (even OT has sprouted epicycles since its birth, to deal with opacity). If we truly believe that the human language capacity came about as a small number of evolutionary changes, then something is wrong with theories that require such complexity. Perhaps what we need is a simple but powerful theory of reduplication: one which will inevitably over-generate possible grammars, in particular allowing for unattested kinds of reduplication, but which is a plausible result of evolution. The unattested sorts of reduplication can be ruled out by other means, such as the lack (or rarity) of a diachronic path to produce them in the first place.
It may be that Frampton's analogy with transcription in biology is intended to address this complaint, since it has become clear that the biological process by which DNA is read and converted into proteins is far from simple. If so, I'm afraid the analogy fails. The evolution of DNA transcription and translation has been going on for billions of years in trillions of individual organisms, whereas the evolution of language must have been a relatively rapid process in a group of human ancestors which cannot have numbered more than a few millions, and was probably far smaller. But I may be reading more than is warranted into the fact that Frampton uses the analogy between language and biology.
Despite my misgivings, I will close by saying that this book is a must-read for anyone wanting to understand the breadth of things that go on in reduplication. My guess is that Frampton's cataloging of unusual processes (particularly his last chapter) will outlast the theoretical contributions of this work. And for that reason, if no other, it is worth reading.
Bane, Max, and Jason Riggle. 2008. Three correlates of the typological frequency of quantity-insensitive stress systems. In: Proceedings of the Tenth Meeting of ACL Special Interest Group on Computational Morphology and Phonology. Association for Computational Linguistics.
Blevins, Juliette. 2004. Evolutionary Phonology: the emergence of sound patterns. New York: Cambridge University Press.
Gordon, Matthew. 2002. ''A Factorial Typology of Quantity-Insensitive Stress.'' Natural Language & Linguistic Theory 20 (3): 491-552.
Haugen, Jason D. 2010. [Review of] Distributed Reduplication. Language 86: 953-957.
Inkelas, Sharon, and Cheryl Zoll. 2000. Reduplication as morphological doubling. [Rutgers Optimality Archive 412, http://roa.rutgers.edu]
Inkelas, Sharon, and Cheryl Zoll. 2005. Reduplication: Doubling in morphology. Cambridge Studies in Linguistics, 106. Cambridge University Press.
McCarthy, John J., and Alan S. Prince. 1995. Faithfulness and reduplicative identity. In Jill Beckman, Laura Walsh Dickey & Suzanne Urbanczyk (eds.), Papers in Optimality Theory. University of Massachusetts Occasional Papers 18. Amherst, Mass.: Graduate Linguistic Student Association. pp. 249–384. [Rutgers Optimality Archive 60, http://roa.rutgers.edu]
Raimy, Eric. 2000. The morphology and morphophonology of reduplication. Berlin: Mouton de Gruyter.
Riggle, Jason. 2006. Infixing reduplication in Pima and its theoretical consequences. Natural Language and Linguistic Theory 24 (3): 857-891.
ABOUT THE REVIEWER
ABOUT THE REVIEWER:
Dr. Maxwell is a researcher in grammar description and other computational
resources for low density languages, at the Center for Advanced Study of
Language at the University of Maryland. He has also worked on endangered
languages of Ecuador and Colombia, with the Summer Institute of
Linguistics, and on low density languages with the Linguistic Data
Consortium (LDC) of the University of Pennsylvania.