"Kissine offers a new theory of speech acts which is philosophically sophisticated and builds on work in cognitive science, formal semantics, and linguistic typology. This highly readable, brilliant essay is a major contribution to the field."
Although the subtitle of ''Beyond Morphology'' suggests that this book is mainly about morphology, the title itself represents the contents better. While the data in the book are indeed primarily word formation phenomena, the theoretical coverage does indeed go much beyond that. In essence, Ackema and Neeleman (A&N) argue for a specific grammar model, and then show how this grammar model can be used to account for a wide range of data.
The book is divided into seven chapters. In the first chapter, ''Morphology and Modularity'', A&N outline the grammar model that they assume, and how morphology has its place in it. The second chapter, ''Arguments for Word Syntax'', discusses the question whether morphology is part of syntax or whether there is a separate morphological module with its own principles. Having reached the conclusion that ''word syntax'' is indeed separate from phrasal syntax, chapters three and four (''Competition between Syntax and Morphology'' and ''Generalized Insertion'', respectively) discuss how syntax and morphology interact.
These first four chapters in a way form the theoretical part of the book. Although much data has already been discussed, it is mostly chapters five ''Distributed Selection'', six ''Context-sensitive Spell-Out and Adjacency'' and seven ''PF Feature Checking'' that show how the model can account for word formation data.
The grammar model presented in the book is very attractive, and A&N cover an impressive range of data. The analysis is well thought through, carefully executed and often convincing. I believe it should prove a worthwhile read for anyone interested in morphology, but also for those interested in theoretical linguistics, and for syntacticians interested in knowing more about the relation of the syntactic module to other modules of grammar.
OVERVIEW
Chapter One Morphological data being the main focus of the book, A&N start out with the question what the place of morphology in the grammar is. Perhaps surprisingly, they argue that there is no separate morphology module in the grammar. Instead, they propose a grammar model that consists of three so-called macromodules: semantics, syntax and phonology (cf. Jackendoff 1997). Each of these macromodules has two submodules: a generative system for phrasal structures and a generative system for word structures. The syntax macromodule therefore has a module for phrasal syntax and one for word syntax. (A&N adopt the habit of calling the former ''syntax'' and the latter ''morphology'', and I will do the same, though it should be kept in mind that they are just convenient shorthands.)
The central point of A&N is that the interaction between different modules and the mapping between the outputs of different modules have a large impact on linguistic forms. Most of the book is therefore dedicated to exploring the ways in which this interaction and mapping takes place.
Chapter Two Chapter two argues extensively for the point that syntax and morphology are separate from each other, and against a view of morphology where words are formed in syntax, for example through head-movement. The main point is basically that any syntactic account of morphology must assume extensions to the syntactic theory in order to account for the different behaviour of word formation processes and phrase formation processes. Such a qualitative extension to syntax, A&N argue, is tantamount to separating morphology and syntax. Furthermore, if morphology were syntactic, one would expect that one could condition the other. That is, one would expect rules such as 'compound verbs are right-headed when they occur in C, but left- headed when they occur in situ', or 'the object is placed after the verb when it is underived, but before it when it is morphologically complex'.
Syntax and morphology being separate modules, A&N then pose the question how these two modules interact. They argue that there are two forms of interaction between the two: competition and insertion. Chapter three discusses competition, while insertion is the topic of chapter four.
Chapter Three In A&N's model, when the syntactic macromodule combines two elements, it can do so both in syntax and in morphology. Which of the two is actually used, is determined by competition. In principle, all else being equal, syntax wins out, at least in the languages under discussion (mainly English and Dutch; A&N suggest that in polysynthetic languages, the situation may be reversed, morphology being the default choice). Therefore, when a verb combines with an object, the structure that will be built will be a syntactic one (1a), rather than a morphological one (1b):
(1) a. [_VP [_V drive] [_N trucks]] b. [_V truck drive]
However, syntax will only win when there are no special further requirements. Specific requirements imposed by elements in the structure may lead to morphology being preferred. This is for example the case when a V is merged with an object, and the resuling V+N structure is to be embedded under an affix. Since (the morphological form of) an affix requires a word to attach to, the V+N structure cannot be merged in syntax but must instead be merged in morphology. This derives for example so-called synthetic compounds such as 'truck driver'. Here, it is the affix -er that requires a word, and as a result, 'drive' and 'truck' are merged in morphology, not in syntax. Thus, competition accounts for the fact that although English does not have verbal compounds with incorporated objects (*to truck-drive), it does have derived nouns with incorporated objects (truck driver).
Chapter Four As mentioned, chapter four deals with another kind of interaction between morphology and syntax, insertion. The term ''insertion'' is not to be understood in its traditional sense of taking an element and using it as a building block in a larger structure. Rather, A&N argue, insertion is to be seen as a form of feature matching. Features in a node in one structure, either syntactic or morphological, are matched to features of a node in another structure, also either syntactic or morphological. When the host structure is eventually spelled out, the morphological material associated with the inserted material is placed in the linear string at the corresponding position.
Other than that, however, the inserted structure is not visible for the host structure. This derives, among other things, the lexical integrity effect. But insertion is not limited to lexical insertion. A&N argue that insertion is not sensitive to the types of structures that are matched, nor to the nodes that are matched. Hence it is possible to insert a phrase into a morphological terminal position (deriving such complex words as '[animal to human] transplant experiments', '[go anywhere anytime] access', etc.), and also to insert phrases into non-terminal positions of other phrases, which, A&N argue, is what parentheticals are. (Note that because insertion is not really insertion but rather feature matching, this proces does not derive a ternary branching node.)
Chapter Five Chapters three and four deal with so-called intramodular interaction, i.e., interaction between the two submodules of the syntactic macromodule. Chapter five deals with so-called intermodular interaction, specifically between the syntactic and the phonological macromodules. As stated above, in a modular grammar model such as the one A&N develop, there is no single module responsible for morphology. Rather, there are several modules that deal with aspects of morphology, i.e., the syntactic, the semantic and the phonological word-level submodules. Therefore, such a typically morphological phenomenon as selection must also be distributed over these three modules. Structures generated in the three submodules must then be mapped onto each other.
This mapping is partially idiosyncratic, in that it simply states that a semantic predicate such as READ is mapped onto a syntactic structure [<+V,-N>, (Th1, Th2)] and onto a morphological structure /ri:d/. At the same time, however, there are general mapping principles that govern the realization of complex words. A&N recognize three general mapping principles: Linear Correspondence, Input Correspondence, and Quantificational Correspondence (note that AFFIX denotes a morphosyntactic element, while /affix/ denotes its corresponding morphophonological form):
(2) Linear Correspondence: If X is structurally external to Y, then /x/ is linearly external to /y/. (3) Input Correspondence: If an AFFIX selects (a category headed by) X, then /affix/ takes /x/ as its host (4) Quantitative Correspondence: No element in the morphosyntax is spelled out more than once.
The distinction between an AFFIX and an /affix/ is of course a direct consequence of the assumption that morphosyntax and morphophonology are submodules of different macromodules. Note that it is not necessary that one AFFIX is always mapped onto the same /affix/ whenever it occurs. For example, the English agentive suffix ER is normally mapped onto the suffix /-er/, but it can idiosyncratically be spelled out differently:
(5) DRIVE + ER <--> /drive/ /er/ ('driver') (6) TYPE + ER <--> /type/ /ist/ ('typist')
The form 'typist' is accounted for by assuming that there is an idiosyncratic mapping rule that maps TYPE+ER onto /type/+/ist/, rather than /type/+/er/.
Furthermore, the mapping of the suffix ER can be changed when one of the three general mapping principles would otherwise be violated. For example, in the morphosyntactic structure [[STAND IN] ER], which derives the agentive noun from the particle verb '(to) stand in', spelling out ER as /er/ would violate one of the mapping principles. In theory, there are three ways to realize [[STAND IN] ER] morphophonologically:
(7) a. /stander in/ b. /stand inner/ c. /stander inner/
(7a) would violate Linear Correspondence, because ER is external to the particle IN. (7b) would violate Input Correspondence, because ER selects for a category [+V,-N], and hence should not attach to the morphophonological spell-out of a [-V,-N] category. Finally, (7c) violates Quantitative Correspondence, because ER is spelled out twice.
The only way to realize [[STAND IN] ER] phonologically without violating one of the mapping principles is to spell out ER as a phonologically null element, on the assumption that the mapping principles refer to overt material, and that covert phonological material will satisfy them vacuously. Hence, the mapping principles can be adhered to when there is an idiosyncratic mapping rule that maps ER to // when it takes [STAND IN] as complement, yielding the phonological form /stand in/.
Traditionally, affixes are thought to attach to words. This is a morphophonological selection requirement, however, and therefore only holds in the morphophonological submodule, i.e., for /affix/es. There is no reason to assume that it also holds in the morphosyntactic module. Rather, AFFIXes have another selectional restriction: they require a host of a specific category. There is no restriction on the projection level of this host, however.
In other words, AFFIXes can attach to any syntactic structure of the right category. It is irrelevant whether this structure is a head or a phrase. (Though it is possible that additional requirements restrict this freedom; e.g., an affix such as ER satisfies the external theta role of a verbal predicate, and can therefore not attach to verbal projections in which the external role has already been assigned.)
This assumption, virtually inescapable in A&N's model, allows for a straight-forward account of so-called mixed categories. Mixed categories are for example deverbal nouns that retain verbal characteristics internally. Examples are nominal infinitives in Dutch, and gerunds in English. It is well-known that these categories can occur in constructions in which they assign accusative case to their objects, and be modified by adverbs. However, they can also occur in constructions in which objects are licensed through a preposition, and in which only adjectives can modified them:
(8) a. John's constant singing of the Marseillaise b. John's constantly singing the Marseillaise
The idea is that the nominalizing suffix simply attaches to projections of different levels. (Essentially an insight developed by Abney 1987.) Note that this nominalizing /suffix/ is not /ing/, because /ing/ does not derive nominal categories: it also derives participles, which are not nominal. Therefore, the actual nominalizing /suffix/ must be a null /suffix/.
This assumption is essential for A&N, because if /ing/ were the nominalizing /affix/, Linear Correspondence would be violated: the object '(of) the Marseillaise' is structurally internal to the AFFIX, and hence the /affix/ must be linearly external to it, which /ing/ in (8) is not.
This means that languages in which the nominalizing /affix/ is overt, mixed categories are not possible. This seems indeed to be the case for Norwegian, which has an affix /ing/ which, unlike its English homophone, does not derive any non-nominal categories, and can therefore be assumed to be the nominalizing affix. Interestingly, a verbal structure similar to (8b) is not possible with this affix:
(9) den ulovlige kopieringen *(av) populaere sanger that illegal-DEF copying-DEF *(of) popular-PL songs-PL
That is, Norwegian has no way to map [[COPY POPULAR SONGS] ING] to a morphophonological structure, because attaching /ing/ to /kopier-/ violates Linear Correspondence: [POPULAR SONGS] is structurally internal to ING, but in /kopieringen populaere sanger/ it would be linearly external to it. Therefore, Norwegian must attach ING at a lower level, yielding a morphosyntactic structure of [_N COPY ING], forming a nominal category, which can only license an object with a marker such as OF.
Another way to maintain Linear Correspondence is to have a head- final VP to which the /suffix/ attaches to. This is seen in one Quechua nominalizing /affix/, /sqa/. In main clauses, Quechua allows OV and VO, but when a VP is nominalized with /sqa/, the VP must obligatorily be OV. This creates a syntactic structure [[OBJ V] SQA], which can be mapped onto a phonological structure /obj/ /v/ /sqa/. The corresponding VO structure [[V OBJ] SQA] would have to be mapped onto /v/ /obj/ /sqa/, which again violates Linear Correspondence.
Chapters Six and Seven In chapters six and seven A&N discuss another kind of interaction between syntax and phonology. There are many phenomena in syntax that are subject to adjacency. Typical examples are case assignment in e.g. English, agreement weakening under subject-verb inversion in Dutch and Standard Arabic, certain types of cliticization and pro-drop, complementizer agreement, that-trace effects, etc. A&N argue that such phenomena are best analyzed as PF phenomena, subject to rules that are sensitive to phonological rather than syntactic domains. (A&N base themselves on Selkirk 1986 and similar work for defining the phonological domains they assume.)
Two types of processes can take place in phonological domains, A&N argue. One is feature suppression, where features of an element are reduced when it appears in a phonological domain with an agreeing element that has full feature specification, and feature checking, where features on an element are checked against identical features on another element in the same phonological phrase.
As an example of feature weakening, take agreement weakening in Arabic. It is well-known that Arabic has both SVO and VSO structures, and that only SVO shows full agreement. VSO, in contrast, only shows gender agreement, but not number agreement. Many attempts have been made to analyze this phenomenon syntactically, but none has really been satisfying. A&N argue that the phenomenon should not be seen as purely syntactic. Instead, in VSO structures, the verb and the subject form a single phonological domain (an observation known from the literature) and this configuration allows feature weakening on the verb at PF ({} indicate phonological phrases):
So if a verb with a plural feature appears in the same phonological phrase as its subject, the plural feature of the verb may be elided.
PF feature checking, A&N argue, can account for the phenomenon of complementizer agreement seen in certain Germanic languages:
(11) Hellendoorn Dutch dat-te wij speul-t that-1pl we play-1pl
Here, A&N argue, agreement between the complementizer 'dat' and the subject 'wij' takes place at PF, and is made possible by the fact that the complementizer and the subject are in the same phonological phrase.
DISCUSSION
It must be said that the short description given above hardly does justice to the analysis that A&N propose. They discuss all of the phenomena mentioned here in much more detail, and many others that I haven't mentioned. The breadth of the data covered is impressive, and on the basis of their model, A&N offer many different and new solutions to well-known problems in morphology. Some of their solutions are truly novel, some are reinterpretations of existing solutions, but invariably they offer interesting new perspectives, new ways of thinking about about them.
This is of course not to say that there are no comments to be made. I will first mention some empirical problems, and then go on to discuss some aspects of the theoretical model.
First, one of the arguments given to support the idea that syntax and morphology are not a single component of the grammar is that the two do not condition each other. As mentioned above, there are no rules such as 'compound verbs are right-headed when they occur in C, but left-headed when they occur in situ', or 'the object is placed after the verb when it is underived, but before it when it is morphologically complex'.
While generally true, there are possible exceptions. In Dutch, most (though not all) preposition do not allow non-human pronouns as their complement. When a preposition takes a non-human pronoun as complement, the pronoun must raise, attaching as a proclitic to the pronoun:
(12) a. *in het -- er-in in it -- there-in
(12) b. *op dit -- hier-op on this -- here-on
This happens with personal pronouns, but equally with demonstrative, interrogative, and relative pronouns. The crucial point is that when this happens, the pronoun changes form: it takes what might be called an r- form (due to the fact that all these forms end in -r). One preposition, 'met' (with) even changes form itself, to 'mee':
(13) ?met het -- er-mee /*er-met with it -- there-with
Prepositions that do not require (or even allow) the complement to be extracted, or prepositions that (marginally; they improve with demonstratives) allow the complement to stay in situ, require the original, non-r-form of the pronoun:
(14) a. zonder dat -- *daar-zonder -- *zonder daar b. met dat -- daar-mee -- *met daar
The change is a purely morphological one, in that the form change does not entail semantic change, and seems conditioned by pure syntax, in that the r-form occurs when the complement is extracted. Note that this morphological change cannot be accounted for in terms of some PF adjacency rule as discussed in chapters six and seven, because the r-pronoun can be extracted further (e.g. to the CP domain), stranding the preposition. The phenomenon, therefore, seems to contradict A&N's assumption that morphology and syntax do not condition each other.
Other empirical problems are found with Standard Arabic. A&N argue that agreement weakening in Arabic (the fact that VSO structures do not allow plural marking on the verb) is PF conditioned, because it takes place under adjacency. As A&N note themselves in a footnote on page 205, this is in fact not entirely true. Agreement weakening also takes place in VOS structures, which are rare but grammatical in Arabic. A&N tentatively assume that in VOS structures the weak agreement has another source, but that is hardly a satisfying situation. A uniform analysis of the phenomenon would seem to require a syntactic analysis after all.
Arabic also poses a problem for the analysis of mixed categories, discussed above. Arabic has deverbal noun structures headed by a nominal form, the so-called masdar (see, e.g. Fassi Fehri 1993, Kremers 2003). Like English gerunds and Dutch nominal infinitives, masdars can have an internal verb-like structure, assigning accusative case and licensing adverbials, but can also appear in purely nominal constructions, where the object is licensed through a preposition and adjectives modify the masdar.
The problem is that the masdar form is a purely nominal form, (it is not a form that can also have non-nominal meanings, such as English ing- forms), and more importantly, that it is formed non-concatinatively. Depending on the verbal base, the masdar form can be irregular or regular, but it is *always* formed by inserting a vocalic pattern into a consonantal stem. E.g., in the masdar form '(i)ntiqaad' (criticizing), the verbal consonantal stem is N-T-Q-D, while the vocalic pattern is CCiCaaC.
It is not clear how A&N would deal with such non-concatinative morphology. The fact that Arabic masdars are nominal forms suggests that the masdar affix is itself the nominalizer. It must be the case, then, that the mapping principles are met, but it is not clear how, exactly. Especially Linear Correspondence seems problematic.
Nonetheless, in my opinion, the success in accounting for a large set of varied data far outweighs the empirical problems that do still exist. Still, a few points about the theory itself can be made.
First, the argument for a modular grammar is very strong. It has already been clear from other work (Jackendoff 1997, for example) that we do indeed need separate modules for semantics, phonology and syntax, that these generate structures independently and that these structures must be mapped onto each other. Yet, mapping is not the only interaction that A&N allow in their theory. Insertion, the matching of features in one structure with those in another, is also a form of interaction between modules. It must be said that given the grammar model, assuming insertion is inevitable, but I believe it does raise some questions.
The same, I believe, is true of kind of PF rules that A&N propose. They appear to require a qualitative addition to the theory that is not clearly empirically motivated.
Taking insertion first, A&N argue that insertion is not sensitive to the kinds of structures it applies to. Hence, it can insert morphological material in syntactic structures, but also syntactic material into morphological structures. At the same time, insertion is not sensitive to which nodes of the insertee or the host are involved.
That is, under a traditional view of insertion, what happens is that the *root* node of a structure A is inserted into a *terminal* node of structure B. But in A&N's view, insertion is no more than feature match, and therefore not restricted to root and terminal nodes. It is equally possible to associate the root node of e.g. a phrase with a non- terminal node of another phrase (which accounts for parentheticals, as A&N argue), or the non-root node of a phrase to a terminal node of another phrase (which, A&N suggest in a footnote, may account for graft structures (van Riemsdijk 2000, 2001), and similarly, of a morphological structure into non-terminal nodes of another morphological structure, which may account for ''expletive insertion'' such as 'un-fucking-believable' (McCarthy 1982), a suggestion that A&N somewhat optimistically introduce with the remark that ''[t]here should in fact be a fourth type of insertion (...)'' (p. 131 fn.)
Things are not so easy, however. There are four variables: 1) the insertee can either be morphological or syntactic; 2) the host can either be morphological or syntactic; 3) the inserted node can be a root or not; 4) the target node can be terminal or not. Given that there are four binary variables, we expect not four but sixteen possible forms of insertion. A&N only discuss two (plus what they call 'simultaneous insertion', in which multiple insertions take place at the same time, thus accounting for syntactic idioms), and mention two others in footnotes. Most of the other theoretically possible types of insertion, however, are not mentioned at all. Yet, A&N's model predicts that they can occur. The question then becomes, do they occur, and if not, why not?
A fifth type, insertion of the root a syntactic structure into a syntactic terminal, is mentioned, but A&N argue that this yields a structure that will always have an equivalent identical structure that is formed through merge alone, and therefore cannot be distinguished. However, an inserted structure is not visible in the host structure, while structures formed by merge are expected to be fully transparent. One would, therefore, expect empirical differences that would allow us to distinguish the two types. In fact, it might be interesting to see if such an account is feasible for the so-called Condition on Extraction Domains effects (Huang 1982), or whether phase theory could be linked to it.
Secondly, in their chapters on PF rules (six and seven), A&N first present their analysis as a derivation from syntax to phonology, a process during which a certain mixed syntactic/phonological representation is formed that they call PF. However, in a modular theory, there shouldn't be such a one-way derivation. Rather, we expect mapping principles that are non-directional (or bi-directional). A&N notice this, of course, and argue that PF should be seen as an independently built representation, linked to syntax and phonology through mapping principles.
This solution raises a few questions, however. If PF is a separate representation, just like the semantic, syntactic and phonological representations of a clause, it should be a macromodule, on a par with semantics, syntax and phonology. That would mean that it should have two submodules as well, one for word-level PF and one for phrasal PF. Yet, in their final representation of the grammar module, A&N do not present it as such. Rather, they give PF a box below the three macromodules, rather than next to them, without submodules but with mapping principles linking it to all three. (In fact, the lexicon is represented in exactly the same way, except that it is placed above the three macromodules.)
So, what is the exact nature of the PF representation? It is obvious from the discussion that it is partly phonological, in that it represents ''initial prosodic phrasing''. At the same time, it has syntactic features, because the rules operating at PF are sensitive to agreement and to traces. It is a representation that is halfway between syntax and phonology, combining features of both of them.
I also miss an empirical motivation for PF, in the sense that we cannot directly observe the representation it builds. The semantic, syntactic and phonological representations of a phrase are observable in linguistic utterances, and it is to a large extent possible to research each of them independently from the others. For PF, however, this is not the case.
The question then is, do we really need a PF representation? Wouldn't it be better to eliminate it, and describe its properties as part of the mapping from syntax to phonology?
Obviously, I have no direct answer to these questions myself. The fact that the book raises such questions in the reader, is a positive point: A&N present a very well thought-out theory, which I believe will prove a fruitful base for further research.
REFERENCES
Fassi Fehri, A. 1993. Issues in the structure of Arabic clauses and words. Dordrecht: Kluwer Academic Publishers
Huang, C-T. J. 1982. Logical Relations in Chinese and the Theory of Grammar. MIT Phd Dissertation
Jackendoff, R. 1997. The Architecture of the Language Faculty. Cambridge, MA: MIT Press
Kremers, J. 2003. The Arabic Noun Phrase: a Minimalist Perspective. PhD Dissertation, University of Nijmegen.
McCarthy, J. 1982. 'Prosodic Structure and Expletive Infixation'. Language 58: 574-590
Riemsdijk, H. van, 2000. 'Free Relatives Inside Out: Transparent Free Relatives as Grafts'. In: B. Rozwadowska (ed.) Proceedings of the 1999 PASE Conference. University of Wroclaw, 223-233.
Riemsdijk, H. van, 2001. 'A Far from Simple Matter'. In: I. Kenesei and R. Harnish (eds). Perspectives on Semantics, Pragmatics and Discourse. Amsterdam: John Benjamins, 21-41.
Selkirk, E. 1986. 'On Derived Domains in Sentence Phonology'. Phonology Yearbook 3: 371-405.
ABOUT THE REVIEWER:
ABOUT THE REVIEWER
Joost Kremers studied Arabic and linguistics at the University of
Nijmegen, The Netherlands, and received his PhD there in 2003
(thesis title: "The Arabic noun phrase: a minimalist analysis). Apart
from Arabic syntax, his main research focus is linearisation. His central
idea is that linear structure is not determined in narrow syntax. Rather,
it is the modality in which language is expressed (speech, sign) that
imposes linear order. Hence, linearisation should be part of the
derivation or mapping from syntax to phonology. Future research will
focus more on synchronicity, i.e. the mutual spell-out of multiple
syntactic elements. He is currently a post-doc at the Graduate
College "Sentence Types" of the University of Frankfurt, Germany.
