Semantico-Phonetic Form: A Unitarianist Grammar Ahmad Reza Lotfi, Ph. D. Department of English Language Azad University at Khorasgan, Esfahan, IRAN. E-mail: lotfi@www.dci.co.ir ABSTRACT Semantico-Phonetic Form is a unitarianist theory of language in two different but inter-related senses: first, it assumes that the Concep- tual-Intentional and Articulatory-Perceptual systems (responsible for semantic and phonetic interpretations respectively) access the data at one and the same level of interpretation; hence a single interface level--Semantico-Phonetic Form, SPF. Second, it is unitarianist in that (although it is still a formalist theory of language) it potentially permits the incorporation of both formalist and functionalist explana- tions in its formulation of the architecture of language. Within the framework of Semantico-Phonetic Form, and as an alternative proposal to Chomsky's minimalist thesis of movement, the Pooled Features Hypothesis proposes that "movement" is the consequence of the way in which the language faculty is organised (rather than a simple "imper- fection" of language). The computational system CHL for human language is considered to be economical in its selection of formal features from the lexicon so that if two LIs (to be introduced in the same derivation) happen to have some identical formal feature in common, the feature is selected only once but shared by the syntactic objects in the deriva- tion. It follows that the objects in question must be as local in their relations as possible. The locality of relations as such, which is due to economy considerations, results in some kind of (bare) phrase struc- ture with pooled features labelling the structural tree nodes that dominate the syntactic objects. Pooled features, in a sense, are structurally interpreted. Other features, i.e. those not pooled, will be interpreted at SPF. KEY WORDS: bare phrase structure, economy, faculty of language, feature checking, feature sharing, formal features, imperfections, lexicon, logical forms, minimalist syntax, Semantico-Phonetic Form, strength, unitarianist theory 1. Introduction 1. Semantico-Phonetic Form is a unitarianist theory of language in two different but inter-related senses: first, it assumes that the Concep- tual-Intentional and Articulatory-Perceptual systems (responsible for semantic and phonetic interpretations respectively) access the data at one and the same level of interpretation; hence a single interface level--Semantico-Phonetic Form, SPF. Second, it is unitarianist in that (although it is still a formalist theory of language) it potentially permits the incorporation of both formalist and functionalist explana- tions in its formulation of the architecture of language. 2. The paper is organised as follows. Section 2 reviews orthodox mini- malist accounts of interface levels, and introduces a unitarianist theory of the interface between competence and performance systems. Section 3 examines the limitations of a theory of language that con- fines itself to economy-oriented explanations of language. Instead, it offers a theory of language that favours economy on a par with distinctness. Sections 4 and 5 introduce the Pooled Features Hypo- thesis as a unitarianist alternative to Chomsky's minimalist thesis of movement. Section 6 concludes the paper with some final comments on the relationship between feature sharing and Semantico-Phonetic Form. 2. The Unitarianist Interface Hypothesis 2.1 Logical Forms: The Background 3. Orthodox Minimalists assume that the faculty of language FL con- sists of a cognitive system (a computational system and a lexicon) responsible for storing information, and performance systems (the "external" systems A-P and C-I interacting with the cognitive system at two interface levels of PF and LF respectively) responsible for using and accessing information. In accord with the requirements of conceptual simplicity, it is further assumed that "there is a single computational system CHL for human language and only limited lexical variety" (Chomsky, 1995:7). Being a theory of Universal Grammar (UG), the minimalist programme considers a structural description (SD) to be "the optimal realizations of the interface conditions, where 'optimality' is determined by the economy of UG" (Chomsky, 1995:171). Since the programme does not assume the existence of any conditions (such as the Projection Principle) relating lexical properties and interface levels (p. 220), one may conclude that (for Chomsky) the economy of UG can be best viewed as a function of such operations of the computational system CHL as Merge, Move, and Agree rather than other operations of the cognitive system, such as LI/FF selection for the lexical array LA, or other components of the system like the lexicon. 4. The introduction of Logical Form--a "minimalist" interface level where semantic interpretation takes place--goes back to the late 70s with Revised Extended Theory as the dominant version of Transformational Grammar. However, REST considered shallow structure--the level of syn- tactic representation following the application of all transformational rules except filters and deletion--to be the input to the semantic rules. For May (1991), Logical Form was the representation of the form of the logical terms, or the expressions with invariant meanings, of a language. As a result, "on this view the syntax of natural language does have a logical form, in that at LF it represents the structure required for the application of semantic rules for logical terms" (May, 1991: 55). 5. The operations Wh-Movement and Quantifier Raising were assumed to derive LF from S-structure (in GB). In other words, such a designated level of syntactic representation was motivated by some theory-internal considerations such as the principles of the Binding Theory of the time. "Indeed, if the Binding Theory could be shown to require the particular articulation of structure found just at LF for its full application, this would constitute a sort of 'existence proof' for LF, and the devices employed in deriving it" (May, 1991:339). Typical empirical support for such "invisible" LF operations were sentences with quantifiers like (22) in May (1991), reproduced here as (1a), with a structure satisfying Principle A ONLY AFTER the application of QR at LF (1b) so that "both 'the women' and 'the men' locally c-command an occurrence of 'each other' ": (1) a. The men introduced each other to everyone that the women did. b. [everyone that the women [VP introduced each other to e-i]]i [the men introduced each other to e-i] In other words, instead of revising the (binding) theory to make it more compatible with such empirical data, it was decided to change (the designated architecture of) language to fit in. 6. Instead, they could designate an elliptical VP with a visible move- ment of 'everyone' from the ellipsis site and the deletion of elements recoverable from the linguistic context without the unnecessary formal step of "copying" the WHOLE visible VP into the ellipsis site: [1] (2) The men-j introduced each other-j to everyone-i that the women-k did [VPellip introduce each other-k to e-i] Now one of the *each other*s c-commands 'everyone' while the other its trace. Then QR, and as a result, LF, would be dispensed with. But the purported formal restrictions inherent in any logical system (natural languages included) did not allow the move: (2) was simply illogical as copying would blindly proceed with the whole VP 'introduced each other to everyone that the women did' copied into the site, naturally with another elliptical VP created in the embedded clause, and this would proceed ad infinitum.[2] 7. The GB solution--LF as the exclusive input to semantic rules, which has been carried over to the minimalist era as the C-I interface level--suffers both conceptual and empirical shortcomings.[3] Firstly, one can still keep logical operations operating in natural languages without assuming language to be nothing but a perfect logical system. To be more specific, it is not conceptually necessary for a phonetically realized VP to be fully copied into an elliptical one. A partial copying will do as long as the semantic system (whatever it is) can assign a plausible interpretation to the structure. Interest- ingly, even May's formulation of LF cannot fully avoid this logical trap: even in (1b), the cyclicity shows up as e-i is co-indexed with the phrase [everyone that the women [VP introduced each other e-i]]i in which it occurs. May's formulation, however, is at a disadvantage as it is conceptually more complicated than (2). Secondly, LF as specified solves one empirical problem for the Binding Theory but creates some other. If QR saves binding principles for sentences like (1a), it actually refutes them for some other sentences that are ungrammatical in terms of binding prior to the movement of the quantifier to the left-most position of the sentence at LF but fulfill the binding requirements after QR: (3) * a. She-i met everyone that Mary-i knew. b. [everyone that Mary [VP knew e-i]]i [she met e-i] 8. In (3b), 'Mary' is outside the c-command domain of 'she'. Then 'Mary' can antecede 'she' with no violation of binding requirements. The pre- diction proves to be empirically false. Based on similar cases, Chomsky (1981) concluded that Principle C was satisfied at s-structure, a solu- tion is not available in MP with S-structure dispensed with.[4] 9. LF made it possible to afford such syntactic operations as Procras- tinate, Attract, and covert raising in order for UG principles to remain maximally generalizable, and to downgrade cross-linguistic variation. Perhaps there is nothing wrong a priori with such a theoretical device. However, and despite the purported theoretical elegance due to the introduction of logical forms, it is a pity that the generativist has not been interested in the question of whether it is possible in the real world for LF (so vastly divorced from the phonetic reality of sentences, i.e. PF) to be the interface level at which *real speakers* semantically interpret whatever they hear (or whatever they don't). The issue is presumably dismissed as a matter of performance, one that the generativist has traditionally found an irre- levant question to ask. More on this below. 2.2 Logical Form as the I-C Interface Level 10. Chomsky (1995) takes a particular language L to be a procedure of constructing pairs (pi, lambda) out of LIs selected from the lexicon into a lexical array/numeration to be introduced into the derivation by the computational system. The operation Spell-Out strips away pi elements from the structure sigma to be mapped to pi by the phonological component of the computational system and leaves the residue sigma-L for its covert component to map to lambda so that they are interpreted at the A-P and C-I interfaces respectively as "instructions" to the rele- vant performance systems. If they consist entirely of interpretable objects, i.e. those that are legible for the external systems, the derivation D converges as it satisfies the condition of Full Interpreta- tion. "A derivation converges at one of the interface levels if it yields a representation satisfying FI at this level, and converges if it converges at both interface levels, PF and LF; otherwise, it crashes" (1995:219-220). In case there are more than one convergent derivation possible, the most economical one blocks all others. Uriagereka (1998) diagrams the procedure as follows: Lexicon | V A | V Merge and Move | V Spell-Out PF component / \ LF component / \ PF LF level level / \ / \ V V A/P I/C component component Figure 1. Uriagereka's formulation of the MP model (from Uriagereka, 1998:536) 11. It is not clear, however, whether Chomsky meant the model to be one for the speaker, for the listener, or both. Chomsky talks about "instruc- tions" sent to performance systems. This means the procedure is what the speaker goes through in order to articulate a sequence of sounds.[5] But the model cannot represent what the speaker does because this means the speaker first selects LIs, has the structure derived with all its complexity via the application of Select, Merge, Agree, Move, etc, strips sound from meaning (Spell-Out), and then and only then "understands" what she said means. On the other hand, only a miracle can make the model work for the listener. The only pieces of informa- tion she receives are those legible to her A-P system. What happened to sigma-L? How can she access the information that does not travel through the air? Is she supposed to reconstruct the derivation at her own LF and in reference to the PF information she receives?[6] How likely is it to take place given the mismatches between LF and PF? If such mismatches are trivial enough to make her reconstruction of LF possible, why should one hypothesise the existence of Logical Form in the first place?[7] If the computational system can distinguish pi elements from lambda ones (spell-Out), why isn't it possible for the performance systems them- selves to do so while accessing a single interface (instead of two) representing both types of information? Accidentally, Chomsky distin- guishes formal features from semantic ones (1995:230) although both are presumably interpreted at LF. If the performance systems responsible for semantic interpretation can access and interpret features as different as formal and semantic features, why is it necessary to strip away phonological features after all? Couldn't the performance systems access and interpret all these features--phonological, semantic, and formal--at a single interface level? Such a level is not only concep- tually possible but also more desirable in terms of economy considera- tions. Chomsky would remind us that this is an empirical question: given any empirical proposal, we can always ask why things should work that way rather than some other way. Thus why do mammals have two eyes in- stead of three (an eye in the back of the head would be very useful for escaping predators)? But what is the empirical support (apart from the dubious "existence proof" for LF reviewed earlier in the paper) for the split? On the other hand, there is some empirical evidence to suggest that such a split does not substantiate (see 2.3 below for details). 12. As mentioned earlier in paragraph 9, such questions are usually dismissed as aspects of performance, which is basically irrelevant to what the generative grammarian does. For Chomsky, "[l]inguistic theory is concerned with an ideal speaker-listener in a completely homogeneous speech community [...]" (1965:3). This formulation of grammar (I think still correctly) excludes "such grammatically irrelevant conditions as memory limitations, distortions, shifts of attention and interest, and errors [...] in applying [...] knowledge of the language in actual per- formance" (1996:3). He further asserts that "[a] generative grammar is not a model for a speaker or a hearer" but one that "attempts to charac- terize in the most neutral possible terms the knowledge of the language that provides the basis for actual use of language by a speaker-hearer (1965:9)." Thus the differences between the speaker and the hearer (those between production and understanding of language respectively) are dismissed as matters of performance. 13. In practice, however, this neutrality of terms has proved to be unattainable. Chomsky takes " L to be a generative procedure that constructs pairs (pi and lambda) that are interpreted at the articula- tory-perceptual (A-P) and conceptual-intentional(C-I) interfaces, respectively, as 'instructions' to the performance systems" (1995:219). Moreover, he thinks of the computational system "as mapping some arrays A of lexical choices to the pair (pi, lambda)" (p.225). Apparently, lexical selection (among many other operations of the system such as Merge, Move, Delete, Form Chain, etc) is what only the ideal *speaker* rather than the listener can afford. On the other hand, interpreting the linguistic expressions of L (already computed by "the ideal speaker") at LF is what only a listener is liable to indulge in. Chomsky's account of competence is then concerned with "the ideal speaker-listener" in the sense of SOMETIMES being a model of the ideal speaker, and SOME OTHER TIMES that of the ideal listener. It can hardly be a model of the abstract idealisation of linguistic knowledge divorced from all aspects of language production and perception; a "speaker- and hearer-neutral" description of what they both have in common as the basis for their actual use of language. 14. Chomsky shows his concern for performance systems when he writes, "[f]or each language L (a state of FL), the expressions generated by L must be 'legible' to systems that access these objects at the interface between FL and external systems--external to FL, internal to the person. [...] SMT (the strongest minimalist thesis) or a weaker version, becomes an empirical thesis insofar as we are able *to determine interface conditions* (emphasis mine) and to clarify notions of 'good design'. [...] While SMT cannot be seriously entertained, there is by now reason to believe that in nontrivial respects some such thesis holds [...]" (Chomsky, 1999:1). To me, Chomsky's "Aspects" formulation of the re- search scopes for the generativist grammarian cannot hold for minimalist research anymore if one is not to contradict oneself in terms of re- search questions set to address (see Chomsky's Introduction to MP, 1995 for these questions). Performance considerations as "external cons- traints" on the functioning of the language faculty now need play some role not only in psycholinguistics but also theoretical linguistics. 15. Chomsky (1995) states that "[t]he language L [...] generates three relevant sets of computations: the set D of derivations, a subset Dc of convergent derivations of D, and a subset Da of admissible derivations of D. FI (Full Interpretation) determines Dc, and the economy conditions select Da" (p. 220). Given the assumption that the convergence of a derivation is conditional upon its interpretability at both interface levels, he hypothesises that "there are no PF-LF interactions relevant to convergence--which is not to deny, of course, that a full theory of performance involves operations that apply to the (pi, lambda) pair" (p. 220). Now some tough empirical questions for the minimalist to address:[8] (1) Suppose the derivation D converges at PF but crashes at LF. This means D is expected to crash in the final run. Now how does PF "under- stand" that D has crashed at LF, then NOT to be articulated phonetic- ally? How do PF and LF communicate? Are sensori-motor instructions sent to PF temporarily stored somewhere (where?) so that the case of D is decided on at LF, and then PF is informed (how?) to proceed with its articulation of D? (2) Also suppose that two rival derivations have converged but only one of them, say Da, passes the test of optimality. For example, (4a) below is more economical than (4b) in terms of the DISTANCE/STEPS needed for the Wh-word to move from its canonical position to [Spec C]. (4) a. Whom did you persuade t to meet whom? b. Whom did you persuade whom to meet t? Da (4a) must be blocking the less economical but still convergent derivation (4b). How is it signalled to the other interface level to phonetically articulate this single admissible derivation and not the other? How long should PF wait before deciding to articulate a pi (it is too risky to articulate pi even if D has converged at LF as it may simply prove to be less economical than another)? Can one take care of such a mapping between PF and LF without violating the independence assumption of interface levels? Is it the computational system that monitors PF and LF in this respect? Or perhaps all these questions are to be dismissed as the concern of "a full theory of performance" rather than those of the minimalist syntax as a theory of competence? 16. One way out of the dilemma is to stipulate that the convergence/ crash of each derivation is decided on in advance as LIs are mapped onto the lexical array LA or introduced into the derivation. This stipulation does not solve the problem but only displaces it. Moreover, this means that (a) ill-formed derivations do not crash; they are always cancelled, and (b) we need another interface level--LA, d- structure, or whatever you wish to call it--at which crash/convergence issue is taken care of prior to any interfacing with performance sys- tems. Whatever the case, Chomsky's formulation of such basic tenets of the theory begs the empirical questions outlined above. 2.3 Semantico-Phonetic Form 17. In absence of empirical support for Chomskyan split-interface claims, and in agreement with Liberman's (1993) requirement "that in all communication the processes of production and perception must somehow be linked; their representation must, at some point, be the same (Place, 2000: par. 40)", an attempt is made here to propose a more conservative and conceptually simpler ALTERNATIVE HYPOTHESIS--THE UNITARIANIST INTERFACE HYPOTHESIS--according to which at one and the same interface level, say the semantico-phonetic form, the derivation D containing bundles of diverse information types--phonological, formal, and semantic features--is accessible to both C-I and A-P performance systems. Compa- tible features (phonological features for the A-P system, and formal- semantic features for the other) are processed by each system, which ignores incompatible features, leaving them to the other system to interpret.[9] The derivation crashes if it still contains uninterpreted features when the processing is over. Otherwise, it converges. Although the truth of this hypothesis is not obvious either, I will try to show that it is still essentially possible to explain language data in its light without the extra step of stipulating LF covert operations, and (for the computational system) to generate optimal convergent deriva- tions at no extra cost. I will take this to be the empirical support for the unitarianist claims made here. More on this below. 18. As it was pointed out earlier, LF-PF mismatches inherent in gen- erativist works of the past two decades make it less probable for a more communication-oriented model (than the MP) to entertain the possibility of a level of representation solely interfacing the C-I performance system. It is further claimed here that sound-meaning and sound-syntax correspondences also suggest that C-I and A-P access the relevant pieces of information at the same interface level. 19. Metrical Phonology research in the organization of prosodic struc- tures known as metrical trees (with phonological constituency assumed to be binary branching with two sisters of a branch to be [S(trong) W(eak)] or [W(eak) S(trong)] affords a formal representation of strength relationships at a sentential level that is only comparable with X-bar representations in generative syntax: ------------------------------------------------- U /\ / \ / \ / \ --------------- W ------------- S ---------------- phi, I /\ / \ / \ / \ -----------W ---- S-------- W ------ S ----------- F /\ /\ / \ / \ / \ / \ / \ / \ --------S----W-S----W------S----W---S-----W--------- sigma | | | | | | | | ma ny lin guists go to Ess ex Figure 2. A metrical tree (from Durand, 1990:225) Such striking similarities between prosodic structures (shaped by phonological features, which are the input to A-P) and grammatical ones (shaped by formal features of L not interpretable at PF under minimalist assumptions) need to be explained in terms of Chomsky's split-interface claims outlined earlier. Furthermore, the direct asso- ciations between such prosodic features and Semantico-pragmatic con- siderations like pragmatic emphasis, information structures, and illo- cutionary force suggest that sound-meaning direct, i.e. non-symbolic, correspondences are real. Also data from Romance and Germanic languages (Zubizarreta, 1998) suggest that phrasal prominence (nuclear stress) reflects syntactic ordering with two major varieties: the asymmetric c-command ordering and the ordering between a head and its complement. These properties of language suggest that formal and/or semantic features and some prosodic features are mysteriously pied-piped together so that the selection of one requires that of another. The pied-piping of a formal feature and the relevant prosodic feature, say [Q] and the phonological feature of raising intonation, cannot take place in the lexicon itself: a Q particle, e.g. 'aya' in Persian, has no high tone of its own when it appears in isolation. Then the prosodic feature must be added later to the whole sentence for some unknown reason via some unknown operation. The other alternative explanation is the possibility of having one and the same feature interpreted differently by both A-P and C-I performance systems, e.g. the formal feature [Q] interpreted by A-P and C-I systems as "raising intonation" and "asking a question" respectively, an explanation which is possible only under the Unitarian- ist Interface Hypothesis. 20. Research in motor theories of language [10] suggests that gesticulation--the (involuntary) body movements accompanying speech-- highly correlates with articulatory properties of language (prosodic features included), which by their own turn match syntactic, semantic, and pragmatic ones. Informally speaking, language spreads across all body organs like a wave: it is one's existence in its totality (and not just one's speech organ) that "speaks" a language. In other words, even features intended to be interpreted by the performance system S may happen to be "read off" by the other performance systems S' and S". This supports a single interface between the competence system and the performance systems. Moreover, it implies that the original unitarianist hypothesis formulated earlier in the paper needs to be modified in this respect. A-P and I-C performance systems do not simply ignore ALL the "incompatible" features present at SPF. Synatctico- semantic information (formal and semantic features of LIs) may happen to be also interpreted by A-P as those prosodic properties of speech (like phrasal stress patterns, intonation, etc.) that are NOT inherently available in the lexical entry for LI. Gesticulation, on the other hand, could be the "translation" of such information into the language of the body. While the Unitarianist Interface Hypothesis is silent on the issue of the mechanism of implementation for the phenomena in question, it takes them to be the empirical support for the unitary nature of the interface between the computational system for human language and exter- nal performance systems. 21. The findings of some minimalist research into ASL also seem to lend support to a unitarianist theory of language. Wilbur (1998) shows that a purely functionalist approach to "brow-raising" in ASL--that "br" marks non-asserted information--cannot be the whole story. She argues that there is some syntactic motivation behind "br". Using a minimalist framework of study, she hypothesises that "br-marked structures are as- sociated [...] with [-Wh] operators" (p. 305). The brow furrow (bf), on the other hand, is still assumed to be associated with [+Wh] operators spreading across the c-command domain (Aarons et al 1992). In a unita- rianist reading of such findings, formal features associated with these operators are available for interpretation at SPF. For a language like English, gesticulation and articulation are both possible since the re- levant biological systems access the features in question. For ASL, on the other hand, such formal features--not available at PF in orthodox minimalist accounts as they are illegible there--are interpreted as brow-raising and brow furrow. It is not very probable that such features were consciously incorporated by Charles Michel de l'Epee, Thomas Gallaudet or other educators behind sign languages back in the eighteenth and nineteenth centuries. If they were, the issues were not as controversial as they are today since the origins of ASL forms are presumably more accessible than those of normal languages. It is more probable that hearing users of sign languages, who had access to such features in their native language, unconsciously incorporated the fea- tures in their signed performance, too. In other words, the relevant formal features normally associated with their L1 equivalents of ASL lexical items, e.g. Wh-words, AUX, and Q particles in articulated languages, were still accessible to their signing performance system while communicating in ASL. As a result, while they were signing LIs to other speakers of the language, a formal feature like br crept into their performance NOT as a conscious attempt to signal a [-Wh] operation of the language faculty but an unconscious move on the user's part because her signing performance system could not help accessing the relevant feature at SPF: hence, a unitary interface level between the competence system and the performance systems. 22. Brody (1995) recognises that LF representations must be "regularly recovered quite fully on the basis of PF evidence" (p. 3), which seems impossible in Chomsky's minimalist framework given the mismatches between PF and LF. Brody argues that LF representations "vary from language to language only to the extent to which language learner can determine the relevant parameters on the basis of PF data" (1995:3). He further hypothesises that "semantic interpretation rules and the lexicon have access to the same interface, the level of Lexico-Logical Form (LLF)" (1995:2). In order to achieve a higher level of conceptual economy in his theory, Brody dismisses the operation Move altogether on the grounds that "chains and Move Alpha cover the same class of pheno- mena [...]" (1995:8). Since for Brody "the concept of chains is in- dependently motivated by the principle of Full Interpretation and by the condition that determines the distribution of the set of thematic positions [...]" (p. 5), he concludes that a theory with both concepts of chains and movement is wrong. Then a minimalist theory with a single syntactic structure (LLF), which is input to SPELLOUT, emerges. "Since the theory has no movement, categories in LLF representations will have to occupy their PF positions" (p. 20). Then LLF representa- tions are comparable with S-structures "with respect to the positions lexical categories occupy [...]" (p.21). If I understand the proposal correctly, a schematic representation of LLF may look like the diagram below: Lexicon | | V Form Chain | | V LLF | | V Spell-Out | | V PF Figure 3. A formulation of Brody's LLF model 23. The model is superior to Chomsky's as it can better explain the LF- to-PF mapping. Mismatches disappear because it is LLF that is spelled out as PF. Then it comes closer to my SPF. Despite that, LLF is dis- tinctly different from SPF in two important ways. Firstly, LLF is still a split-interface model presumably with the C-I representation as the input to the A-P system. In other words, the C-I system reads the expression first, and the residue is sent to the A-P system for inter- pretation. SPF, on the other hand, dispenses with the operation Spell- Out altogether on the grounds that there is no need to separate pi and lambda at the interface level as pi and lambda features--phonological and Semantico-formal ones--are mainly unreadable BUT/THEN harmless to an incompatible performance system (with significant exceptions like some prosodic features and gestures, which LLF cannot explain). Then no mapping is needed in SPF between PF and LF at all: [11], [12] _ Meaning | | ... | | D Lexicon | E | | WHAT THE R | | SPEAKER I V | DOES V Lexical Array | A | | T | | I V | O Move and Share | I | | N | | V | I A-P. . . . . .> SPF <. . . . . . Motor | N system system _| T /\ - E . | R . | P . | R . | WHAT THE E . | LISTENER T . | DOES A . | T . | I . | O C-I | N System _| Figure 4. The SPF Model 24. Brody's (1997) formulation of the level seems similar enough to SPF as he considers LLF to be "the input to both semantic interpretation and the SPELLOUT component" (Brody, 1997:139). Following the standard minimalist terminology, Brody (1995:34) still considered SPELLOUT to be an operation of some sort, but now Brody (1997) refers to it as a component. It might be possible then to consider LLF as the input to both semantic and phonological interpretation systems. If this is really the case, then Brody's LLF, like SPF, is a unitarianist (in the first sense of the word) theory of language. Despite that, the LLF model (contrary to SPF) remains ambiguous with regard to the question of the speaker/hearer orientation of a model discussed earlier in 2.2. 25. Secondly, SPF does not reduce Move to Form Chain. Brody dispenses with Move because whenever an element moves overtly, a trace is left be- hind that is linked to the new position via an invisible chain. Then Form Chain must suffice to explain the phenomena under study. This re- ductionism, however, suffers the same weaknesses that typical reduction- ist approaches are open to: it equates {P iff Q} to {P=Q}, which ignores the causal relation between P and Q--chains as the consequence of Move. Since chains are the product of Move, Bordy's reductionist thesis misses the possibility of other syntactic effects due to Move: {P iff Q} and {P AND S iff Q} are not contradictory. With Move banished from our theories, one has to either (a) reduce other effects to Form chains again, or (b) dispense with them. In both cases, some empirical and conceptual losses will be inevitable. Movement is NOT self-motivated. It is then X, X being a morphological requirement in Chomsky's frame- work or anything else conceivable, that triggers movement. Then: X ---> Move ---> Chains Eliminating Move establishes a direct relation between X and chains, which is at least inaccurate. It is like saying John drinks iff he has a quarrel with his wife, and the police officer gives him a ticket iff he drinks, then the police officer gives him a ticket iff he has a quarrel with his wife--drinking now can be reduced to having a ticket. Although the statement 'the police officer gives John a ticket iff he has a quarrel with his wife' is true, the reduction itself is not an acceptable move because it misses the whole point. It fails to explain why things work that way. Similarly, Brody's reduction of Move to Form Chain offers a simpler description but a less adequate explanation. 3. Language and the Trade-off between Economy and Distinctness 3.1 Extrema: Minima and Maxima 26. Chomsky's strict attention to the economy of derivation and (to a less degree) the economy of representation along with his contention that "[o]ne expects 'imperfections' in morphological-formal features of the lexicon [...]" (Chomsky, 1995:9) raises both theoretical and empirical questions about the relevance of economy to other aspects of human cognition in general and the language faculty in particular, and the extent to which a minimalist grammarian can develop any taste for such facets of human cognition. Moreover, he does not explain the cost at which the purported economy of the human cognition is at- tained--what economy is counterbalanced by. More on this in the rest of the section. 27. We should bear in mind that whether the phenomenon under study is a minimum or maximum solely depends upon the angle from which we observe things. Think of two cities, say Teheran and Algiers. You can then con- nect these two points with a straight line (an arc to be more precise because the earth is a sphere). You may call the length the shortest distance. But there is inevitably another arc connecting the two the other way round the earth which is definitely the longest distance bet- ween these two points. It is now a text-book piece of reality for mathe- maticians (among many other scientists) that maximum and minimum always co-exist, that in almost all problems of the living world the organism is indulged in a search for extrema--a generic term for maxima and minima, that the graph for many of such organic problems looks like a parabola with a single point on the curve denoting both the minimal value of one variable and the optimal value of the other: an animal picks up the shortest route (minimal value) in order to maximize its efficiency of locomotion, a TV engineer minimizes the reproduction error of the set in order to maximize the quality of the picture, the reader of this paper maximizes his or her reading rate in order to reduce the reading time to a minimum, and even a person standing still over there is constantly looking for extrema in order to stay in a position of equilibrium. Likewise, any economical endeavour of the human mind must be directly associated with another extremum in the opposite direction. Then the maximal economy of representation as formulated in a model of knowledge like semantic memory might be related to some other extremum such as the least amount of retrieval time, the least expenditure of energy, etc. 28. Going back to language economy and minimalist syntax, one should ask what the thing is that a language-user minimizes, and what the thing is that she maximizes in return when features/syntactic objects are dislocated across sentential boundaries. The most straightforward "minimalist" answer to this question seems to be one with the shortest move/covert movement (Procrastinate)/ no movement operation (move as the last resort) as the minima and economy of derivation/representation as the maxima. However, Chomsky explains nowhere what exactly and technically the nature of this economy is. Do we spend less and less energy as we proceed from left to right through the ranked sequence of shortest move--> covert movement--> no movement? If yes, then Chomsky's will be a model of performance, which he has repeatedly denied. If no, then what? In other words, what are our concrete criteria for the economy of the language faculty? Secondly, and even more importantly, from Chomsky's colourful but vague expressions such as PF "extra bag- gage", it may be inferred that a wh-in-situ language is more economical than one with overt movement, that a pro-drop language with no pronominal subjects is more "perfect" than a non-pro-drop language with expletives, that synthetic languages are more elegant than analytic ones, that the more the number of "strong" features in a language the less economical and then less perfect that language would be, etc because in all these cases the amount of the speaker's PF "extra baggage" to carry would be substantially different from one language type to another. All these seem to be very undesirable (even catastrophic) but logical inferences for a theory of language that is expected to reduce parametric variation among particular (I-) languages to choosing among a number of deriva- tions all of them convergent and maximally economical. It reminds one of (now dead and buried) reflections of such historical linguists as Friederich Muller, and August Schleicher (see Otto Jesperson 1993 for a review) on the "perfectness" of "flexional" languages in comparison with agglutinating and isolating ones. 29. Semantico-Phonetic Form as a unitarianist theory of language, on the other hand, seeks to explain language phenomena as different aspects of an innately available system--most probably an exaptation of some sort with its original biological function associated with motor activities of the body and their representations in the brain (see Allott 1994, and Calvin and Bickerton, 2000)--which inevitably inter- acts with other systems of human biology and sociology in the fulfill- ment of its major function in the human society, namely communication. Formulated as such, a theory of the language faculty is expected to function as a point of convergence for two different types of theories of language: speaker-oriented theories, which focus on what happens to the speaker as she produces language (then having a potential interest in such considerations as economy), and functionalist theories of lan- guage, which inevitably have a keen interest in the communicative functions of language and how they shape language use and usage (then implying a significant role for negotiations between the speaker AND the listener). Such a negotiation of interests is characterised with (a) the speaker's natural tendency to conform to the principle of the least effort (economy), on the one hand, and (b) the listener's interest to urge the speaker to remain as distinct as possible (distinctness). Neither of these two per se can explain why language is structured as it is. This puts both formalist and functionalist explanations of language in perspective and encourages a more unified model of language. 30. In this sense, as the speaker minimises her PF "extra baggage" (i.e. she tries to maximise her conformity to the principles of natural econo- my), the property of distinctness of speech will be minimised, which disfavours the interlocutor as now he must maximise his efforts to make sense of what the speaker means. Maximal economy means uttering noth- ing, which minimises distinctness to zero--a failure to communicate. Then sentences we normally produce embody some compromise between these two tendencies so that a reasonable balance is hit between what the speaker and the listener each demands. A strict observation of economy principles, such as those assumed in purely formalist models of compe- tence (like Chomsky's), is out of the question in real performance. Semantico-Phonetic Form, on the other hand, does NOT assume its formal restrictions (see Section four below) as strict, inviolable require- ments of Universal Grammar but the natural tendency of the language faculty to minimise the energy needed in order for the speaker to com- pute structural descriptions; a tendency which is constrained by (the speaker's observation of) the listener's tendency to optimise communica- tional utility via the maximal distinctness of the message. This means that a Unitarianist Grammar has speaker economy counterbalanced by the listener's desire for distinctness. When one is maximised, the other gets minimised. 3.2 Semantico-Phonetic Form and the Convergence of Functional and Formal Models of Language Architecture 31. Many linguists in both formalist and functionalist camps maintain that a convergence between functional and formal explanations of lan- guage is possible (if not necessary) in principle. Newmeyer (1998a, b) argues that formalist and functionalist approaches can complement each other in that the former is concerned with the autonomous system at the core of language while the latter focuses on the functional motivation of syntactic structures in general. Each approach has its own merits and demerits. Formalists' focus on purely formal grammar-internal solutions has resulted in unnaturally complex treatments of phenomena while func- tionalists go to the other extreme of rejecting the existence of struc- tural systems. On the other hand, functionalists (rightly) incorporate some discourse-based explanations for syntactic phenomena that may prove to be more adequate than merely formalist accounts of language. But formalists do not forget that there are serious mismatches between forms and functions. For Newmeyer, these two approaches can converge on (a) what a model is constructed of, (b) developing a synchronic model of grammar-discourse interaction, and (c) explaining the mechanism by which functions shape forms. 32. With more specific issues in mind, Hale (1998) observes that overt nominals in Navajo--a pronominal argument language--must not be adjuncts as it is possible to extract from NP. Otherwise, the Condition on Extraction Domains will be violated. On the other hand, some other data from Navajo strongly suggest that such nominals must be adjuncts so that no pronominal can c-command any overt nominal argument. Then some coreference phenomena seem to be in conflict with Principle C of the Binding Theory while some other suggest that the principle is observed in Navajo. 33. Kaiser's (1998) insightful paper on the Japanese post-verbal cons- truction (PVC) suggests that both formalist and functionalist accounts are needed in order to explain the Japanese PVC. From a formal point of view, PVCs are subject to such structural constraints on movement as Subjacency. From a functional point of view, there are certain dis- course contexts and not others in which some PVCs can occur. Kuno (1978), for instance, considers a PV element to be discourse- predictable. Vallduvi's (1992) theory of Informatics may serve as a basis for a unified explanation of the PVC in terms of iconicity with both formal and functional properties. 34. Functionalism and formalism also complement each other in explaining topicality and agreement (Meinunger, 1998): formalism affords a grammatical description of agreement (e.g. in terms of Chomsky's Minimalist Program) while the phenomenon can be explained functionally in reference to Givon's understanding of topicality. Meinunger proposes that "the properties which characterize the degree of topicality of a given noun phrase are linked to concrete morphological features [...] which trigger certain operations like movement or clitic doubling" (p. 213). Examples concerning the behaviour of direct objects in Turkish, Spanish, and German, the differences between Spanish and Greek clitic doubling, and object shift in Icelandic and Danish support the proposal. 35. Nettle (1998) focuses on the parallelism between linguistics and biology with respect to functionalism: for both adaptation is the result of the process of replication, variation and selection. Structural pat- terns are passed from one generation to another. But this replication is not perfect as random errors and novel solutions to specific discourse problems leak in. The linguistic equivalent of natural selection has something to do with plasticity, economy, and communicational utility that language forms can afford within the user's linguistic and cogni- tive system. 36. The core of functionalist accounts of language seems to the adoption of forms into grammar due to their communicational or cognitive useful- ness (Nettle, 1998:449). It follows that formalist theories of language economy are not necessarily incompatible with functionalism as the speaker naturally adopts a more functional system that helps to achieve maximal economy of production in real speech (Lindblom et al, 1995). Such a system saves the speaker some physical (articulatory/cognitive) effort, which is just desirable to any organism. Despite that, there are times when the speaker has to "hyper-articulate forms [...] to make herself understood, but will otherwise produce the most reduced variants she can as her speech output tends towards maximal economy of produc- tion" (Nettle, 1998:448). A more functionalist (than Chomsky's) approach to economy can better explain "imperfections" in the economy of speech as it is the communication force--the listener's demand that the speaker remain comprehensible--that may force deviations from the principles of economy. 4. The Pooled Features Hypothesis 4.1 Lexical Economy 37. Cognitive psychologists of the sixties had already thought of a model of knowledge called SEMANTIC MEMORY structured as a network of nodes and some paths between them (Collins and Quillian, 1969). In this network, two types of nodes were hypothesised to exist: set nodes, such as Animal, and property nodes, such as Can Move Around. The model was primarily concerned with economy of representation achieved through "strict hierarchical organization and placement of properties at their highest level of generalizability" (Komatsu, 1994:184). 38. Although subsequent studies suggest that semantic memory is not strictly hierarchical in its organization (e.g. Rips, Shoben, and Smith 1973), nor organized with such properties at the highest level of generality (e.g. Conrad 1972), some of the latest developments in the field of cognitive psychology such as connectionist or distributed arrays models of concepts (like McClelland and Rumelhart, 1985; Estes, 1986; Gluck, 1991; Schyns, 1991; Shanks, 1991; and Kruschke, 1992) still have an eye on economy of representation as in such PDP (Parallel Dis- tributed Processing) models one single network of simple, interconnected units can represent a good number of categories. Then although a concept is assumed to be a collection of individual representations of the members of a category, connectionist networks capture both abstracted and specific instance data while they are neither abstractive nor enumerative. 39. Assuming the lexicon to be a network of concepts and categories with some phonetic labels and certain formal features characterizing grammatical limitations on their use, one can hypothesise that the lexicon is economical in its internal organization and retrieval process both. Perhaps Chomsky has no objection to this contention. While he still endorses de Saussure's view that the lexicon is "a list of 'exceptions', whatever does not follow from general principles", he further assumes that "the lexicon provides an 'optimal coding' of such idiosyncrasies" (Chomsky, 1995:235). 40. If we are concerned with the cognitive system of the language faculty, and if "for each particular language, the cognitive system [...] consists of a computational system CS and a lexicon" (Chomsky, 1995:6), then it is quite natural to assume as THE NULL HYPOTHESIS that the system is economical in all respects--organization and retrieval of LIs, selection from the lexicon for the numeration, and derivation of structural SDs included--unless proved otherwise. Based on this, it may be hypothesised that those formal features that happen to be common bet- ween two LIs (selected for the same derivation) are copied from the lexicon onto the lexical array only once so that such LIs will share these features among themselves in order to satisfy the requirements of the principles of economy of derivation and representation such as simplicity, nonredundancy, and the like. Naturally, ALL identical features cannot be ALWAYS shared as such pooling of identical features requires the adjacency of the relevant lexical items: a very strong version of this "sharing condition" may necessitate syntactically im- possible constellations--e.g. one in which some LIs, say A through E, are arranged as pairs A-B, B-C, C-D, and D-E (with some features shared for each) but no such union as A-D (although A and D can have some features in common) because this will inevitably nullify some other unions. A weaker version, advocated here, merely requires all lexical items in the structure to have SOME feature in common with a neighbour [13]. The hypothesis formulated as such is termed here the Pooled Features Hypothesis. 4.2 Feature Sharing and Phrase Structure 41. The postulation of such a sharing mechanism has theoretical con- sequences for the unitarianist syntax; hence, more distanced from minimalist accounts of movement. Firstly, the Pooled Features Hypothesis reduces the phrase structure to a bare phrase structure in which tree diagrams are labelled with shared formal features rather than category labels. The assumption is that the phrase structure is NOT computed by the computational system: it is universally available in its barest form as a means to present an array of lexical items ((5) below). However, as lexical items are plugged into the structure, certain and not other local relations are imposed on their hierarchical organization, mainly (but not exclusively) due to the featural composition of each lexical item and the formal features it happens to share with some others (see (6) below as an illustration). In other words, due to certain economical considerations, lexical items with common formal features enter into the most local relations possible (between two LIs or their projections) so that the common formal features can be pooled. Feature sharing, in a sense, is a necessary (though not sufficient) condition on the locality of structural relations. (5) . / \ . / \ . / / \ . / / \ . / / / \ . / / / \ . / / / / \ . / / / / \ 42. Secondly, no distinction is made between such pairs as strong/weak or interpretable/uninterpretable features. Then it is not a question of un/interpretablity when a difference is detected between two features. It is rather a question of how and/or where, i.e. at which stage of the derivation, the features are supposed to be interpreted. If a formal feature is shared by two LIs, the feature is structurally interpreted in that it has made these two LIs assume the most local structural relation in a bare phrase structure:[14] (6) a. He may marry her. [3MSD] [Pres I] [V] [3FSD] . may . / \ . / may . / / \ . / / marry . / / / \ . He may marry her . [3MSD] [Pres I] [V] [3FSD] b. He married her. [3MSD] [Past V] [3FSD] . married . / \ . / married . / / \ . He married her . [3MSD] [Past V] [3FSD] Unpooled features, however, cannot have any structural interpretation. As a result, they have to wait in line until interpreted at SPF. Fortunately, pooled features, as specified here, happen to be roughly the same as those that Chomsky refers to as uninterpretable ones. The inventory of unpooled features, on the other hand, corresponds to Chomsky's set of interpretable features. Although the Pooled Features Hypothesis does not hold the distinction between interpretable and uninterpretable features, the distributional similarities between (un)interpretable and (un)pooled features minimize our theoretical and empirical losses. For Chomsky, such formal features are checked and deleted. For me, (when pooled) they shape the structure. 43. (7) represents a definition of Feature Sharing. (7) F is shared by alpha and beta iff F is a common formal feature that labels a node immediately dominating both alpha and beta or their projections. The shared feature will label the node which is on the shortest path between alpha and beta or their projections. To put it more formally: (8) SHFab <---> (Ex) [Fx & CMFab & Lxn & Dna & Dnb V Dna'' & Dnb''] (9) ~SHFab ---> (Ex) [(Fx) & ~CMFab & SPFx] where SHF stands for "share the feature", a for "alpha", b for "beta", F for "feature", CMF for "Common feature", L for "labels", n for "node", D for "dominates", '' for "a projection of", and SPF for "is interpreted at Semantico-Phonetic Form". The Pooled Features Hypothesis is compatible with Brody's (1997) radical interpretability that requires ALL features to have semantic interpre- tation. They are even similar in that Brody's bare checking theory assumes that "multiple instances of what is in fact one feature are not tolerated at the interface" (Brody, 1997:159). But feature sharing and bare checking cease to be similar at this point as for Brody, checking does take place, i.e. a feature is deleted after all, because "the mul- tiple copies of F are interpretively redundant and would violate the principle of full interpretation" (Brody, 1997:158). Feature sharing, on the other hand, assumes that an LI in the lexicon is a set of codes each pointing to some feature from one of the inventories of features-- there are three of these inventories: those of phonological, semantic, and formal features respectively. When LIs are selected, their features are copied from the lexicon onto a temporary buffer one by one so that features common between two LIs are copied only once in the fulfillment of the principles of natural economy. Note that such sharing of fea- tures can work ONLY FOR FORMAL FEATURES as FF(LI) is different from other subcomplexes, namely PF(LI) and SF(LI), in that formal features are grammatical in nature, thus INTERLEXICAL. This model is even more economical than Brody's bare checking theory which seems to introduce a feature onto LA first and then check and delete all of its copies except one in order to fulfill the principle of full interpretation. 5. Movement and Formal Features: How Imperfect Are "Imperfections"? 5.1 A Critique of Chomsky's Thesis of Movement 44. In Chapter Four of "The Minimalist Program", 'Categories and Trans- formations'(1995), Chomsky advances several claims with the aim of establishing a relation between certain morphological requirements of a language and the operation Move. According to Chomsky, "the operation Move is driven by morphological considerations: the requirement that some feature F must be checked" (Chomsky, 1995:262). Then F (a feature) raises to target beta (a full-fledged category) in K = {gamma, {alpha, beta}} to form K = {gamma, {F, beta}}, or it raises to target K to form {gamma {F, K}}. However, due to the economy condition, "F carries along just enough material for convergence. [...] Whatever 'extra baggage' is required for convergence involves a kind of 'generalized pied-piping'. [...] For the most part--perhaps completely--it is properties of the phonological component that require such pied-piping" (p.262). Chomsky (1995) argues that a principle of economy (Procrastinate)[15] requires that this movement be covert unless PF convergence forces overt raising (p.p. 264-265). 45. This formulation of Chomsky's thesis of movement, however, crucially relies on how the terms checking and the PF convergence condition are defined. Otherwise, one cannot explain why alpha (whether F or K) moves at all nor why covert raising is preferred to overt raising. Although checking is such a central concept to Chomsky's thesis, he avoids an explicit definition of the term. Instead, he appeals to intuitions, illustrations, and such comments on checking as follows (which seem to be the closest ones to a definition of the term): * "We can begin by reducing feature checking to deletion [...]. This cannot be the whole story" (p. 229). * "A checked feature is deleted when possible. [...] [D]eletion is 'impossible' if it violates principles of UG. Specifically, a checked feature cannot be deleted if that operation would contradict the overriding principle of recoverability of deletion [...]. Interpre- table features cannot be deleted even if checked" (p.280). * "[-Interpretable] features [...] must be inaccessible after checking. [...] Erasure of such features never creates an illegitimate object, so checking is deletion, and is followed by erasure without exception" (p.281). * "Mismatch of features cancels the derivation. [...] We distinguish mismatch from nonmatch: thus, the case feature [accusative] mismatches F' = [assign nominative], but fails to match F' = I of a raising infinitival, which assigns no case" (p. 309). 46. Chomsky's formulation of PF convergence as the condition on the "extra-baggage" accompanying F in its movement is even less clear in that he seems to associate it with the strength of features in question. Accordingly, a strong feature, one which is a feature of a nonsubstantive category checked by a categorial feature (p. 232), is a feature that can trigger movement (whereby both phonetic and formal features are moved together). Chomsky asserts that "if F is strong, then F is a feature of a nonsubstantive category and F is checked by a categorial feature. If so, nouns and main verbs do not have strong features, and a strong feature always calls for a certain category in its checking domain [...]. It follows that overt movement of beta targeting alpha , forming [Spec, alpha] or [alpha beta alpha], is possible only when a is nonsubstantive and a categorial feature of beta is involved in the operation" (1995:232). 47. In Chapter Four of his "Minimalist Program", Chomsky drops the stipulation underlying his formulation of strength because, as he puts it, "formulation of strength in terms of PF convergence is a restatement of the basic property, not a true explanation" (p. 233). Since he cannot think of any better formulation of strength either--"[i]n fact, there seems to be no way to improve upon the bare statement of the properties of strength" (p.233)--we have to conclude that a strong feature is one that "triggers a rule that eliminates it: [strength] is associated with a pair of operations, one that introduces it into the derivation (actually, a combination of Select and Merge), a second that (quickly) eliminates it" (p. 233). Thus: (10) (A) If F is a feature of the target so that the target is not a substantive category, AND (B) alpha is a substantive category that contains a categorial feature SO THAT ALPHA MOVES, AND (C) it enters into a checking relation with the target, AND (D) its categorial feature eliminates F (A through D altogether as equivalent to saying "F is strong"), THEN (E) alpha moves. (10) is of little interest because it is redundant and trivial: (10') [ (A) & (alpha moves) & (C) & (D) ---> (alpha moves) ] It merely tells us that if (among many other events) alpha moves then alpha moves. This reduces Chomsky's thesis of overt movement to a triviality. At best, it is as informative as saying: (11) If an element does not move overtly, then its unchecked features moves covertly in order to be checked. 48. The thesis is problematic with regard to the PF convergence condition on movement, Procrastinate, and feature strength, too (see note 15 above). Procrastinate, a natural economy condition, minimizes to zero the PF "extra-baggage" F carries with itself as it is raising to a new position to be checked. For LF movement is "cheaper" than overt movement. Then the strength of a feature (as the PF convergence con- dition on triggering overt movement) necessitating the overt movement and LF movement (as a requirement by Procrastinate) are always in com- plementary distribution so that: (12) (PF convergence ---> overt movement) V (Procrastinate ---> LF movement) This seems to be a violation of the independence assumption according to which PF and LF are two independent interface levels. 49. The thesis does not meet the condition of falsifiability either. If an element moves overtly, then the theory explains the movement in reference to some strong feature of the element. When required to offer some existence proof for such strong features, it resorts to the overt movement of the element as the syntactic evidence. If confronted with some disconfirming cross-linguistic evidence, the theory replies by saying that the feature must be weak in that language. Even if the confirming and disconfirming pieces of evidence happen to come from the same language, e.g. the feature Q in English which is to be eliminated despite being Interpretable, vaguely defined notions such as strong/weak, delete/erase, and checking relation/checking configuration make it difficult to challenge the thesis on empirical grounds. 50. Finally, Chomsky's checking theory does not explain why [-Interpretable] features should exist after all if (a) they have no interpretation at all, (b) they must always be checked, deleted, and erased without exception (p.281) in order for the derivation to con- verge, and (c) it is not uninterpretability but strength that triggers overt movement. Perhaps one needs such formal features in order to justify Chomsky's hypothesis of covert movement--the remainder of his thesis of movement. It is not so clear, however, why the language faculty should need them. 51. Chomsky's 'Minimalist Inquiries: the Framework'(1998) (henceforth, MI)--although still exploratory like "The Minimalist Program" (1995)--is intended to be a major rethinking of MP issues and "a clearer account and further development of them" (p.1). As far as Chomsky's thesis of movement is concerned, however, there seems to be no significant improvement in the original ideas discussed earlier in MP. Once more, a set of fresh terms and distinctions are introduced in order to explain the complexities of the functioning of language faculty. Chomsky seems to recognize this terminological strategy himself in the footnote 110 while discussing the featural composition of AGR when he says: "In MP, it could be avoided only by recourse to the (dubious) distinction bet- ween deletion and erasure" (1998:55). 52. Chomsky (1998) assumes movement--or "dislocation", the term Chomsky prefers in his MI--to be an apparent "imperfection of language" or a "design flaw" which makes the strong minimalist thesis untenable (p. 32). Chomsky assumes "two striking examples" of such imperfections to be: (I) Uninterpretable features of lexical items (II) The "dislocation" property "Under (I), we find features that receive no interpretation at LF and need receive none at PF, hence violating any reasonable version of the interpretability condition [...]" (p.33). "The dislocation property (II) is another apparent imperfection (as) [...] the surface phonetic relations are dislocated from the semantic ones" (p.35). Since "such phenomena are pervasive, [...] (t)hey have to be accommodated by some device in any adequate theory of language, whether it is called 'transformational' or something else" (p.35). 53. What is, according to Chomsky, the role of the minimalist programme for the syntactic theory in this regard? "The function of the eye is to see, but it remains to determine the implementation; a particular protein in the lens that reflects light, etc. Similarly, certain semantic properties may involve dislocated structures, but we want to discover the mechanisms that force dislocation. Minimalist intuitions lead us to look at the other major imperfection, the uninterpretable inflectional features. Perhaps these devices are used to yield the dislocation property. If so, then the two imperfections might reduce to one, the dislocation property. But the latter might itself be required by design specifications. That would be an optimal conclusion [...]" (Chomsky, 1998, p 36). 54. Then Chomsky seems to dispense with the concept strength altogether saying: "The concept strength, introduced to force violation of Procrastinate, appears to have no place. It remains to determine whether the effects can be fully captured in minimalist terms or remain as true imperfections" (p. 49). 55. It is too early, however, to conclude that Chomsky's MI rethinking of movement is significantly different from his MP formulation of the phenomenon. Although he dispenses with the term (but not the concept) strength, he introduces a new one--EPP-features--which is functionally similar (at least as far as movement is concerned) to strength as formu- lated in MP, and a new operation--Agree--in order to explain the mecha- nisms underlying movement: "(The) operation [...] Move, combining Merge and Agree (,) [...] establishes agreement between alpha and F and merges P(F) (generalized 'pied piping') to alphaP, where P(F) is a phrase determined by F [...] and alphaP is a projection headed by alpha. P(F) becomes SPEC-a. [...] All CFCs (core functional categories) may have phi-features (obligatory for T, v). These are uninterpretable, constituting the core of the systems of (structural) Case-assignment and "dislocation" (Move). [...] Each CFC also allows an extra SPEC beyond its s-selection: for C, a raised Wh-phrase; for T, the surface subject; for v, the phrase raised by Object Shift (OS). For T, the property is the Extended Projection Principle (EPP). By analogy, we can call the corresponding properties of C and v EPP-features, determining positions not forced by the Projection Principle. EPP-features are uninterpretable [...] though the configuration they establish has effects for interpretation" (Chomsky, 1998:14-15). He then formulates the configuration (22) below for CFCs "with XP the extra SPEC determined by the EPP-features of the attracting head H: (22) alpha = [XP [ (EA) H YP ]] Typical examples of (22) are raising to subject (yielding (23A)), Object Shift (OS, yielding (B), with XP= DO and t its trace), and overt A'-movement (yielding (C), with H = C and XP a Wh-phrase [...]: (23) (A) XP - [T YP] (B) XP - [SU [ v [V t ]]] (C) XP - [C YP] The EPP-features of T might be universal. For the phase heads v/C, it varies parametrically among languages and if available is optional. [...] [T]he EPP-feature can be satisfied by Merge of an expletive EXPL in (A), but not in (B) / (C)" (Chomsky, 1998:23). 56. The arguments against Chomsky's MP thesis of movement presented earlier seem to be relevant here, too. Chomsky's thesis is still a tautology in that it does not provide any useful information about the phenomenon. The thesis merely states that things move simply because some mysterious EPP-features up there make them move (as strong features did in his MP account of the thesis). And by EPP- features he means those features we understand must be there because of the raising of an element to the new position. Since "[c]hoice of Move over Agree follows from presence of EPP-features" (p.19), and since such features are uninterpretable ones presumably doomed to deletion in the course of the derivation, we are once more left with the question of why they should be there after all, and with the other questions discussed earlier. 57. Chomsky's allusion to "certain semantic properties" involving dis- located structures seems to have something to do with such functionalist theories as parsing or theme-rheme structure in explaining the why of movement. Chomsky has set himself on the exploration of the mechanisms involved in movement. Then one may wonder how the nature could antici- pate (if it did) our future need to such (then useless) uninterpretable features as that part of the computational mechanism we will happen to employ later when we want to move things for meaning's sake. One possi- bility is that such features evolved later to take care of our already existing needs to communicate meaning. The other possibility, which is more in line with the ideas expressed in Gould (1991) and Uriagereka (1998), is to consider uninterpretability an exaptation--a property of the language faculty that was NOT adapted for its present function, i.e. affording movement so that certain semantic effects are achieved, but later co-opted for that purpose. Uninterpretability as an adaptation must not be particularly attractive to Chomsky as it implies that uninterpretable features, which are illegible to the C-I system, are still semantically motivated in origin. Uninterpretability as an exaptation, on the other hand, makes the proposal less falsifiable than ever. 58. Roberts and Roussou in their manuscript 'Interface Interpretation' criticise Chomsky's (1995, 1998) thesis of movement on similar grounds, namely (a) the introduction of uninterpretable features that have no other role except to be deleted, (b) Chomsky's formulation of a strong feature as one with a PF reflex while they must be deleted/erased as soon as possible, (c) F-checking requiring the presence of the same feature twice, (d) checking theory imposing a ranking of principles (a conceptual anomaly in minimalist approaches), (e) case features being uninterpretable for both the attractor and the attractee, and finally (f) its failure to provide a formal account for parametric variation. 59. Instead, they propose another minimalist model--Interface Inter- pretability--in which there are only (LF) interpretable features, with strength associated with morphophonological realisation. The system is claimed to take care of parametric variation, too: "[t]he lexicon pro- vides the information determining the mapping" designated in this model as the syntactic symbols +p for PF-mapping, +l for LF-mapping, and F* when a feature "must have a PF realisation" (p.p 5-6). Parametric vari- ation then may be formulated as: (13) a. Is F* ? Yes/No b. If F*, is it satisfied by Move or Merge? Accordingly, "there can be no features that do not receive any inter- pretation at all, that is they are not interpreted in either inter- face (-p, -l)" (p. 7). 60. Roberts and Roussou's proposal, however, seems to be open to the same criticisms already levelled at other split-interface hypotheses. The proposal merely rules out a [-p, -l] feature. They do not explain how the listener/learner can recover a [-p, +l] feature, i.e. one to be interpreted in terms of meaning but bereaved of any phonetic reali- sation. Moreover, their system is based on two mapping features (or whatever they intend these to be), namely [+/- p] and [+/- l]. But it artificially exploits only [F][+p] and [F][-p] as the parametric variants. It is not so clear why "Is F$ ?"(with $=[+l]) cannot be the source of any parametric variation in such a system. 61. Furthermore, Roberts and Roussou assume Merge to be a more economi- cal option than Move/Merge + Move because Merge is costless (Chomsky, 1995). They conclude that Merge is less marked than Move as far as parametric values are concerned. Then "[t]he least economical option (in 13 above) is Move" (p. 7). This means one can think of a hierarchy of parametric options arranged as below in terms of economy: (14) (~F*) > (F* & Merge) > (F* & Move)/(F* & Move & Merge) where > stands for "more economical than". This inevitably means some languages are more economical than others: as one proceeds from left to right in the diagram, the parametric value results in a language that is less and less economical. Since economy is NOT placed on a par with distinctness, as it is done in SPF, such a stipulation can be catastrophic for a theory of language. It implies that some languages are functionally more evolved than some "less perfect" ones. For instance, a language in which a Q particle with a PF index is added to the sentence via Merge to signal questions must be more economical than one that prefers AUX-raising for the same purpose. Then Chinese is a more economical language than English. 62. Roberts and Roussou predict that no (-p, -l) features can exist. They then continue that "[t]hese features (-p, -l) are precisely those that correspond to Chomsky's non-interpretable and in particular weak non-interpretable features, as well as Case features" (p. 7). Apart from the careless phrasing of this stipulation, they forget that Chomsky's non-interpretable features can also be [+p, -l]. The phi-features of a verb or any other non-nominal, e.g. the [plural] feature of AUX in (15) below, which are definitely non-interpretable in Chomsky's system, obviously have a phonetic realisation: (15) a. I WAS going. (WAS: I, past, SINGULAR) b. We WERE going. (WERE: I, past, PLURAL) Then Chomsky's uninterpretable features, which they rightly criticise for reasons similar to those discussed here earlier and also in Lotfi (in press), can actually survive Roberts and Roussou's system because they are NOT [-p, -l]. And perhaps even more than that: Such features are ~F*, then the most economical option the system offers. This is even worse than Chomsky's Procrastinate, which Chomsky himself rejects in his 'Minimalist Inquiries' (1998), because Chomsky has always considered uninterpretability as an imperfection in the design of the language faculty rather than the most economical parametric value. 5.2 Feature Sharing and Movement 5.2.1 Structural Well-Formedness 63. According to the Pooled Features Hypothesis, an LI moves from its original position to a higher position on the hierarchy iff this raising is formally motivated, viz a formal feature of the lexical item requires alpha (the LI to move or its maximal projection) to move either because (a) the nature of the feature in question necessitates such raising (the c-command condition on scope), (b) alpha moves in order to have some feature pooled with the target, i.e. to fulfill the sharing condition of locality, or (c) a head is required to be in a position high enough to dominate the whole grammatical structure it heads. A trace is left be- hind that is connected to the moved element via a chain. Both the head and the foot of the chain continue to share some formal features with their neighbours so that they remain structurally licensed. Although raising of pooled features may be denied in this framework on the grounds that shared formal features of an LI are structurally needed where they originated, it is still possible for a formal feature to be drawn from the lexicon and be located somewhere on the tree without any lexical realization.[16] Furthermore, prior to Spell-Out, it is still possible for an LI to draw upon the lexicon and pick up new formal features. A yet unexplored possibility is to move even a POOLED feature (in order to license a structural target position for an LI) iff the foot remains licensed due to some other features still shared between the trace and its neighbouring LI (or the projection of the latter). 64. A syntactic structure is well-formed if three inter-related conditions are met in the derivation: (a) The feature sharing condition on the locality of relations between adjacent LIs (or their projections) For each node at least one formal feature of the LIs involved must be pooled in order to form a legitimate local relation. When a mother node and its daughter(s) are identical in their lexical labels, no pooled feature is needed to appear between < > as the featural label because for such elements all formal features are actually identical and, as a result, pooled. For others, the pooled features will cement lexical items or their projections together. 65. The sharing condition on lexical/phrasal adjacency is not bereaved of a functional motivation: since formal features are all assumed to have some interpretation, the adjacent LIs/their projections will be inevitably those with some Semantico-pragmatic links. For instance, the feature pooled between a transitive verb and its object links these two in terms of semantic predication. Both the speaker and the listener seem to benefit from such links as they must find it easier now to compute these LIs/phrases one after another with one leading naturally to another, and as a result, minimising the amount of the information one has to store on the working memory while computing sentences. Otherwise, the speaker/listener has to keep track of such lexical items while producing/processing the interrupting data, say the material separating the transitive verb and its complement in (16): hence, the adjacency condition on case assignment. 16. a. Mary met him yesterday. * b. Mary met yesterday him. This approach to adjacency is compatible with Hawkins' notion of Early Immediate Constituents according to which the parser prefers linear orders that maximise the IC (immediate constituent) to non IC ratios of constituent recognition domains (Hawkins, 1994:77). (b) The c-command condition on scope 66. The c-command condition on scope is formulated here in reference to the (non-)locality of a feature, viz. whether it semantically pertains to a complete sentence, or to only part of it. I define a global feature as one whose sphere of influence is (due to semantic considerations) the complete sentence. A local feature, on the other hand, pertains the interpretation of a part of a sentence. In 'We will meet him at the station', the feature [Male] pertains the interpretation of 'him' as the internal argument of the verb 'meet', while the feature [Declarative] pertains the whole sentence. As a result, [Male] is local while [Dec] global. The c-command condition on scope requires that a global feature should be in a structural position high enough to c-command all the elements of the construct and take scope over them. Mood category features--defined here as those which are concerned with the illocution- ary force of the sentence, such as declarative, interrogative, and imperative--are good examples of such global features. [Q] is distinct from [Wh] in this respect as the former exclusively requires either the proposition P or its negation ~P to be true: hence a global feature. [Wh] per se, on the other hand, is concerned with the identity of a missing argument (among other things) but not the truth value of the whole proposition: hence a local feature. 67. The c-command condition on scope may prove to be more than a formal constraint on the use of global features. Distinctness as a communica- tive constraint on production encourages the speaker to generate sen- tences that are easier to process by the audience. The existing literature on parsing effects and iconicity support the view that these performance explanations are necessary in order to afford a more comprehensive account of grammaticality and language universals (Givon 1979, 1995; Hawkins 1989, 1994; Haiman 1985; and Bybee 1985 among many others). Having a global feature like [Q] at the beginning of a sen- tence, by means of a Q particle in the initial position and/or a raising intonation for the whole sentence, presumably facilitates the audience's processing of the sentence as a question. This is in line with Givon's account of the pragmatic aspects of meaning according to which we attend first to the most urgent task: SUBJ V word order is more frequent than V SUBJ; and 10,253 is ten thousand, two hundred, and fifty three rather than three, fifty, two hundred and ten thousand. Similarly, the lexical carrier of a global feature tends to appear in the initial position so that the listener can attend the task of determining the mood category of the sentence first. This will give her more time to decide what the speaker expects her to do--the illocutionary force of the utterance--as the latter gallops towards the end of the sentence. It is also compa- tible with the functionalist theories of thematic structure (Halliday, 1970, 1973) according to which speakers place the "frame"--a point of departure for the sentence--at the beginning of the sentence in order to orient their listeners toward a particular area of knowledge. The remainder of the sentence, i.e. the "insert", enables them to narrow down what they want to say. With [Q] in the initial position, speakers give notice to listeners that they are going to ask a question. (c) The dominance condition on head position 68. A head of a phrase must be in a position high enough in order for its maximal projection to dominate all its phrasal elements. The rele- vant phrases for our discussion here seem to be root and subordinate clauses as the maximal projections of AUX/VERB and COMP respectively. The condition is implied as a formal requirement on projections in both GB and MP as in neither framework an [X] [BAR 3] is allowed. In terms of the operation Merge, "[a] derivation converges only if this operation has applied often enough to leave us with just a single object [...]" (Chomsky, 1995:226). Then the operation stops (in the fulfillment of principles of natural economy) the moment [X] projects as [X] [BAR 2] because it is now a single object open to other syntactic operations like Move and Share, which naturally apply to [X] [BAR 0] and [X] [BAR 2] as single syntactic objects but no [X] [BAR 1]. Since [X] [BAR 3] cannot project--the merger has already produced a single syntactic object, a head like AUX has to raise under certain circumstances [17] so that it can further project up there and fulfill the dominance con- dition: (17) [AUX] [BAR 2] / \ / \ / [AUX] [BAR 1] / / \ / / [AUX] [BAR 2] / / / \ / / / \ / / / [AUX] [BAR 1] / / / / \ / / / / doing / / / / / \ What-i are-j you t-j doing t-i ? ^ | |_________________| 69. In (17), the final derivation is still a root clause, then the maxi- mal projection of AUX. Since the movement of 'what' to the initial posi- tion is a forced move--in the fulfillment of the c-command condition on scope, the auxiliary also has to move to a new position high enough to satisfy the dominance condition, but not too high to violate the scope condition for the global feature [Q] carried by the Wh-word. If Merge could derive [AUX] [BAR 3], AUX would remain in its original position. In subordinate clauses, on the other hand, AUX does not need to head the clause, which is the projection of COMP. Then there is no SUBJ-AUX inversion: (18) what / \ / [AUX] [BAR 2] / / \ / / \ / / [AUX] [BAR 1] / / / \ / / / doing / / / / \ (I wonder)... what-i you are doing t-i 70. My thesis of movement states that whenever one or more of these conditions are not satisfied in a derivation, overt raising takes place. Otherwise, the derivation will be cancelled as ill-formed. These possi- bilities are explored below in further details. 5.2.2 Languages with Q-Particles 71. Japanese-type languages are well known to require no subject- auxiliary inversion rule in the formation of yes-no questions: (19) a. Kore-wa hon desu. this-TOP book is 'This is a book' b. Kore-wa hon desu ka? this-TOP book is Q-particle 'Is this a book?' (from Kuno 1973) Persian, though Indo-European, follows the same pattern in that a question particle 'aya' is added to sentence instead of AUX movement. The particle, however, is placed at the very beginning of the sentence: (20) a. Pedar khahad raft. Father will-3S go 'Father will go' b. (Aya) pedar khahad raft? Q-particle father will-3S go 'Will father go?' Both Japanese and Persian Q-particles, presumably carrying [Q], are high enough in the structure to have scope over the whole sentence. Since the Q-particle is an operator that addresses the truth value of the whole sentence, it is necessarily located in the highest position available in the sentential structure. It is hypothesised that such lexical items as Persian 'aya' occupy the head position immediately above IP. Moreover, such LIs contain the structural feature [Inflection] with a potentiality of being shared with a finite verb or auxiliary whose maximal projection is c-selected by C. Then such a Q-particle heads a root clause, and dominates all elements within it (21b). For Persian declarative sentences, the formal feature [Dec] does not have a lexical carrier. However, all the three well- formedness conditions are fully met in (21a). Under such circumstances, [Dec] is assumed to be attached to C, a null element whose maximal projection both dominates all the elements of the sentence, and shares a structural feature with 'khahad' the head of IP. (21) a. . C . / \ . / \ . / \ . / khahad . / / \ . / / khahad . / / / \ . C Pedar khahad raft . [Dec] . . . b. (Aya) pedar khahad raft? . . Aya . / \ . / \ . / \ . / khahad . / / \ . / / khahad . / / / \ . / / / \ . Aya pedar khahad raft? . [Q] 72. Japanese and Persian are also wh-in-situ languages; that is, Wh- interrogatives are formed with no Wh-fronting. In both languages, Q- particles are available in a position high enough to have scope over the whole construct. In Persian, however, the Q-particle is not obligatory. Again, this must be a case of [Q] located in the initial position without any phonetic realization. One should bear in mind that the Persian Q-particle is not obligatory in yes-no questions either. Moreover, even in standard spoken Persian, the Q-particle 'yani' can be employed to signal a (yes/no or Wh-) question (see note 18 below). (22) Japanese Wh-interrogatives: a. John-wa dare-o korosita ka? John-TOP who-DO killed Q-particle 'Who did John kill?' b. John-wa Mary-ga dare-o kiratte-iru to John-TOP Mary-particle who-DO hating-is that sinzite-ita ka? believing-was Q-particle 'Who did John believe that Mary hated?' (from Kuno 1973) (23) Persian Wh-interrogatives: a. (Aya) ke khahad raft? Q-particle who will go 'Who will go?' b. (Aya) pedar fekr-mikonad Hasan ke-ra did? Q-part. father thought-do-3S-Pres Hasan who-DO saw-3S 'Who does father think that Hasan saw?' Here questions are formed with very similar requirements and mechanisms as those of yes-no questions: The Q-particle is located in the initial position so that the scope requirement on the use of [Q] is fulfilled. The position is structurally granted via [Inflection]-sharing between the Q-particle and auxiliary/finite verb. The Wh-phrase remains in situ because (a) [Wh] is a local feature whose sphere of influence is not the whole sentence but one of its constituents, and (b) [Wh] in such languages is not piedpiped to the global feature [Q]. Then: (24) . Aya . / \ . / \ . / \ . / khahad . / / \ . / / khahad . / / / \ . / / / \ . Aya ke khahad raft? . [Q] [Wh] . 5.2.3 Languages with Overt Movement 73. Let's begin the study of overt movement in such languages with the structure of English subordinate clauses, for the differences between such structures and those of root clauses in wh-in-situ languages are minimal. These similarities could be mainly due to the presence of a lexical item with a mood category feature in English subordinate clauses. Then no other sentential element is needed to host the feature for structural reasons: (25) English subordinate clauses a. that you could see her [Dec] b. whether you could see her [Q] c. whom-i you could see t-i [QWh] But even here, some differences can be observed because in English subordinate clauses, the Wh-word still moves since even here [Q] is not lexicalized. Instead, the Wh-word contains both [Q] and [Wh]. These two features seem to be piedpiped together in a sense as the covert raising of a feature (such as [Q] in order to take scope) is denied in this model. 74. Like Persian declarative/interrogative root clauses, English subordinate clauses (in 26a, b, c) conform to well-formedness conditions outlined earlier. (26) a. . that . / \ . / \ . / could . / / \ . / / could . / / / \ . / / / see . / / / / \ . that you could see him . [Dec] . . b. whether . / \ . / \ . / could . / / \ . / / could . / / / \ . / / / see . / / / / \ . whether you could see him . [Q] . c. . whom . / \ . / \ . / could . / / \ . / / could . / / / \ . / / / see . / / / / \ . whom you could see t . [Q] . In all these cases, appropriate local relations are established between nodes with pooled features. Moreover, mood category features in these trees are carried by LIs that both c-command all other elements, and head the whole construct. 75. Interrogative roots in English, however, show different syntactic patterns than those of Japanese-type languages because of no lexical- ization of mood category features in English roots. As a result, other sentential components compete for hosting the feature as required by well-formedness conditions. In order to form a yes-no question in Modern Standard English, the formal feature [Q] is introduced into the derivation. Contrary to Japanese-type languages, there is no specific lexical item comparable with 'ka' and 'aya' to carry [Q]. Still contrary to Early Modern English, it is the auxiliary rather than the verb that carries [Q] in MSE. This seems to be a lexical difference between EME and MSE. Whatever the case, an existing LI functions as a host to [Q] so that the scope condition on the use of [Q] is satisfied. In EME, when the finite verb (27a, b) moves to the initial position, other well- formedness conditions are satisfied automatically. Then no more structural modifications are needed. In MSE yes-no questions, on the other hand, it is the auxiliary rather than the finite verb which is introduced into the derivation in order to host [Q]. Then Aux moves in the fulfillment of well-formedness conditions (27c). (27) a. You saw him. (EME/MSE) . C . / \ . / \ . / saw . / / \ . / / saw . / / / \ . C you saw him. . [Dec] . . b. Saw you him? (EME) . saw . / \ . / \ . / \ . / saw . / / \ . / / saw . / / / \ . Saw you t him ? . [Q] . . c. Did you see him? (MSE) . did . / \ . / \ . / did . / / \ . / / did . / / / \ . / / / see . / / / / \ . Did you t see him ? . [Q] 76. Other Germanic languages like German and Dutch seem to follow the same pattern as that of EME whenever an auxiliary is not present in the sentence. Otherwise, the auxiliary is fronted with similar specifica- tions to MSE auxiliary inversion. Other structural differences are due to SOV word order in German and Dutch. (28) German yes-no questions a. Kauft Karl das Buch? buys Karl the book 'Does Karl buy the book?' b. Hat Karl das Buch gekauft? Has Karl the book bought 'Has Karl bought the book?' (from Haegeman 1991) (29) Dutch yes-no questions a. Koopt Wim het boek? buys Wim the book 'Does Wim buy the book?' b. Heeft Wim het boek gekocht? has Wim the book bought 'Has Wim bought the book?' (from Haegeman 1991) 77. In order to form a Wh-question in MSE, the formal feature [Q] and [Wh] are to be introduced into the derivation. Contrary to wh-in- situ languages, however, [Q] and [Wh] seem to be piedpiped together in interrogative words for such languages with obligatory Wh-fronting. Then the Wh-word has to move overtly to the beginning of the sentence in order for [Q] to take scope. This can explain the syntactic configurations depicted in (30a) and (30b). In (30a), the Wh-word containing both the formal features [Q] and [Wh] is already in a position c-commanding all other sentential elements. Then no Wh- raising needs to take place. The feature-sharing account of (30a) seems to be more elegant and economical than orthodox accounts of the phenomena according to which even in (30a) the Wh-word is raised to Spec-CP position. In (30b), on the other hand, both the Wh-word and Aux (could) raise to new positions in the fulfillment of the scope condition on [Q] insertion, the sharing condition on structural local- ity, and the dominance condition on head position. (30) a. . saw . / \ . / saw . / / \ . Who saw her ? . [QWh] . .b. could . / \ . / could . / / \ . / / \ . / / could . / / / \ . / / / could . / / / / \ . / / / / see . / / / / / \ . Whom could you t see t . [Q Wh] 78. Brody's (1997) analysis of such sentences also incoporates an inter- pretable Wh-feature. However, his [+Wh] seems to be more similar to my [Q] as he assumes [+Wh] to be loaded onto AUX even in yes/no questions. Furthermore, he assumes the feature to be carried by AUX (and not Wh- words) in Wh-questions, which is in sharp contrast with the analysis offered here because Brody's [+Wh] does not require the c-command condition on the scope of global features outlined earlier. 79. Given sharing assumptions, however, the ungrammaticality of (31a, b, and c) needs to be explained: (31) . a. whom . / \ . / \ . / saw . / / \ . / / saw . / / / \ . * Whom you saw t ? . [Q Wh] . . . b. whom . / \ . / whom . / / \ . / / \ . / / could . / / / \ . / / / could . / / / / \ . / / / / see . / / / / / \ . * Whom you could see t ? . [QWh] . . c. whom . / \ . / \ . / whom . / / \ . / / \ . / / could . / / / \ . / / / could . / / / / \ . / / / / see . / / / / / \ . * Could whom you t see t . [Q Wh] . In both (31a and b), the dominance condition on head position is violated. The finite verb 'saw' and the auxiliary 'could' are the legitimate heads of the relevant interrogative roots while in both cases it is the Wh-word that occupies the head position. (31c) is even worse than (31a and b) because both the c-command condition on scope and the dominance condition on head position are violated in this structure. 80. Data from other Germanic languages seem to provide further support for this account of movement. In both German and Dutch, the finite verb is raised to the initial position of a yes-no question whereby the dominance and scope conditions are fulfilled. The sharing condition is also satisfied because three projections of 'kauft' in (32) below are identical in structural features. In case of Wh-interrogatives, Wh- raising is a forced move, too. Otherwise, the scope condition would be violated. Since German and Dutch data are syntactically identical in this regard, only German trees are provided below. (32) German interrogatives . . a. kauft . / \ . / \ . / kauft . / / \ . / / kauft . / / / /\ . / / / /____\ . Kauft Karl t das Buch? . [Q] . 'Does Karl buy the book?' . . b. kauft . / \ . / kauft . / / \ . / / \ . / / kauft . / / / \ . / / / kauft . / / / / \ . Was kauft Karl t t? . [QWh] . 'What does Karl buy?' . . . . c. hat . / \ . / \ . / hat . / / \ . / / hat <+n> . / / / \ . / / / gekauft . / / / / \ . / / / / \ . / / / / \ \ . / / / /___\ \ . Hat Karl t das Buch gekauft ? . [Q] . 'Has Karl bought the book?' . 5.2.4 Multiple Questions and Movement 81. It is a common observation that single-pair answer is impossible in English multiple questions like 'Who bought what?': it is infelicitous in a situation like a store to ask such a question when someone sees that someone else buys an article of clothing but does not see who it is and what exactly is bought. A pair-list answer, on the other hand, is felicitous when someone, say the store clerk who has been out for an hour, asks his assistant the same question expecting an answer like 'Mr Brown bought a jacket, Mrs Smith bought a sweater, ...' (see Grohmann 1999 for some other situations). In Japanese, Chinese, and Hindi, which are all wh-in-situ languages, either a single-pair or a pair-list one is possible. Hagstrom (1998) argues that a single-pair multiple question is a set of propositions, while a pair-list question is a set of sets of propositions. The Q-morpheme in wh-in-situ languages is an existential quantifier that originates in a clause internal position and then moves into CP. If Q moves from the lower wh-phrase, it will be a pair-list question. If it moves from a position higher than both wh-phrases, a single-pair question will be the result. 82. French also permits both interpretations depending on the use of the in-situ or the Wh-movement strategies: single-pair answers are pos- sible in French only with the in-situ strategy. "It is possible that the obligatoriness of syntactic movement of a wh-phrase to SpecCP for some reason forces the pair-list interpretation" (Boskovic, 1998). Then what happens in Japanese is due to the semantically motivated movement of Q while the overt Wh-movement in English is motivated by a strictly formal syntactic requirement. Boskovic also observes some variation in multiple Wh-fronting languages. Bulgarian (in which the overt movement of a Wh- phrase to SpecCPs is obligatory) patterns with English, and Serbo- Croatian (no Wh-phrase overt movement to interrogative SpecCPs) with Japanese. Grohmann (1999) extends Boskovic's (1998) adaptation of Hagstrom's (1998) semantics to German. Both Wh-elements in a German multiple question move overtly with one of WHs targeting SpecFocP and the other the lower projection FP. Then German patterns with Bulgarian rather than English with regard to multiple questions. 83. Significantly, the data on multiple questions in Persian suggest that associating single-pair interpretations with wh-in-situ languages is not empirically borne out: Persian, a wh-in-situ language with a Q particle in the initial position to mark its interrogatives, seems to pattern with English rather than Japanese, Chinese, and Hindi. It normally affords only pair-list interpretations in multiple Wh- questions, which sheds doubt on both Hagstrom's (1998) semantics of multiple questions and Boskovic's (1998) and Grohmann's (1999) adapta- tions. 84. A group of 40 adult native-speakers of (Esfahani) Persian studying at Azad University were asked to indicate on a five-point-scale how in- felicitous a multiple question was in each of the two situations des- cribed below. In all cases, the Q particle 'yani' [18] was employed to signal the question rather than 'aya' because in informal Persian 'yani' (or preferably no Q particle at all) is used in Wh-questions. Since multiple questions are rarely employed in written Persian, and also because the use of '-o/-ro' as the cliticised form of 'ra' has certain important consequences for the felicitous interpretations of such questions, it was decided to assume a more conversational style in writing the items in question. Their ratings are tabulated for each case separately: Situation I You are in a store and off in the distance see somebody buying an article of clothing, but do not see who it is and does not see exactly what it is being bought. You go to the shop-assistant and ask: - Yani ki chi kharid? Q-particle who what bought 'Who bought what?' - Ali ye pirhan kharid. Ali a shirt bought 'Ali bought a shirt.' Scale of infelicitousness: 0 1 2 3 4 Number of participants who preferred each point on the scale: 0 3 2 17 18 Number in percentages: 0% 7.5% 5% 42.5% 45% Rating scores: 0 3 4 51 72 Possible MAX: 160 Possible MIN: 0 Total: 130 (out of 160) Situation II You are paying a social visit to a newly-married couple in their apart- ment. While having a friendly conversation about the wedding presents they received from their friends, you ask about both what they received and who sent them each: - Yani ki chi avord? Q-particle who what brought 'Who gave you what?' - Ali ye sa'at avord, Maryam ye angoshtar avord, Mina ye goldan Ali a clock brought, Maryam a ring brought, Mina a vase avord, ... . brought 'Ali gave us a clock, Maryam gave us a ring, Mina gave us a vase, ... .' Scale of infelicitousness: 0 1 2 3 4 Number of participants who preferred each point on the scale: 15 15 2 5 3 Number in percentages: 37.5% 37.5% 5% 12.5% 7.5% Rating scores: 0 15 4 15 12 Possible MAX: 160 Possible MIN: 0 Total: 46 (out of 160) 85. The ratings suggest that the PL-reading of a multiple question in Persian is about 2.83 times more felicitous than its SP-reading. Then: (33) # a. Yani ki chi kharid? (infelicitous '#') (single-pair answer) b. Yani ki chi avord? (felicitous) (pair-list answer) Despite that, some other data from Persian multiple questions seem to be more in harmony with what happens in other wh-in-situ languages: Situation III While resting in her office, the teacher notices that some of her students are hitting their friends. Not wearing her glasses, she fails to identify them. Later she goes to her class and asks: - Yani Ki ki-o zad? Q who whom hit 'Who hit whom?' - Hasan Ali-o zad, Hamid Arash-o zad, Sina Pedram-o zad, ... . Hasan Ali-DO hit, Hamid Arash-DO hit, Sina Pedram-DO hit 'Hasan hit Ali, Hamid hit Arash, Sina hit Pedram, ... .' (felicitous, pair-list answer) Situation IV While resting in her office, the teacher notices that one of her stu- dents is hitting his friend. Not wearing her glasses, she fails to identify them. Later she goes to her class and asks: - Yani Ki ki-o zad? Q who whom hit 'Who hit whom?' - Hasan Ali-o zad. Hasan Ali-DO hit 'Hasan hit Ali.' (felicitous, single-pair answer) Situation V Ali knows that his friend Hasan used to have three cars on sale: a white B.M.W, a red Chevrolet, and a black Jaguar. Hasan tells him that a close friend of Ali's bought one of these three cars last week. Ali wants to know who he was and what car he bought: - Yani ki chi-o kharid? Q who what bought 'Who bought what?' - Hamid Jaguar-o kharid. Hamid Jaguar-DO bought 'Hamid bought the Jaguar.' (felicitous, single-pair answer) Situation VI Back from his holiday, Hasan notices that his assistant has sold all the cars on sale. Hasan asks: -Yani ki chi-o kharid? -Hamid Jaguar-o kharid, Bahram Chevrolet-o kharid, Ahmad B.M.W-ro Hamid Jaguar-DO bought, Bahram Chevrolet-DO bought, Ahmad B.M.W-DO kharid, ... bought (felicitous, pair-list answer) 86. Interestingly, in all the multiple questions with a SP-reading the direct object marker 'ra' follows the Wh-phrase (cliticised as '-o' or '-ro' in spoken Persian) that is the internal argument of the verb. Al- though 'ra' carries an ACC feature, which will be shared by the object, there seems to be also a [+ specific] borne by 'ra'. Browne (1970) as- sumes that Persian 'ra' is comparable with Turkish '-i' in this respect as both carry the feature, and as a result make the object specific, too. Karimi (1989, 1990) follows Browne in this regard and considers 'ra' as a specificity marker for the direct object. She further argues that the interrogative element 'chi' may or may not cooccur with 'ra' when it is in the object position. If it does, it receives a specific reading: (34) a. emruz ketab xarid-am today book bought+1sgS 'I bought books today.' b. chi-(*ro) xaridi? what bought+2sgS 'What did you buy?' (from Karimi, 1990:149.30) (35) a. ketab-i-ro ke be to gofte bud-am xarid-am book+indef+ra that to you told was+1sgS bought+1sgS 'I bought the book that I had told you (about).' b. Chi-ro xaridi? what+ra bought+2sgS 'What did you buy?' (from Karimi, 1990:149.31) 87. Then in '(Yani) ki chi kharid?' both 'ki' and 'chi' are [-specific] while in '(Yani) ki chio kharid?' the first Wh-word is [-specific] but the second [+specific]. It follows that in (36a) below, the only possible interpretation is a non-specific, generic one. It actually means 'someone, who can be anyone and I want to know who one is, bought something, which can be anything and I want to know what it is'. In (36b), on the other hand, what is bought is marked as specific. Then it might be thought to have a specific 'ki' subject, too. Still on the other hand, 'ki' in (36b) is [-specific] and can take scope on 'chio' in that for each non-specific person that antecedes 'ki', there can be a [+specific] 'chio' object. Then the question is ambiguous in that if the speaker decides that specificity of the Wh-word 'chio' lies within the non-specificity of the c-commanding Wh-word, i.e. 'ki', then the sen- tence will still have a non-specific pair-list interpretation. Other- wise, the specificity of the Wh-word 'chio' makes a single-pair inter- pretation possible. (36) a. Ki chi kharid? [-specific] [-specific] b. ki chio kharid? [-specific] [+specific] 88. The ambiguity of (36b) is comparable with that of a sentence with two quantifiers; one a universal quantifier in a c-commanding position, and the other an existential quantifier in a lower position. Then 'everyone loves someone' is ambiguous in scope as it may be paraphrased either as 'everyone has somebody or other that one loves' or 'there is some particular individual whom everyone loves'. The data provided by Fox and Sauerland (1997) suggest that a generic tense/context can also be behind such scope effects. Then ignoring tense, (37a) is identical to its counterpart (37b). (37b) is generic while (37a) is not: (37) a. Yesterday, a guide ensured that every tour to the Louvre was fun. b. In general, a guide ensures that every tour to the Louvre is fun. (from Fox and Sauerland, 1997) 89. Similarly, it may be hypothesised that the feature [+/- specific] carried by Wh-phrases in multiple questions determines PL and/or SP reading(s) of the question. Then the differences among such languages as English, Japanese, Bulgarian, and Persian may be due to certain dif- ferences in the featural composition of their Wh-phrases rather than the semantically motivated movement of Q, or the Wh-phrase overt move- ment to interrogative SpecCPs. The hypothesis correctly predicts that for a Persian multiple question with Wh-phrases like 'koja' (where), 'kay' (when), and 'chera' (why), a pair-list answer is obligatory as no DO-marker can be attached to such Wh-phrases: (38) a. Ki kay raft? Who when went [-spec] [-spec] 'Who went when?' b. Ki koja raft? [-spec] [-spec] who where went? 'Who went where?' c. Ki chera raft? [-spec] [-spec] 'who why went?' 90. Interestingly enough, for Persian it is the arithmetic sum of the number of Wh-phrases with [+/- specific] feature (irrespective of their structural height) that in the final run determines the pair-list/ single-pair reading of the sentence. Then in a sentence with two [-specific] Wh-phrases, the pair-list reading is obligatory. When one of these two Wh-phrases but not the other is [-specific], both PL and SP readings are possible. Finally, in a sentence with one [+specific] Wh- phrase and two [-specific] ones, a PL reading is again obligatory: (39) a. (Aya/Yani) ki chi kharid? (pair-list answer) [-spec] [-spec] b. (Aya/Yani) ki chio kharid? (pair-list/single-pair answer) [-spec] [+spec] c. (Aya/Yani) ki chio koja kharid? (pair-list answer) [-spec] [+spec] [-spec] 91. Cross-linguistic variation is then possible with regard to three different factors: (a) A Wh-phrase can be either [-specific] or [+specific]. For a lan- guage like Persian both types are possible. Some other languages are conceivable that afford only a [-specific] feature. (b) A marker marks the (non)specificity of the Wh-phrase. For Persian, it is the DO marker 'ra' that marks the Wh-phrase as [+specific]. The negative value of the feature is available by default. Then Persian [+specific] wh-phrases can only occupy the object position. For other languages, completely different markers are conceivable. (c) The structural height of the Wh-carrier of [+/- specific] feature may or may not matter in determining the PL/SP reading of the question. In Persian, the structural height of the Wh-phrase does not matter. 6. Conclusion 92. The Pooled Features Hypothesis is a unitarianist hypothesis. It dis- penses with LF in its generative account of movement, which is always overt because the well-formedness conditions have to be satisfied first in order to license the derived structure. It is only then that a sentence can be interpreted by performance systems. Therefore there is no need to postulate any LF interface level to explain how the well- formedness of a sentence is guaranteed after Spell-Out with no PF realisation. Moreover, it is possible to think of some functional explanation for feature-sharing as it presumably makes a sentence easier to process (see par. 63). It follows that the Pooled Features Hypothesis is unitarianist in a third sense, too: since feature-sharing is both an economy requirement on production and a processing facility, the hypothesis can "unify" the speaker and the listener in a single model. The speaker saves time and energy as she minimises the number of features to be mapped onto the lexical array. She must also benefit from the formal links, i.e. features to be pooled, between the adjacent LIs and/or their phrasal projections in her selection of items from the lexicon: hence, easier to produce 'She will be doing that' than 'she be that will doing'. The listener has her own share of sharing as (due to the links mentioned above) she finds the former easier to process than the latter. Notes [1] As May points out, (1a) is ambiguous as it could mean 'the men introduced each other to everyone that the women introduced the men to'. Then we could have (2') instead of (2): (2') The men-j introduced each other-j to everyone-i that the women-k did [VPellip introduce them-j to e-i]. In both (2) and (2'), however, elliptical VP could be still hypothesised to contain the elements they do with their phonetic content deleted. [2] That two copies of 'each other' in (2) are different in their references would pose no problem as the two copies of 'each other' are not co-referential in (1b) either. [3] Isn't it too late to criticise generativists for what they did about two decades ago? To me, the answer is 'no'. The argument still has a target to address: although Minimalist syntax today is radically different from GB in both its conceptualisation and manipulation of logical forms, the empirical motivation for this level of represen- tation goes back to the GB framework. It appeared in MP as an un- questionable assumption of the field. Chomsky (1993) had already dis- pensed with internal interfaces S-structure and D-structure. Perhaps the move was so radical that the existence of the interface level (LF) was not doubted at all. Hence no need to look for more existence proof for LF. Although this does not necessarily make matters worse, it does explain why I have targeted at the GB empirical evidence for LF. [4] Extending Chomsky's conclusion, Barss (1986) had already argued for the satisfaction of Principle A at s-structure, too. Otherwise, (i) would be grammatical with the anaphor licensed at LF: (i) * David-i wonders who showed which picture of himself-i to Mary. [5] Also note the direction of arrows on Uriagereka's diagram. [6] As far as I remember, nowhere in Chomsky (1995), (1998), and (1999), any reference is made to the listener's reconstruction of LF based on PF as a possible interpretation of the model. There is a strong tendency in Chomsky (1999), however, to give more weight to interactions between syntax and phonology, e.g. the phonological edge, leftward TH/EX as the function of phonological component, etc. Although I welcome this shift of interest from LF to PF, it seems to widen the existing gap between these two levels, which makes Chomsky's model even less comprehensible than before. Anyway, I discuss the issue of reconstruction as a mere possibility. [7] The LF-to-PF mapping may be equated to language learning to answer this question. The answer, however, is not satisfactory because this "explanation" could be equally used for relating any other LF with any other PF that the linguist wishes to link to save his theories. It simply displaces the problem as now one can ask how the learner can learn about such mismatches. Because the PLD is not rich enough to show the mismatches, one should hypothesise some innate mechanism--like parameterised mapping--to explain the problem of learn- ing inherent in this case. But even this cannot be the whole answer. To be more specific, a language-user encounters the sentence 'Jane knows that Mary distrusts herself'. Due to his innate knowledge of the binding principles, he assumes 'Mary' and the anaphor to be coreferential. However, if the pragmatics of the language repeatedly proves 'Mary' in such cases to be co-indexed with 'Jane', then he sets a parameter, like Wexler and Manzini's (1987) Governing Category Parameter, at the relevant marked value to take care of the data. So far so good. But what LF-advocates expect the learner to do with regard to LF-to-PF mapping is next to a miracle. Encountering the sentence 'the men introduced each other to everyone that the women did', he raises the quantifier to derive its logical form: hence, 'each other' can also be coreferential with 'the women' after Quantifier Raising. This presumably needs no learning as for Chomsky LF is universal, then most probably innately available. What is miraculous in this case is the learner's knowledge NOT to do the same thing to some other sentences like 'He phoned everyone that John knew'. Else, 'he' and 'John' could be co- referntial after QR at LF, which is impossible in English. [8] I say "tough" because I doubt that the standard generative methodology, which relies on the linguist's intuitions and grammatical judgements, is always adequate for empirically addressing such cogni- tively inclined questions as those put forward by the minimalist, like those forwarded by Chomsky in his introduction to MP, e.g. "what con- ditions are imposed on the language faculty by virtue of [...] its place within the array of cognitive systems of the mind/brain [...]" (1995:1)? [9] This part of the hypothesis will be weakened later to take care of sound-meaning and sound-syntax correspondences with regard to prosodic features of language. [10] See Kendon (1972, 1991), McNeil (1992), Brownman and Golstein (1992), Allott (1994) for relations between language and gestures. [11] In actual performance, the listener's C-I system "reads" a copy of SPF perceived (via the listener's A-P system) as a "reconstruction" of the speaker's copy. Hence, one single SPF with two actual copies: the speaker's, and the listener's. SPEAKER: | V A-P. . . . . .> spf1 system | | V ACTUAL SPEECH - - - - - - - - - - - - - - - - - - - - - - - - - - - - - LISTENER: /\ . . . A-P system - - - - - -> spf2 /\ . . C-I system interpret . . .> produce - - -> At any rate, spf1 and spf2 must be similar enough in order for actual speakers and hearers to communicate. The listener's A-P system can manage the reconstruction of spf1 as spf2 due to the similarities between the speaker's and the listener's A-P systems. The SPF model (depicted in Figure 4) captures this with its assumption of a single interface level at which a derivation becomes accessible to both the speaker and the listener. The LF-to-PF mapping is dispensed with. [12] An interesting empirical question to ask is whether the listener's A-P and motor systems can also access SPF via the listener's copy: . . A-P Motor system - - - - - - > spf2 < . . . . . . system . /\ /\ . . . . . . . . .: . . . C-I system If yes, the listener may subconsciously "echo" the speaker's speech and/or "mirror" her gesticulations and physical movements even when the listener has no visual access to the speaker's performance. Even if true, such "echoing-mirroring" behaviour does not need to be easily observable under normal circumstances. It could be as silent as lips movement, muscle tension, or laryngeal movements. [13] As mentioned earlier, a stronger version of the hypothesis, which requires ANY formal feature to be ALWAYS selected only once and then shared by ALL relevant lexical items, is obviously false. Otherwise, as one of the referees notes, a sentence like 'all the dog-s which Mary feeds will bite chicken-s' will be problematic because the plural feature in this case is selected twice. 'Dogs' and 'chickens' cannot share the feature because in order to afford that they first need to be structurally adjacent, which is barred due to the feature shared by 'bite' and 'chickens'. [14] Pooled features appear between < >; others between [ ]. In (6a), is shared by 'he' and 'may', and by 'may' and 'marry'. Pooled features appear here on that side of a tree node that is closer to the LI/NODE to function as a partner: hence, to the right and to the left of the first and second projections of 'may' respectively. [15] Chomsky (1998, 1999) finally abandoned the principle Procrastinate. He thinks the principle is not formulable as before anymore as the overt-covert distinction has collapsed (Chomsky, 1999:12). The prin- ciple is abandoned because it is dispensable as another case of look- ahead (1998:49). He even advocates "something like the opposite: per- form computations as quickly as possible, the 'earliness principle' of Pesetsky (1989)" (Chomsky, 1999:12). Since the concept strength was originally introduced to explain the violation of the principle, he concludes that strength has no place in this framework either. Later I argue that Chomsky's new term "EPP-features" is essentially the same (at least in terms discussed here) as "strong features". [16] The feature must normally have some SPF realization, however. Otherwise, the listener has simply no chance to become aware of it. A suprasegmental like stress or intonation can be a convenient SPF real- ization of such isolated formal features without a lexical item specifically selected to bear the feature in question. In Persian, for instance, the lexical item 'aya' can be dropped from a yes/no question. Despite the absence of this lexical device to carry [Q], the rising intonation of the sentence makes the listener understand the presence of [Q] because even in the 'aya'-less question, the feature [Q] still has some SPF realisation. [17] See Section 5.2.3. [18] Contrary to 'aya', 'yani' is an Arabic word which has been bor- rowed into Persian. In standard formal Persian, the word means 'means/is equal to'. Then: 'Rehlat' yani 'margh'. demise means death ' 'Demise' means 'death'.' Despite that, and contrary to Persian verbs, it cannot be inflected at all nor occur in the final position of the sentence. In spoken Persian, on the other hand, 'yani' also replaces 'aya' in both yes/no and Wh- questions: Hasan raft. Hasan go+3sg+past 'Hasan went.' Yani Hasan raft? Hasan went 'Did Hasan go?' Yani Hasan koja raft? Hasan where went 'Where did Hasan go?' References Aarons, D., B. Bahan, J. Kegl, and C. Neidle. (1992). Clausal Struc- ture and a Tier for Grammatical Marking in ASL. Nordic Journal of Linguistics 15:103-142. Allott, R. (1994). Gestural Equivalence (Equivalents) of Language. Berkeley, [http://www.percep.demon.co.uk/index.htm]. Barss, A. (1986). Chains and Anaphoric Dependence: On Reconstruction and its Implications. Doctoral dissertation, MIT. Boskovic, Z. (1998). On the Interpretation of Multiple Questions. Chomsky Celebration Website, [http://mitpress.mit.edu/\ celebration]. Brody, M. (1995). Lexico-Logical Form: A Radically Minimalist Theory. Cambridge, MA: MIT Press. Brody, M. (1997). Perfect Chains. In Elements of Grammar: Handbook of Generative Syntax, ed. L. Haegeman. Dordrecht: Kluwer. Browne, W. (1970). More on Definiteness Markers: Interrogatives in Persian. Linguistic Inquiry 1:359-363. Browman, C. and L. Goldstein. (1989). Gestural Structures and Phonolo- gical Patterns. Status Report on Speech Research SR-97/98 pp. 1-23. NewHaven, Conn.:Haskins Laboratories. Browman, C. and L. Goldstein. (1991). Gestural Structures: Distinctive- ness, Phonological Processes, and Historical Change. In Modu- larity and the Motor Theory of Speech Perception, eds. Mattingly, I. M. and M. Studdert-Kennedy. Browman, C. and L. Goldstein. (1992). Articulatory Phonology: An over- view. Phonetica 49:155-180. Bybee, J. (1985). Morphology: A Study of the Relation Between Meaning and Form. Philadelphia: Benjamins. Calvin, W. and D. Bickerton. (2000). Lingua ex Machina: Reconciling Darwin and Chomsky with the Human Brain. Cambridge, MA: MIT Press. Chomsky, N. (1965). Aspects of the Theory of Syntax. Cambridge, MA: MIT Press. Chomsky, N. (1981). Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. (1993).A Minimalist Program for Linguistic Theory. In The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, eds. Hale, K. and S. Jay Keyser, Cambridge, MA: MIT Press. Chomsky, N. (1995). The Minimalist Program. Cambridge, MA: MIT Press. Chomsky, N. (1998). Minimalist Inquiries: The Framework. MIT Occasional Papers in Linguistics 15, Cambridge, MA: MIT Press. Chomsky, N. (1999). Derivation by Phase. MIT Occasional Papers in Lin- guistics 18, Cambridge, MA: MIT Press. Collins, A. M.& M. R. Quillian (1969). Retrieval Time from Semantic Memory. Journal of Verbal Learning and Verbal Behavior 8. 240-247. Conrad, C. (1972). Cognitive Economy in Semantic Memory. Journal of Experimental Psychology 92 . 149-154. Estes, W. K. (1986). Array Models for Category Learning. Cognitive Psychology 18. 500-549. Fox, D. and U. Sauerland. (1997). Illusive Scope of Universal Quantifiers, [http://web.mit.edu/afs/athena.mit.edu/user/s\ /a/sauerlan/www/Scope_Illusions.html]. Givon, T. (1979). On Understanding Grammar. New York: Academic Press. Givon, T. (1984). Syntax: A Functional-Typological Introduction. Amesterdam: Benjamins. Givon, T. (1995). Functionalism and Grammar. Amsterdam: Benjamins. Gluck, M. A. (1991). Stimulus Generalization and Representation in Adaptive Network Models of Category Learning. Psychological Science 2. 50-55. Gould, S.J. (1991). Exaptation: A Crucial Tool for Evolutionary Psycho- logy. Journal of Social Issues 47, 43-65. Grohmann, K. (1999). German is a Multiple Wh-Fronting Language! Colloque de syntaxe et semantique a Paris 3. Haegeman, L. (1991). Introduction to Government and Binding Theory. Basil Blackwell. Hagstrom, P. (1998). Decomposing Questions. Doctoral dissertation, MIT. Haiman, J. (1985). Natural Syntax. Cambridge: Cambridge University Press. Hale, K. (1998). Conflicting Truths. In Functionalism and Formalism in Linguistics, eds. M. Darnell, E. Moravcsik, F. Newmeyer, M. Noonan, and K. Wheatley. Philadelphia: Benjamins. Halliday, M. A. K. (1970). Language Structure and Language Function. In New Horizons in Linguistics, ed. J. Lyons. Baltimore: Penguin Books. Halliday, M. A. K. (1973). Explorations in the Functions of Language. London: Edward Arnolds. Hawkins, J. A. (1989). Competence and Performance in the Explanation of Language Universals. In Essays on Grammatical Theory and Universal Grammar, eds. D. Arnold, M. Atkinson, J. Durand, C. Grover, and L. Sadler. Oxford University Press. Hawkins, J. A. (1994). A Performance Theory of Order and Constituency. Cambridge: Cambridge University Press. Jesperson, O. (1993). Progress in Language with Special Reference to English. Philadelphia: Benjamins. Kaiser, L. (1998). Representing the Structure-Discourse Iconicity of the Japanese Post-Verbal Construction. In Functionalism and Formalism in Linguistics, eds. M. Darnell, E. Moravcsik, F. Newmeyer, M. Noonan, and K. Wheatley. Philadelphia: Benjamins. Karimi, S. (1989). Aspects of Persian Syntax, Specificity, and the Theory of Grammar. Doctoral dissertation, University of Washington. Karimi, S. (1990). Obliqueness, Specificity, and Discourse Functions: Ra in Persian. Linguistic Analysis 20:139-191. Kendon, A. (1972). Some Relationships between Body Motion and Speech: An Analysis of one example. In Siegman, Aron Wolfe and Benjamin Pope eds. Studies in Dyadic Communication. New York: Pergamon. Kendon, A. (1991). Revisiting the Gesture Theory of Language Origins. Paper for LOS Meeting, De Kalb, Illinois. Komatsu, L. K. (1994). Experimenting with the Mind: Readings in Cognitive Psychology. California: Brooks/Cole Publishing Company. Kruschke, J. K. (1992). ALCOVE: An Exemplar-Based Connectionist Model of Category Learning. Psychological Review 99. 22-44. Kuno, S. (1973). The Structure of the Japanese Language. Cambridge, MA: MIT Press. Kuno, S. (1978). Japanese: A Characteristic OV Language. In Syntactic Typology, ed. W. Lehmann. Austin: University of Texas Press. Liberman, A. M. (1993). Haskins Laboratories Status Report on Speech Research 113:1-32. Lindblom, B., S. Guion, S. Hura, S. Moon, and R. Willerman. (1995). Is Sound Change Adaptive? Revista di Linguistica 7:5-37. Lotfi, A. R. (to appear). Minimalist Program Revisited: Chomsky's Strength to Trigger Movement. Proceedings of the 34th Col- loquium of Linguistics. May, R. (1985). Logical Form: Its Structure and Derivation. Cambridge, MA: MIT Press. May, R. (1991). Syntax, Semantics, and Logical Form. In The Chomskyan Turn, ed. A. Kasher. Oxford: Blackwell. McClelland, J. L. & D. E. Rumelhart (1985). Distributed Memory and the Representation of General and Specific Information. Journal of Experimental Psychology 114. 159-188. McNeill, D. (1992). Hand and Mind: What Gestures Reveal about Thought. Chicago: University of Chicago Press. Meinunger, A. (1998). Topicality and Agreement. In Functionalism and Formalism in Linguistics, eds. M. Darnell, E. Moravcsik, F. Newmeyer, M. Noonan, and K. Wheatley. Philadelphia: Benjamins. Nettle, D. (1998). Functionalism and Its Difficulties in Biology and Linguistics. In Functionalism and Formalism in Linguistics, eds. M. Darnell, E. Moravcsik, F. Newmeyer, M. Noonan, and K. Wheatley. Philadelphia: Benjamins. Newmeyer, F. (1998a). Language Form and Language Function. Cambridge, MA: MIT Press. Newmeyer, F. (1998b). Some Remarks on the Functionalist-Formalist Controversy in Linguistics. In Functionalism and Formalism in Linguistics, eds. M. Darnell, E. Moravcsik, F. Newmeyer, M. Noonan, and K. Wheatley. Philadelphia: Benjamins. Pesetsky, D. (1989). Language-Particular Processes and the Earliness Principle. Ms. MIT. Place, U. T. (2000). The role of the hand in the evolution of language. Psycoloquy 11(007), [http://www.cogsci.soton.ac.uk/cgi/psyc/n\ ewpsy?11.007]. Rips, L. J., E. J. Shoben, & E. E. Smith (1973). Semantic Distance and the Verification of Semantic Relations. Journal of Verbal Learning and Verbal Behavior 12. 1-20. Roberts, I. and A. Roussou. ms. Interface Interpretation. University of Stuttgart. Schyns, P. G. (1991). A Modular Neural Network Model of Concept Acquisition. Cognitive Science 15 . 461-508. Shanks, D. R. (1991). Categorization by a Connectionist Network. Journal of Experimental Psychology: Learning, Memory, and Cognition 17. 433-443. Uriagereka, J. (1998). Rhyme and Reason: An Introduction to Minimalist Syntax. Cambridge, MA: MIT Press. Vallduvi, E. (1992). The Informational Component. New York: Garland. Wilbur, R. (1998). A Functional Journey with a Formal Ending: What do Brow Raises Do in American Sign Language? In Functionalism and Formalism in Linguistics, eds. M. Darnell, E. Moravcsik, F. Newmeyer, M. Noonan, and K. Wheatley. Philadelphia: Benjamins. Zubizarreta, M.L. (1998). Prosody, Focus, and Word Order. Cambridge, MA: MIT Press. Author's address: Department of English Language Azad University at Khorasgan, Esfahan, IRAN. E-mail: lotfi@www.dci.co.ir