A Companion to Chomsky. Группа авторов
neither the top nor the tail of a dependency D is a gap (and therefore both the top and the tail are overt items), in cases where the dependencies concern nominal expressions, the status of the tail often determines which constraints are emplaced on establishing D.9 For instance, the locality conditions on establishing the dependency vary according to whether the tail member is a personal or possessive pronoun (e.g. she, her) or a reflexive or reciprocal pronoun (e.g. herself, each other). Roughly speaking, a personal or possessive pronoun P can establish a syntactic dependency with a c‐commanding nominal expression N only if N does not occupy P's binding domain (23),10 whereas a reflexive or reciprocal pronoun R can establish a syntactic dependency with N only if N occupies R's binding domain (24).
1 (23)a.William thinks that [BD Edith dislikes him].b.*Edith thinks that [BDWilliam dislikes him].(BD = binding domain for him)
2 (24)a.*William thinks that [BD Edith dislikes himself].b.Edith thinks that [BDWilliam dislikes himself].
Interestingly, common and proper noun phrases, such as the red apple and London, can never be tails of non‐gap dependencies, as (25) shows. In this example, the sentence is judged as unacceptable if she and Virginia refer to the same person. The sentence is only acceptable if she and Virginia refer to different people, in which case no dependency is established between the two phrases.
1 (25) She regrets that Virginia hurt Joseph's feelings.
When a nominal tail engages in such dependencies, it is referred to as bound. The three facts about nominal binding exemplified in (23) to (25) have endured because they are incredibly useful for diagnosing syntactic structure. For instance, if one wants to know the size of particular syntactic phrase P (i.e. whether it is as large as a binding domain), one can test to see how bound pronouns behave within P. Alternatively, if one wants to know if a particular hierarchical position A in a sentence c‐commands another position B, one can see if a nonlocal dependency between two co‐referring proper nouns can be established using these positions. If it cannot, then A c‐commands B.
It should be clear at this juncture that one can organize the discoveries discussed so far in this subsection such that they form an emerging typology of nonlocal syntactic dependencies. This typology is presented in diagrammatic form in (26).
1 (26)
We must emphasize here that, although the table in (26) consolidates the preceding text in a clear and instructive way, it does not attempt to provide an exhaustive typology of the nonlocal dependencies uncovered by generative linguists over the last 60 years. The main reason for this is that the position of certain types of nonlocal dependencies within this table is still debated. For instance, it remains undecided whether the phenomenon of extraposition, which involves phrases appearing rightward of their canonical position (compare (27a) and (27b–c)) and which was first documented by Rosenbaum (1967), involves a gap (27b) (as Ross 1967 and Baltin 1981 contend) or not (27c) (as Culicover and Rochemont 1990 and Haider 2010 argue).
1 (27)a.Someone that I don't know has left a message on your answer machine.b.Someone Δ has left a message on your answer machine that I don't know.c.Someone has left a message on your answer machine that I don't know.
This typology (or one similar to it) has endured because it represents the agreed‐upon generalizations that form the bedrock for much recent generative linguistic research. This research has typically aimed to either (i) subsume a newly discovered or neglected nonlocal dependency under an established class or (ii) show that certain classes in the established typology are only superficially different, and that, at the correct level of analysis, they are indistinct.
In research that aims to fulfill (i), the discoveries that constitute (26) are used as diagnostic tools. Consider the phenomenon of scrambling (Ross 1967), which refers to the process that derives noncanonical word orders, typically within the same clause (compare (28a) and (28b), from Japanese).
1 (28)a.Mary‐gasonohon‐oyonda(canonical word order)Mary‐NOMthatbook‐ACCread‘Mary read that book.’b.sonohon‐oMary‐gayonda(scrambled word order)thatbook‐ACCMary‐NOMread‘Mary read that book.’
Recall that A‐ and A′‐dependencies behave differently with respect to crossover and parasitic gaps (recall the discussion surrounding the examples in (16) to (18)). By using these facts and others as diagnostic tools – i.e. by observing how scrambling behaves with respect to these phenomena – linguists were able to determine whether scrambling is an A‐ or A′‐dependency. It transpires that there are actually two forms of scrambling (A‐scrambling and A′‐scrambling) (Fanselow 1990; Mahajan 1990; Webelhuth 1992; see Karimi 2008 for an overview), an important discovery demonstrating that the same surface syntactic pattern – namely, noncanonical word‐order – can arise from distinct grammatical processes.
In attempting to reduce the typology in (26) to a more fundamental picture, research that aims to fulfil (ii) frequently tackles the question of why natural language contains the seemingly different classes of nonlocal dependencies that it does. Chomsky's linguistic research from the mid‐1970s onward has reductionist tendencies in this vein, and has been explicitly reductionist since 1995's Minimalist Program. A common target for this type of linguistic research is locality constraints. For instance, Chomsky (1977) famously proposed that the locality constraints on establishing an A′‐dependency – namely, that an A′‐dependency cannot extend across a syntactic island or two clausal boundaries – are reflections of one underlying locality constraint called Subjacency. According to conceptual framework behind Subjacency, certain phrases (the Tense Phrase and the Noun Phrase in English) are, by their nature, partial disruptors of A′‐dependencies that extend across them. This idea has been retained in Subjacency's successor, the Phase Theory (Chomsky 2000, 2001a, 2001b), in which certain phrases (the Complementizer Phrase and the vP (as in (9)) are again considered to be inherent disruptors of nonlocal dependencies. In removing the notion of syntactic islands from syntactic theory, Subjacency and its successor Phase Theory in effect unite the locality conditions on A and A′‐dependencies, with the ban on Improper Movement (see (18b)) yielding the apparent finite‐clause‐boundedness of A‐dependencies.
4.3.3 A Typology of Gaps
When coupled with the fact that functional heads are often “empty” (see Section 4.3.1), the typology in (26) also underscores the fact that, at least according to most interpretations of fruitful generative inquiry, natural language contains many gaps. The ubiquity of gaps in natural language – not only as the tops and tails of nonlocal dependencies but also as independent items – has prompted extensive research into the precise status of gaps and whether all gaps have the same status. With regards to the former issue, generative linguists are divided as to whether gaps are linguistic items in their own right, akin to “silent” phrases, words, or morphemes, depending on the gap in question (this tradition starts with Chomsky 1973), or whether they are merely indications of the special grammatical or categorial status of the syntactic phrase in which the gap is found (this tradition took hold with Gazdar et al. 1985). With regards to the issue of whether all gaps have the same status, the picture is also unclear, with differences posited in earlier generative research being subject to reductivist reanalysis, especially within the current post‐1995 Minimalist paradigm. However, many of differences between gaps posited in earlier research can nonetheless be classified as enduring discoveries, as few – if any – reductivist analyses of gaps enjoy universal acceptance.
Famously, different types of gaps have been implicated in explanations of the semantic and syntactic behavior of predicates that select infinitival