A Companion to Chomsky. Группа авторов
– and which are enduring discoveries in their own right – were observed to apply to some nonlocal dependencies but not others. It was quickly demonstrated that, for any nonlocal dependency D, the syntactic type of the elements involved in the dependency, whether the dependency includes a gap or not (recall (1) versus (3)), and the syntactic position occupied by the top member of the dependency can each determine what constraints are emplaced on successfully establishing D in the first place.
For instance, dependencies between syntactic heads of phrases (e.g. (10)) are more constrained than dependencies between syntactic phrases themselves, as the former can only be established across a very short structural distance (two structurally consecutive syntactic heads; Travis 1984), whereas certain dependencies between syntactic phrases can be unbounded. With regards to dependencies between phrases, those involving gaps as tail members typically cannot be established if the gap is contained in one of a handful of distinguished phrases known as islands (Ross 1967), whereas those without gaps are typically untroubled by islands (compare (11) and (12), with an “adjunct island” headed by because). Therefore, the status of a dependency's tail also determines across what distance the dependency can be established.
1 (10) Will Polly Δ be attending the party?
2 (11) Every auntie thought that baby Bobby smiled [ISLAND because she had just arrived].
3 (12) * I know who baby Bobby smiled [ISLAND because Δ had just arrived].
When there are no syntactic islands present in a sentence, it appears at first glance that the distance across which a dependency with a phrasal tail gap can be established is unrestricted. In the sentences in (13), for instance, the apparent tops and tails of each dependency occupy different clauses, and the dependencies themselves extend across multiple clausal boundaries (which are denoted by C1, C2, and C3).
1 (13)Xavier seems [C1 to appear [C2 to want [C3 to be hired Δ by Sue]]].Who does Bo think [C1 Zoë hopes [C2 Jo believes [C3 that Sue hired Δ]]]?
Another enduring result of generative syntax research has been to show that, in these cases, appearances are deceptive. In reality, nonlocal dependencies of this type cannot extend across more than one finite clausal boundary.7 Thus, the sentences in (13) actually contain multiple dependencies, none of which extend across more than one clausal boundary, as illustrated in (14). The presence of each dependency in (14a,b) – a number of which consist entirely of gaps and therefore appear as “hidden” dependencies – can be confirmed through following the generative methodology outlined in Section 4.3.1: i.e. by applying syntactic tests known to diagnose the presence of such hidden dependencies and by showing that visible versions of these dependencies are attested in other languages (Frisian: Hiemstra 1986, German and Romani: McDaniel 1989; Chamorro: Chung 1994; Irish: McCloskey 2001; among many others). In certain dialects of German, for instance, the intermediate gaps in (14b) are replaced by question words, as (15) shows.
1 (14)
2 (15)WenglaubstduwenPetermeintwenSusiheiratet?(Felser 2004 (10))whobelieveyouwhoPeterthinkswhoSusimarries‘Who do you believe Peter thinks that Susi is marrying?’
We previously mentioned that the structural position occupied by the top member of a dependency may determine the syntactic properties of that dependency. This is especially noticeable for dependencies between phrases with gaps as their tail members. In such cases, a correlation exists between the structural position occupied by the top member of the dependency – whether it occupies an argument position (an A‐dependency), such as the syntactic positions reserved for subjects and objects, or a non‐argument position (an A′‐dependency) – and how the dependency interacts with other grammatical phenomena. If a nominal expression that refers to the same person or thing as the top member of a dependency D intervenes between the top and tail members of D, D is licit if it is an A‐dependency but illicit if it is an A′‐dependency (this is a crossover effect; Postal 1971) (compare (16a) and (16b)). A‐ and A′‐dependencies differ in other ways, too. A′‐ but not A‐dependencies can license the presence of parasitic gaps (compare (17a) and (17b), where “ΔP” represents a parasitic gap) (Ross 1967, Engdahl 1983), and the top member of an A‐dependency can also be the tail member of an A′‐dependency, but not vice versa (compare (18a) and (18b)) (Chomsky 1973). Their behavior also differs with respect to reconstruction (May 1977, Chomsky 1977, van Riemsdijk and Williams 1981; see Barss 2001 for an overview), a pervasive linguistic phenomenon to which we return in Section 4.3.4.
1 (16)a.Fiona seems to her mother Δ to be a genius.[A](acceptable when Fiona and her are interpreted as referring to the same person)b.*Who does her mother love Δ?[A′](unacceptable when who and her are interpreted as referring to the same person)
2 (17)a.This secret file should be burned Δ after reading {it / *ΔP}.[A]b.Which secret file did the FBI agent burn Δ after reading {it / ΔP}?[A′]
3 (18)
Having uncovered this rich empirical landscape of nonlocal dependencies with tail gaps, an obvious question that arose was this: For each type of dependency with a tail gap, is there a mirror version, whereby the top is a gap and the tail is an overt item? And if so, do these mirror versions display the same properties as the originals? Thanks to another handful of enduring discoveries, generative linguists have demonstrated that both questions must be answered with “yes.” Although originally couched in the language of transformations, a “top‐gap” dependency between heads was first postulated by Chomsky (1957) for English verbal inflection (19), and the universal presence of such dependencies in natural language has been confirmed by much succeeding research (see Adger, Harbour, and Watkins 2009; Harley 2013; and Harizanov and Gribanova 2018 for discussion).
1 (19) Polly Δ often eat‐s raw carrots for breakfast.
In many languages, question phrases such as who, why, and which student do not occupy the sentence‐initial position in a standard question – as in English – but instead occupy their typical position in non‐question counterparts (e.g. the subject, object, or adverbial position). Such languages are known as wh‐in‐situ languages. Generative linguists discovered that question phrases in wh‐in‐situ languages instantiate top‐gap A′‐dependencies (see the Turkish example in (20)), in which the gap occupies a position high above the sentence, therefore allowing the question phrase, via its connection with the gap, to take logical scope over the entire sentence. The establishment of these top‐gap dependencies is constrained in precisely the same manner as their tail‐gap counterparts (see examples (11) and (12) and accompanying text).8 To see this clearly, compare the simplified phrase markers in (21). Putting irrelevant differences between these languages aside, one observes that their question‐formation strategies are the same, differing only in which member of the A′‐dependency is pronounced (i.e. top‐gap in English vs. tail‐gap in Turkish).
1 (20)ΔAyşekim‐iöp‐tü?Ayşewho‐ACCkiss‐PST‘Who did Ayşe kiss?’
2 (21)
In addition, generative linguists have also discovered that some languages, such as Adyghe (a Northwest Caucasian language; see Potsdam and Polinsky 2012), utilize top‐gap A‐dependencies. In (22), the subject of root clause is an unpronounced gap, yet this gap co‐refers with the demonstrative pronoun a‐xe‐me “these/they” in the embedded infinitival clause.
1 (22)Δ[ a‐xe‐mepj&ip.schwa;sme‐ra‐tx&ip.schwa;‐new ]ø‐fjež'a‐&ip.rscpi;e‐xDEM‐PL‐ERGletter‐ABS3PL.ERG‐write‐INF3ABS‐begin‐PST‐3PL.ABS‘They