Feature 13. Subject islands

found in question(s): 31a, 31b, 31c, 31d, 32a, 32b, 33a, 33b, 33c, 33d, 33e.i, 33e.ii, 33f.i, 33f.ii, 34, 36a, 36b, 36c, 36d.i, 36d.ii, 37a, 37b, 38a.i, 38a.ii, 38b.i, 38b.ii, 38c, 38d, 38e.i, 38e.ii, 38f.i, 38f.ii

Definition and illustration

In some languages such as English, subextraction is generally only possible from objects but not from subjects or adjuncts. The intransparency of subjects has been linked to a subject island constraint. In later work, the intransparency of non-complements has been accounted for by the more general Condition on Extraction Domains (Huang 1982), according to which only complements are transparent. The following pair shows the asymmetry for subjects vs. objects:

(1)

In some of the literature, the intransparency of subjects is related to the fact that they occupy a derived position (they have to move to Spec,TP because of the EPP). The derived position is then too high for further extraction ("anti-locality" cf. Bošković 2016) or the movement itself causes a phrase to become an island ("freezing"). In other languages, extraction from subjects has been shown to be possible, as shown in the German example in (2):

(2)

This has been correlated with a low position of the subject, i.e., in languages where there is no EPP or the EPP is optional, subjects can remain low. In that case, subextraction is taken to be possible (also cf. Rizzi 1982). Given that VO- and OV-languages have been argued to differ in the presence of an EPP-requirement (Fukui 1986, Haider 2010), this predicts that subjects should be more transparent in OV-languages than in VO-languages.

If the link between the EPP and subject islands is true, subjects outside SpecTP should be transparent more generally. This can be illustrated with extraposed subjects in English:

(3)

This, again, aligns with Haider's (2013) view that specifically preverbal subjects should be islands (also see Rizzi 1982).

In general, this feature also tries to gauge the validity and cross-linguistic applicapility of islandhood tests in general and as a diagnostic of the EPP in particular.

Somewhat more complicated predictions arise in approaches where one of the two orders is taken to be basic, while the other is derived from it by means of movement. For instance, in Kayne (1994), VO is taken to be basic with OV-order derived by means of movement of the object. This may predict OV-languages to be less transparent than VO-languages. Conversely, in Haider (2013), OV-order is taken to be basic, predicting it to be more transparent than VO-order.

Testing for island properties requires a series of steps to check for basic extraction properties. Which kinds of movement are available in general (interrogative fronting, contrastive fronting Q35)? Are the relevant phrases transparent in general in object position (Q31/36) and is the adjunct condition obeyed (Q32/37)? Questions 33/38 then address subject islands directly but the outcome can only be interpreted based on the findings in the previous questions.

Correlations

Question 33/38 investigate subject islands, concretely, extraction from clausal and NP-subjects in both their canonical and non-canonical (lower or extraposed) position. It is predicted that extraction from the canonical subject position should be impossible in VO-languages due to the EPP but not in OV-languages (while extraction from a non-canonical subject position should not distinguish between the two types of languages):

• Prediction 1: V-O → EPP → *extraction from subjects in canonical position
• Prediction 2: O-V → EPP → extraction from subjects in general
• Prediction 3: EPP → *extraction from subjects

See also

This feature is closely connected to several others that explore the consequences of the presence/absence of the EPP, i.e.: 5, 7, 12, 17, 18

References