1 Introduction

Sentences in natural languages are not simply a linear sequence of isolated words like beads on a string. Words and parts of words in one position can influence the form and interpretation of words or elements in another position of the sentence: they encode dependencies between one another in the form of shared properties (Koster 1987: 8–9). There are a variety of such dependencies, including singular–plural agreement, anaphor binding, and case-marking (Koster 1987: 13–14). An important type of dependency surfaces in displacement phenomena, which Corver (2006: 567) considers to be “a core property of human languages”. Displacement links an element to another position in the sentence where it is semantically interpreted, even though it does not occur in this position in the surface structure (Chaves & Putnam 2020: 1). Often, dependencies do not apply to elements in linear sequence but rather across other elements; these are called long-distance dependencies (LDDs). In principle, LDDs are not syntactically constrained in the number of constituents or clauses they can span. For example, the sentence in (1a) contains an agreement dependency that stretches across fifteen words, from the plural noun people to the plural verb form are, indicated here by co-indexation. The wh-question in (1b), a filler–gap dependency (FGD), can be formed across multiple CP-boundaries: it links the sentence-initial wh-pronoun what, the so-called filler, to the complement position of the verb in the most deeply embedded clause. This position where the filler is semantically interpreted is called the gap site, here indicated by an underscore. For ease of reference, FGDs are shown by co-indexing.

Dependencies such as these are still subject to non-structural limitations: while there is no theoretical or competence-based limit on the distance for dependency formation, there is a practical, performance-based, one that reflects the finite amount of memory resources available to the parser (Miller & Chomsky 1963; Chomsky 1965; Sprouse & Hornstein 2013; Wagers 2013; Chaves & Putnam 2020). However, there are certain structural domains which apparently do not allow the formation of filler–gap dependencies: these domains are known as islands since the landmark dissertation of Ross (1967). Even a comparatively short dependency cannot be established into these domains and the result is considered ungrammatical. Island domains include subjects, complex NPs, relative clauses, and adjuncts; there are also related phenomena such as wh-islands, factive, and negative islands, where FGDs cannot be established across elements carrying a specific type of feature. The term “island effect” is used to describe the decreased acceptability of a sentence where a FGD, such as wh-extraction, relativization, or topicalization, is established between a filler in the left periphery and a gap inside one of the island domains. Sprouse & Hornstein (2013) do not use the term “island violation” due to its use as an explanatory term in structural or syntactic approaches to islands. The term island effect is more widely applicable in structural as well as non-structural approaches, such as semantic, information-structural or processing approaches because it is an observational term. A violation assumes the presence of a constraint whereas an effect is “agnostic about the source of the unacceptability” (Sprouse & Hornstein 2013: 2, fn. 4).

The main goal of this monograph is to investigate factors that influence the possibility of establishing FGDs into a subset of adjunct islands. Adjuncts are considered to invariably resist extraction in nearly all approaches to islands and are often called strong or absolute islands; in contrast, extraction from weak or selective islands is sensitive to characteristics of the extracted element (Cinque 1990; Szabolcsi 2006). The strong island status of adjuncts is captured in Huang’s (1982) Condition on Extraction Domain (CED) in (2), which restricts extraction to properly governed categories; governed positions generally refer to complement positions in an X′ phrase structure, excluding specifiers and adjuncts.

Thus, extraction from adjuncts such as adverbial clauses is illicit, as shown in the examples in (3); gap sites are again indicated by underscores, and square brackets indicate the adverbial clause:

Since the original formulation of the CED, the concepts it is based on have been fundamentally revised and partially become obsolete in the framework of the Minimalist Program (Chomsky 1993, 1995b). However, extractions out of non-properly governed categories, meaning subjects and adjuncts, are still considered ungrammatical by a majority and there have been numerous attempts at a reformulation of the CED using Minimalist concepts; see G. Müller (2010, 2011) or Johnson (2003) for recent proposals, and Stepanov (2007) for an overview of earlier formulations. The categorical failure to establish FGDs into an adjunct constituent as predicted by the CED does not appear to hold up universally, as there are cases where extraction is reported to be possible; these counterexamples to the CED are introduced in the following section.

1.1 Examples of grammatical extractions from adjuncts

Examples of apparently grammatical extractions from adjuncts have been noted in the literature even before the formulation of the CED, for example in Cattell (1976) or Chomsky (1982). However, they do not offer a comprehensive explanation for these exceptions. Chomsky (1982) considers the example in (4b) to show a “fairly high” degree of acceptability, and Cattell (1976) considers the example in (4a) as a potential problem for his restrictive extraction approach.

There are also more complex examples of grammatical CED violations, such as (5), where the extracted wh-phrase is linked simultaneously to a gap inside a subject and an adjunct; this violates both the subject and adjunct clauses of the CED.

Several attested examples of extraction from adjuncts are collected in Santorini (2019). The sample from these attested examples in (6) shows that extraction out of adjunct constituents occurs naturally despite the fact that this is considered ungrammatical by the CED; square brackets around the adjunct constituent and an underscore as indication for the gap site have been added.