2.1 Pragmatic dominance

The concept of dominance in relation to extraction structures was introduced by Erteschik-Shir (Reference Erteschik-Shir1973), and further developed in Erteschik-Shir & Lappin (Reference Erteschik-Shir and Lappin1979) and Erteschik-Shir (Reference Erteschik-Shir1982, Reference Erteschik-Shir1998). More recent versions of the same basic idea are adopted in works such as Van Valin (Reference Van Valin2005) and Goldberg (Reference Goldberg2006), where embedded environments which are pragmatically foregrounded (i.e. dominant) are distinguished from those that are backgrounded (i.e. non-dominant). In these works, what acceptable extraction constructions are argued to have in common is that the informational center of the utterance lies in the gapped embedded clause rather than in the matrix clause (see also Engdahl Reference Engdahl1997:60). Erteschik-Shir & Lappin's definition of dominance is that ‘a constituent c of a sentence S is dominant in S if and only if the speaker intends to direct the attention of his hearers to the intension of c, by uttering S’ (Erteschik-Shir & Lappin Reference Erteschik-Shir and Lappin1979:43). In turn, dominant constituents are argued to be identifiable by the so-called ‘lie test’, with the crucial property that only dominant constituents can be felicitously negated. Subsequent proposals based on the same idea include Van Valin (Reference Van Valin2005:215) and Goldberg (Reference Goldberg2006:130ff.). Goldberg's test, for example, is known as a ‘negation test’, and the author states that ‘when an assertion is negated, only elements within the potential focus domain are contradicted’ (Goldberg Reference Goldberg2006:135). However, for all relevant purposes, the results of these tests are the same. Applied to example (1) above, the dominance test gives the result in (6), indicating that elements within this (non-dominant) relative clause are unavailable for extraction:

1. (6) Speaker A: We met the mathematician who solved the puzzle.

   Speaker B: That's a lie – you didn't meet him/*he didn't solve the puzzle.

In other words, negating the non-dominant part of the structure (who solved the puzzle) leads to an infelicitous result in this context, presumably because the relative clause in this sentence represents presupposed information (Goldberg Reference Goldberg2006:130). Thus, the claim here is that the acceptability of extraction from complex NPs is determined by pragmatic function rather than by syntactic constraints.

On the assumption that constraints on extraction in at least some cases are related to pragmatic notions such as dominance, an important question concerns the nature of crosslinguistic variation. For example, the Scandinavian languages have often been claimed to allow a greater range of extractions, for example in the case of extraction from complex NPs (see Allwood Reference Allwood1982; Andersson Reference Andersson1982; Engdahl Reference Engdahl1982, Reference Engdahl1985, Reference Engdahl1997; Erteschik-Shir Reference Erteschik-Shir1982; among others). While this claim has sometimes been contested by non-Scandinavian linguists (see Section 2.2 below), I will argue here that the available evidence strongly suggests that although some factors that determine extraction acceptability in Swedish are the same as in English, there are clear examples where there is a marked difference between the two languages.

The first group of sentences that have been shown to be completely natural and regularly occur in Swedish (as well as in Danish) are based on presentational constructions, i.e. structures where a new referent is introduced into a situation through the use of it/there-insertion. Some authentic examples are the following, where (7) is borrowed from Engdahl (Reference Engdahl1997:59), and (8) and (9) from Lindahl (Reference Lindahl2010:33–34):Footnote ⁴

1. (7)

   

2. (8)

   

3. (9)

   

Still, at least in a subset of the acceptable Swedish examples, extraction from such constructions appears to work relatively well even in English, as in Deane's (Reference Deane1991) example (3) above. A second category of sentences where there appears to be a consistent difference between the two languages is one involving matrix verbs of perception or cognition.Footnote ⁵ Whether a particular verb has this function is context-dependent, but some common verbs in this category are känna ‘know’, se ‘see’, möta ‘meet’ and veta ‘know of’. Some authentic and fully acceptable Swedish extractions of this type from Engdahl (Reference Engdahl1997:67) are examples (10) and (11), while example (12) borrowed from Erteschik-Shir (Reference Erteschik-Shir1982:176) shows the parallel situation in Danish:

1. (10)

   

2. (11)

   

3. (12)

   

Erteschik-Shir (Reference Erteschik-Shir1982:177) argues that in the relevant cases such matrix verbs can be perceived as introducing dominant embedded clauses, and that evidence of this is provided by the fact that these embedded clauses can be negated in the dominance test, as in (13), based on (12):

1. (13) Speaker A: I have met many people who have done that.

   Speaker B: That's a lie – nobody has done that!

Thus, the test is argued to identify foregrounded or ‘dominant’ embedded clauses, and certainly, all the Swedish and Danish constructions presented so far in the paper, in (7)–(12), seem to pass the test. Whether the embedded clause is perceived as dominant appears to depend on several interacting factors. Andersson (Reference Andersson1982:40), for example, concludes that (at least) context, relative clause function and sentence content are all relevant factors affecting acceptability in these structures, and he argues that complex NP extraction is often possible in cases where the relative clause has lost its standard ‘identifying’ function.

While the context-dependent aspect of these proposals goes some way towards accounting for the variability of speakers’ judgments within a language, the proposals are more problematic as accounts of crosslinguistic differences. For example, as Erteschik-Shir (Reference Erteschik-Shir1982:177) herself observes, the lie test appears to give the same result in both English and the Scandinavian languages, which means that pragmatic dominance cannot on its own explain perceptions of acceptability in the relevant languages. Her suggestion is that, while the dominance criterion applies in all languages, there is an additional syntactic constraint in English on extraction from complex NPs – a constraint which does not exist in Danish and Swedish. Allegedly, the result is that extraction constructions involving dominant relative clauses in Scandinavian are fully acceptable, whereas extraction constructions involving the most dominant relative clauses in English are perceived as better than those involving non-dominant ones although still not syntactically grammatical.

Clearly, these previous analyses characterize a number of important features of relative clause extractions, and concepts such as dominance, background/foreground, presupposed information, etc. reflect factors which appear to be involved when speakers judge the acceptability of these structures. However, in the following I will focus on a particular set of examples where Swedish clearly and consistently differs from English in terms of the degree to which relative clause extractions are allowed. Moreover, these examples will show that some modification of the previous proposals is necessary to capture the observed facts. The relevant sentences are those where the relative clause is introduced by transitive verbs of emotionFootnote ⁶ such as beundra ‘admire’, avguda ‘adore’, avsky ‘hate’, gilla ‘like’ and respektera ‘respect’, and where a causal relationship exists between the two clauses. Some authentic examples (found on the Internet) will illustrate the point:

1. (14)

   

2. (15)

   

3. (16)

   

4. (17)

   

In fact, it seems quite straightforward to construct examples involving such verbs which are perfectly natural (in a suitable context), as long as the causal relationship between the matrix and embedded clauses is maintained. In other words, there is reason to believe that we are dealing with some kind of productive constructional schema here. The invented sentence (18), for example, seems to be judged as completely acceptable by most speakers, and even (19), which represents an ‘extended’ instance of the construction where the matrix verb has been replaced by an adjectival construction, is regarded as fully acceptable by many speakers (including myself):Footnote ⁷

1. (18)

   

2. (19)

   

In relation to the dominance-based proposals of Erteschik-Shir (Reference Erteschik-Shir1982), Van Valin (Reference Van Valin2005) and Goldberg (Reference Goldberg2006), the interesting fact here is that the matrix clauses rather than the relative clauses appear to represent the informational center of the utterances. A dominance test applied to (14), for example, confirms this intuition:

1. (20) Speaker A: Jag beundrar dom som klarar av det.

   ‘I admire those who get through that.’

   Speaker B: Det är lögn – *ingen klarar av det.

   ‘That's a lie – *nobody gets through that.’

   Det är lögn – du beundrar dom inte.

   ‘That's a lie – you don't admire them.’

Thus, as shown by the lie test, negating the relative clauses in this way does not lead to a felicitous response, indicating that such relative clauses are non-dominant (i.e. backgrounded/presupposed). Furthermore, that a causal relationship between the clauses appears to be necessary to produce acceptable sentences involving matrix clause verbs of emotion is indicated by the stark acceptability contrast between (22) and (23), as replies to the utterance in (21):Footnote ⁸

1. (21) Ser du det stora huset bakom träden?

   ‘Do you see the big house behind the trees?’

2. (22)

   

3. (23)

   

Thus, while (22) is generally perceived as perfectly natural by Swedish speakers, this is not the case with (23). The crucial difference between the two sentences lies in the fact that while the ‘envy’ described by the matrix verb in (22) is causally related to the state described by the embedded clause, such a causal relation does not exist in (23), where the embedded clause simply specifies or identifies the target. Although it cannot be completely ruled out that there are acceptable sentences involving transitive emotion verbs where no causal relation exists, I have not found any such authentic examples, nor have I been able to construct one. An explanation for this difference in acceptability based on pragmatic dominance would have to state that the causal relationship between the clauses improves acceptability due to the resulting relative clauses being foregrounded, even though dominance tests contradict this. Still, it might perhaps be possible to argue that the embedded clause in (22) is less backgrounded than that in (23) and that this leads to the difference in acceptability, but the fact that the dominance test fails in all these examples arguably makes Erteschik-Shir's proposal lose some of its explanatory force. Note, for example, that according to Erteschik-Shir (Reference Erteschik-Shir1982:178), stressing the matrix verb in these structures will ‘force a dominant interpretation of the matrix and block an interpretation of the relative clause as being dominant’. However, in the cause-related extraction examples described here, the natural sentence stress falls precisely on these matrix verbs, but the extractions are nevertheless fully acceptable.

Furthermore, while the examples provided here may not represent conclusive evidence that the mental representation of these constructions cannot be based on degree of dominance, I will argue in Section 3 that there is a simpler and more straightforward way of accounting for the relevant facts. Above all, this alternative explanation fits more naturally with the existence of crosslinguistic variation, and it has the strength of being based on already existing principles of a linguistic theory (Construction Grammar).

2.2 Processing restrictions

Kluender (Reference Kluender1998), Levine & Hukari (Reference Levine and Hukari2006) and Hofmeister & Sag (Reference Hofmeister and Sag2010), among others, argue that constraints on extraction are best explained in terms of processing difficulties. Kluender (Reference Kluender1998:257), for example, argues that the ‘only’ problem in these sentences concerns processing complexity, and gives (24) as an example of an acceptable English sentence, contrasted with the less acceptable sentence in (25), which contains referential material intervening between the filler and the gap (> symbolizing ‘is better than’):

1. (24) That's the article that we need to find someone who understands. >

2. (25) That's the article that we need to find the reviewer who understands.

His claim is that if the individual words and phrases, as well as the context, are adapted to produce maximally efficient processing, then extraction from complex NPs is perfectly fine in English. Addressing crosslinguistic variation and previous claims concerning the Scandinavian languages, Kluender (Reference Kluender1998:257) states the following:

Yet, Kluender (Reference Kluender1998:268) then claims that ‘[i]t is also well known that subjacency is “parameterized” across languages, which in practical terms means that a standard subjacency violation in one language may be perfectly grammatical in another. If this is the case, a processing account of subjacency will have to be flexible and all-encompassing enough to explain these crosslinguistic differences in subjacency effects’. However, Kluender also states that ‘apparent crosslinguistic differences have often turned out to be not so different after all’, and specifically argues that the claim that subjacency must be defined differently in the Scandinavian languages is incorrect, since ‘it has since been shown that English exhibits the same range of so-called exceptions’ (Reference Kluender1998:268).

Levine & Hukari (Reference Levine and Hukari2006:262) elaborate on Kluender's analysis and state that the (so-called) complex NP constraint is ‘just another instance of processing failure’. In fact, Levine & Hukari (Reference Levine and Hukari2006:385) argue that ‘most of the effects which have led investigators to posit a dichotomy within the class of Unbounded Dependency Constructions are not syntactic in nature’, but instead explainable in terms of ‘pragmatics, computation and real-time processing’.

One of the more recent processing-based accounts is that of Hofmeister & Sag (Reference Hofmeister and Sag2010). Since this is a clearly important and exceptionally detailed study, it is worth spending some time on discussing its main claims. Hofmeister & Sag argue that ‘island-violating constructions involve numerous processing pressures that aggregate to drive processing difficulty above a threshold, resulting in unacceptability’ (Reference Hofmeister and Sag2010:366). They subsequently argue that competence grammars ‘overgenerate significantly’, and that processing factors (including semantic and pragmatic effects) explain why only some of the syntactically licensed sentences are perceived as fully acceptable (Reference Hofmeister and Sag2010:368).

While Hofmeister & Sag are somewhat unclear with regard to how ‘syntactic constraints’ interact with processing complexity, and in what sense processing factors ‘include’ semantic and pragmatic effects,Footnote ⁹ their introduction clearly indicates that in their account, unacceptability of e.g. relative clause extractions stems from cognitive resource limitations (Hofmeister & Sag Reference Hofmeister and Sag2010:367):

In order to assess the explanatory potential of this interesting idea, their statement that the crucial factor concerns limitations on speakers’ cognitive resources must be discussed at some length. In particular, I will address here the question of whether this could be reconciled with the existence of crosslinguistic variation in the acceptability of relative clause extraction.

First of all, concerning crosslinguistic differences such as those found between English and Swedish, Hofmeister & Sag seem to argue that there is no syntactic difference between such languages with respect to relative clause extraction. Instead, Hofmeister & Sag (Reference Hofmeister and Sag2010:367) state that ‘the variation in acceptability judgments associated with these constructions, both language-internally and crosslinguistically, can be better explained by appealing to cognitive constraints on language processing’. Note, however, that the constraint they suggest as relevant for variation of acceptability judgments within a language appears to be based on the idea that speakers ‘learn to accept’ processing complexity in a structure:

Yet, the question is whether this type of constraint is sufficient to explain the general pattern of relative clause extraction acceptability in English. Although Hofmeister & Sag do argue that processing factors include semantic and pragmatic effects, it is evident that these are claimed to affect real-time judgments of acceptability (what they refer to as ‘performance-based factors’), rather than conventionalized linguistic forms (‘competence-based factors’). Thus, while the experiments described in Kluender (Reference Kluender1998) and Hofmeister & Sag (Reference Hofmeister and Sag2010) convincingly show that structures which are difficult to process are associated with degraded acceptability ratings, the crucial issue concerns whether such processing restrictions can replace conventionalized syntactic island-constraints (of some kind).Footnote ¹⁰

In the remaining part of this section I will point out a few direct consequences of the processing-based approach as described by Hofmeister & Sag (Reference Hofmeister and Sag2010), in particular in relation to the existence of crosslinguistic variation between English and Swedish. Although I believe that these consequences lead to a somewhat unlikely view of the mental representation of linguistic knowledge, I will not provide empirical evidence against this view. Instead, I will simply argue that it seems worthwhile investigating whether the acceptability judgments as well as the relevant crosslinguistic variation could be explained in an alternative or modified way where such consequences are avoided (to be explored in Section 3).

First of all, Hofmeister & Sag (Reference Hofmeister and Sag2010:367) argue that the processing constraints in extractions are basically of the same type as those found in e.g. garden path structures, such as The horse raced past the barn fell. However, an important property of such structures is that the perception of degraded acceptability generally disappears on a second and third hearing/reading. The reason is that once listeners/readers have familiarized themselves with the structure of the utterance (which is generally ambiguous), processing factors no longer negatively affect acceptability. On the assumption that the acceptability of relative clause extraction is also degraded due to processing difficulties, one might expect such structures to improve in a similar way once speakers have familiarized themselves with the structure of the sentence and the intended meaning of the utterance. However, although experiments involving extractions regularly show that acceptability does increase with repeated exposure (see for example Hofmeister & Sag Reference Hofmeister and Sag2010:402–403), it does not seem that acceptability increases to the same extent as in garden path sentences (where the immediate result is a perfectly acceptable sentence). Of course, it is theoretically possible that English extractions are so much more complex than garden path structures that full acceptability is rarely reached. However, such a view would have the somewhat surprising consequence that (truly) bilingual Swedish–English speakers could not have different acceptability judgments for relative clause extractions in the two languages.Footnote ¹¹ While this does not rule out language-specific processing complexity cut-off points in principle, it does point against an account of processing complexity cut-off points based on cognitive resource limitations, as implied by Hofmeister & Sag (Reference Hofmeister and Sag2010:367). In other words, it seems unlikely that the only reason English speakers do not generally accept relative clause extractions is because they find them too difficult to process.

The main question, therefore, concerns whether there might exist some form of conventionalized ‘processing complexity cut-off points’ in different languages. Yet, there is little indication that e.g. Swedish speakers are less sensitive than English speakers to processing complexity at some general level, for example concerning garden path structures or many other types of extractions. The claim must therefore be reduced to concern different processing complexity cut-off points within the specific category of relative clause extraction constructions. However, when the constraint is reduced to affect only one particular construction, it is no longer clear how it relates to the cognitive resource limitations suggested by Hofmeister & Sag, and the proposal seems to lose much of its independent motivation. Thus, in the next section I will argue that, if what is conventionalized is a construction-specific constraint anyway, its mental representation could be reduced to already existing principles in Construction Grammar, with no need to introduce the additional theoretical concept of conventionalized language- and construction-specific processing complexity cut-off points.

3.1 Theoretical framework

Although there are a number of different versions of construction-based theories, such as Construction Grammar (Goldberg Reference Goldberg1995), Radical Construction Grammar (Croft Reference Croft2001) and Cognitive Grammar (Langacker Reference Langacker1987), they are unified by the idea that grammatical structures (‘constructions’) are pairings of meaning and form, basically in the same way as in the case of words and more complex lexical items. As a consequence, there is argued to be a continuum of linguistic constructions from completely phonologically specified to completely schematic ones, where the latter include what has traditionally been regarded as syntactic rules. The version of Construction Grammar adopted here is essentially that of Radical Construction Grammar (Croft Reference Croft2001), with the slight modification and theoretical clarification presented in Verhagen (Reference Verhagen2009). In this version of the theory, linguistic constructions are made up of form–meaning pairings, as in all Construction Grammar theories. For the present purposes, the crucial aspect concerns the form-side of constructions, and in particular what affects speakers’ perception of what represents acceptable use of a construction.

Verhagen (Reference Verhagen2009:140) makes the following point concerning what it means to know a construction in one's language:

Elaborating on the principles of Croft (Reference Croft2001) in particular, Verhagen (Reference Verhagen2009:140) argues that such schematic slots in constructions represent ‘the traditional structuralist notion of “paradigm”’, and that the notion of paradigm is ‘indispensable in a comprehensive theory of grammatical constructions’. Verhagen (Reference Verhagen2009:145) concludes that the form-side of a construction is organized on the basis of phonological forms (morphemes) and conceptually defined schematic categories (paradigms), where the latter essentially replace the traditional notion of ‘syntactic category’.

In order to clarify these suggestions, consider the following example:

1. (26) I enjoy reading.

   [ACTOR] [enjoy] [EVENT/STATE -ing]

The specific schema in (26) presumably exists as a conventionalized construction in English, which means that it is both associated with a specific meaning and that it specifies required properties of its various slots. As mentioned above, the specifications of the slots in the construction could be of two types, namely morphemesFootnote ¹² (enjoy and the -ing ending) and paradigms (ACTOR and EVENT/STATE). In line with Croft (Reference Croft2001), Verhagen (Reference Verhagen2009), and others, paradigms are represented by schematic conceptual representations based on generalizations across encountered instances of the construction. In combination, these constructional specifications give rise to speakers’ acceptability judgments, for example the fact that the deviating morpheme in I enjoy *to read is generally judged unacceptable by English speakers. It is important to emphasize that these are not specifications ‘on top of’ syntactic structure, instead these constructional specifications are the syntactic structures.

Note that in some specific cases a paradigm in one construction may coincide with a paradigm in another construction, or an existing construction may represent a paradigm in a more complex construction (see Verhagen Reference Verhagen2009:144–147). However, usually the paradigm that defines some specific slot in a specific construction is unique to that construction, and this is equally true within and across languages (Croft Reference Croft2001:34ff.). Consider the following constructional schema in English associated with ‘transfer by using ballistic motion’ (Goldberg Reference Goldberg1995, Croft Reference Croft, Cuyckens, Berg, Dirven and Panther2003):

1. (27) English Ballistic Motion Construction (an instance of the ditransitive give-construction)

   [ACTOR] [BALLISTIC MOTION EVENT] [RECIPIENT] [THEME]

   Example: I threw him the box.

It must be noted here that the rough constructional specifications in (27) are mainly meant as descriptive labels, and they certainly do not capture the full conceptual representation of the paradigms conventionalized in the mental grammar of English speakers. Instead, the ‘true’ mental representation of a paradigm includes all elements one has ever encountered in the same constructional environment, as well as a conceptual generalization across these elements (leading to the productive possibility of using novel elements that are sufficiently similar to the previously encountered ones). The consequence is that the conceptual representation of the ‘Ballistic Motion Event’ paradigm in (27) blocks the acceptability of non-ballistic verbs such as push, presumably because the semantic properties of push are not sufficiently similar to the ‘ballistic motion’ aspect of the conceptual specification in the construction, as seen in (28)–(31):Footnote ¹³

1. (28) I threw the box to him.

2. (29) I pushed the box to him.

3. (30) I threw him the box.

4. (31) *I pushed him the box.

However, if the construction with the form of (30) is compared with the structurally parallel Swedish construction, it is immediately clear that while the Swedish construction includes verbs such as ge ‘give’ and some other transfer verbs, as seen in the acceptability of (33), the ballistic motion aspect has not been conventionalized, leading to the unacceptability of utterances such as (35):

1. (32) Jag gav asken till honom.

   ‘I gave the box to him.’

2. (33) Jag gav honom asken.

   ‘I gave him the box.’

3. (34) Jag kastade asken till honom.

   ‘I threw the box to him.’

4. (35) *Jag kastade honom asken.

   ‘I threw him the box.’

In line with the above discussion, therefore, the constructions in the two languages arguably differ in terms of the conceptual specification of the event paradigm, where English has a more ‘open’ or ‘extended’ conceptual frame in comparison with Swedish (see Löwenadler Reference Löwenadler, Bergh, Bowen and Mobärg2013 for further discussion of this point).

Finally, as noted by Verhagen (Reference Verhagen2009:140–141), not all conceptual specifications of constructional slots can be specified as independent properties of individual slots, but there are also ‘syntagmatically’ defined properties. Such syntagmatic constraints can be associated with both the phonological (morpheme) specifications and the conceptual (paradigm) specifications. While the former can be exemplified by e.g. morphological number agreement between different parts of the structure, the latter type can be exemplified by the following contrastive pairs:

1. (36) The train screeched into the station.

2. (37) *The dog barked into the room. (Goldberg Reference Goldberg1995:62)

3. (38) The car squealed around the corner.

4. (39) *The car honked around the corner. (Jackendoff Reference Jackendoff2002:290)

As Jackendoff (Reference Jackendoff2002:290) notes, whether or not a particular verb is possible in this specific construction appears to be decided by whether ‘the sound emission can be associated directly with the action of moving’; in other words, a requirement for acceptability is that a causal relation exists between the motion expressed by the sentence and the sound expressed by the event verb. As I will argue below, similar types of syntagmatic constraints represent crucial aspects of the relative clause extraction constructions analyzed in the present paper.

3.2 Constraints in relative clause extraction constructions

Following this short introduction to some relevant principles of Construction Grammar, we now turn to the main issue, namely what kind of constraints are associated with relative clause extraction constructions. The argument that will be made here is that acceptability judgments in these constructions are determined by the same principles that are relevant in non-extraction constructions, and that no additional principles concerning language-specific cut-off points based on processing complexity or pragmatic dominance are required.

Focusing first on paradigmatic conceptual aspects of the constructions (i.e. specifications concerning the conceptual content of individual slots), it seems clear that there are differences at least concerning the possible pragmatic functions of the extracted element (or filler argument). As pointed out by Engdahl (Reference Engdahl1997:73), for example, fillers which are continuous topics do not normally exist in English, whereas such fillers are common in Swedish structures of various types. In such sentences, the initial pronominal filler functions as a non-contrastive pronominal topic, linking the sentence in which it occurs to a previous sentence (Engdahl Reference Engdahl1997:72). Consider the following piece of discourse in Swedish, where there is no acceptable utterance in English parallel to (41) in structure and meaning:

1. (40)

   

2. (41)

   

It turns out that continuous topic fillers often occur in Swedish relative clause extraction constructions as well, as in example (42) (see Lindahl Reference Lindahl2010 for other authentic examples).

1. (42)

   

The occurrence of such examples in Swedish indicates that there is at least one conceptual paradigm in relative clause extraction where the conventionalized specifications of English and Swedish differ from each other. However, it is equally clear that this cannot be the only crucial factor, since many authentic and fully acceptable Swedish examples discussed here and elsewhere in the literature include focused or contrastive (rather than topical) fillers. Therefore, following the descriptive format of Verhagen (Reference Verhagen2009:140), I will attempt to sketch a tentative (and obviously in no way complete) formal representation of the three related types of Swedish relative clause extraction constructions discussed here, namely (i) the Presentational complex NP extraction construction, (ii) the Focused relative clause complex NP extraction construction, and (iii) the Cause-related complex NP extraction construc-tion.Footnote ¹⁴

First of all, it must be made clear that the constructional schemas in Tables 1–3 below only contain the relevant and most important characteristics of the constructions.Footnote ¹⁵ For example, the verbal specifications of ‘perception’ and ‘emotion’ must probably be defined in a more restrictive way in order to accurately represent the conventionalized instances and the schematized conceptual properties. Furthermore, it is highly likely that Swedish speakers differ in the extent to which they accept these structures (i.e. the extent to which these constructions exist as representations in their mental grammar). While this is an interesting area for empirical research, the general specifications in Tables 1–3 will suffice to illustrate the theoretical point developed here.

Table 1. Presentational complex NP extraction construction in Swedish.

Table 2. Focused relative clause complex NP extraction construction in Swedish.

Table 3. Cause-related complex NP extraction construction in Swedish.

An important point to note is that the specifications of the individual paradigms in the constructions are likely to be quite general. For example, slot (vi) in the constructions is simply specified as ‘any verbal element’ giving rise to some semantic/pragmatic relationship in the context of the whole sentence. This reflects the fact that the conceptual constructional constraints are often syntagmatic in nature, as specified by the constraints in (vii) and (viii) in Tables 1–3 (see Section 3.1 above). On the other hand, empirical studies such as Lindahl (Reference Lindahl2010), as well as the common characteristics of invented sentences in previous studies, indicate that there are clear tendencies concerning the types of elements that occur in the various slots. Yet, these tendencies do not necessarily reflect ‘true’ paradigmatic constraints, since other types are often possible as long as the syntagmatic constraints specifying the semantic and pragmatic relationships between the sentence parts are satisfied. Instead, the tendencies appear because certain fillers for various semantic and pragmatic reasons correlate with the general (‘whole-sentence’) meanings specified by these constructions.

In the constructions described in Table 1 and Table 2, a crucial conceptual specification concerns the distribution of the informational center of the utterance, which is of course much in line with the dominance-based proposal suggested by Erteschik-Shir, Goldberg and Van Valin. However, with this kind of schema, the mental representation of the constructional possibilities in different languages can also be associated (in the same way) with constraints related to the individual paradigms in the construction, where presumably English is more restrictive than the Scandinavian languages, for example in the range of possible filler types in slot (i) and matrix verbs in slot (ii). In that sense, the concept of dominance (or ‘informational center’) is related to the constructional specifications just like other paradigmatic and syntagmatic properties.

This is particularly important in relation to the cause-related construction described in Table 3. Here, in addition to the paradigmatic specifications associated with the individual slots in the construction (and there may be several others in addition to those indicated here), there is the crucial syntagmatic requirement that (vii) and (viii) express the specified causal relationship. By contrast, there are no requirements in this construction that the relative clause must represent the semantic/pragmatic focus of the utterance (as required in the dominance-based approaches discussed in Section 2.1 above); instead, the pragmatic focus lies on the (stressed) matrix verb. Goldberg (Reference Goldberg2006:130), for example, specifically states that elements that are part of presupposed clauses are backgrounded and therefore unsuitable for extraction, yet the construction in Table 3 appears to be available as a productive schema for Swedish speakers, as seen in Section 2. Crucially, once speakers have acquired the schematic meaning and specifications of this construction, such sentences are presumably no more complex to process than presentational relative clause extractions, even though in this case the relative clause appears to be non-dominant.Footnote ¹⁶

It is interesting to note here the parallelism with the syntagmatic specification in the ‘sound-emission’ construction referred to in Section 3.1, with the examples repeated here for convenience:

1. (43) The train screeched into the station.

2. (44) *The dog barked into the room.   (Goldberg Reference Goldberg1995:62)

3. (45) The car squealed around the corner

4. (46) *The car honked around the corner   (Jackendoff Reference Jackendoff2002:290)

As noted in Section 3.1, a necessary syntagmatic requirement in this construction appears to be that the sound associated with the verb is caused by the motion described in the sentence. In other words, the phenomenon that a specific causal relationship between some parts of a construction may be necessary for acceptability is already observed in other parts of the grammatical system and has a natural place in Construction Grammar theories.

3.3 Syntactic conventionality versus real-time processing

Given the observations so far, a relevant question to ask is whether it is possible that the dominance-based explanation is correct in principle, only incorrectly operationalized. In other words, could it be that Erteschik Shir's ‘lie test’ and Goldberg's ‘negation test’ are insufficient as indicators of whether an embedded clause is ‘dominant’ or belongs to the ‘potential focus domain’? I would argue that although this is possible in principle, there seems to be a risk of circularity if sentences representing acceptable extractions are themselves used to define the notion of pragmatic dominance, and the proposal clearly loses some explanatory force if no independent test can verify this concept.

It should be pointed out here that, while I have argued that the concept of dominance and processing restrictions cannot directly explain conventionalized aspects of island constraints, such factors certainly appear to affect ‘real-time’ acceptability in the sense that acceptability may be degraded if structures are difficult to comprehend or process. However, one of the main arguments in the present paper is that, in order to explain crosslinguistic variation and differences in acceptability judgments between specific instances of these extractions, one must focus on the compatibility between conventionalized constructional specifications and the meanings and forms used in an individual utterance.

To develop this argument, an important difference between the explanations of Erteschik-Shir (Reference Erteschik-Shir1982) on one hand and Kluender (Reference Kluender1998), Levine & Hukari (Reference Levine and Hukari2006) and Hofmeister & Sag (Reference Hofmeister and Sag2010) on the other is that while the former regards complex NP extraction in English as syntactically ungrammatical, the latter regard such extractions as syntactically grammatical. As pointed out in Section 2.2, Hofmeister & Sag (Reference Hofmeister and Sag2010:368) argue that ‘competence grammars overgenerate significantly, leaving it to processing and other factors (including semantic and pragmatic effects) to explain why only a proper subset of the grammatically licensed sentences . . . are judged as fully acceptable’, and an essentially similar view is presented in Levine & Hukari (Reference Levine and Hukari2006:262). By contrast, in Erteschik-Shir (Reference Erteschik-Shir1982) such English structures are claimed to be syntactically inaccessible for extraction, but perceived as relatively acceptable because they have suitable semantic/pragmatic properties.

However, in the present account the dichotomy between syntactic constraints and processing/dominance effects largely disappears. Consider for example the following Swedish sentences, for example as replies to the Swedish equivalent of the utterance Check out these nice shoes in the picture!:Footnote ¹⁷

1. (47)

   

2. (48)

   

3. (49)

   

Despite their semantic and pragmatic similarity, only (47) is possible in Swedish, even though all of (47)–(49) are perfectly acceptable in the related structures with the filler såna skor ‘such shoes’ in the more canonical position at the end of the sentence. Arguably, the reason is (at least partly) that the presence of the phonological form där in (48) and (49) makes these two structures incompatible with the phonological specifications in the constructional schema described in Table 2.Footnote ¹⁸ On this account, the question of whether relative clause extraction constructions are syntactically grammatical or ungrammatical misses the point, since conceptually defined paradigms and syntagmatic relationships along with phonological specifications are sufficient to determine the acceptability of the construction. In other words, the relevant component(s) determining the acceptability of specific relative clause extractions should not be defined as ‘the interaction between syntactic grammaticality/ungrammaticality and language-specific conventionalized processing/dominance cut-off points’, but instead as ‘degree of compatibility with language-specific conventionalized constructional phonological/conceptual specifications’. In the production and reception of real-time utterances there are clearly also additional processing complexity effects involved (of the type investigated by Kluender and Hofmeister & Sag, and others), but there is no relevant sense in which it could be argued that Scandinavian speakers have a greater tolerance of processing complexity as such, in comparison with English speakers.

A final question, then, concerns how the difference between relative clause extraction in English and in Swedish is mentally represented more specifically. In this and previous sections, the characterization of the constructional specifications (albeit tentative and incomplete) has mainly been focused on specifying relative clause extraction possibilities in Swedish. The reason is that, except for the fact that there is essentially agreement that extraction is more prevalent in the Scandinavian languages, there appears to be no clear consensus concerning the acceptability status of parallel English examples (for example concerning sentences such as (3)–(5) above). Still, as has been argued in this paper, what can be said with some certainty is that the cause-related relative clause extraction construction which exists in Swedish does not exist in English. It is argued here that the general difference between English and Swedish acceptability judgments in relative clause extraction reflects the extent to which the forms in the constructional slots and the relationship between these forms match the conceptual and phonological paradigmatic and syntagmatic specifications of some conventionalized relative clause extraction construction in the language.Footnote ¹⁹ In Swedish, in comparison with English, such conventionalized constructions appear to be more deeply entrenched among the speakers and show greater variation in terms of the conceptually defined paradigms and syntagmatic relations.