We propose here a new PF account of the that-trace effect. Adopting the phase-based theory of the syntax-phonology interface and independent principles of prosodic restructuring, we propose that the complementizer that cannot form a prosodic phrase on its own. We show that this analysis straightforwardly derives the core paradigm surrounding the that-trace effect and its well-documented exceptions triggered by focus, adverbs, parentheticals, resumption, and right node raising. We further argue that the relevant prosodic condition can be derived from the interaction of the Lexical Category Condition (Truckenbrodt 1999) with Prosodic Vacuity (Kandybowicz 2015), within the Match Theory of syntax-prosody correspondence (Selkirk 2011).
Since the advent of the Minimalist Program (Chomsky 1995), various phenomena once treated in exclusively syntactic terms have been reanalyzed in terms of the syntax-phonology interface (An 2007a,b, Bošković 2001, 2011, Kandybowicz 2006, 2009, Richards 2010). In this article, we investigate further consequences of the general interface approach to the that-trace effect (that-t effect)—in which the acceptability of embedded subject extraction declines sharply in the presence of the overt complementizer (except in relative clauses) in some languages—which has steadfastly resisted a purely syntactic explanation despite more than three decades of intensive generative research. Our analysis adopts the phase-based theory of prosodic mapping laid out in Dobashi 2003, Kahnemuyipour 2004, and Kratzer and Selkirk 2007, coupled with well-defined principles of prosodic restructuring (Kenesei and Vogel 1995, Nespor and Vogel 1986). We propose that the that-t effect is due to an ill-formed prosodic representation that violates the PF condition in (1).
(1) Function words cannot form a prosodic phrase on their own.
(1) has its roots in Truckenbrodt’s (1999) Lexical Category Condition and Selkirk’s (1984, 1986, 1996) Principle of the Categorial Invisibility of Function Words, both of which state that function words are invisible to the syntax-prosody mapping. Bruening (2009) independently develops a proposal similar to the one offered here. Bruening draws on Postal’s (2004) empirical observation that the relevant effect persists even in as-parentheticals, which are not derived by subject extraction but instead involve a base-generated null (expletive) subject. This observation shows that the so-called that-t effect is not about subject extraction per se; rather, it is about unpronounced subjects (whether they are extracted or base-generated null elements). Bruening then proposes that the effect arises when the overt complementizer that is the sole member of its containing prosodic group, though he does not discuss why this must be so. Our proposed analysis is intended to be a further refinement of Bruening’s proposal, applied to a wider range of empirical data regarding the that-t effect. At the same time, we go beyond Bruening’s proposal and argue that the condition in (1) itself is derivable from the interaction of independently motivated conditions imposed on the syntax-prosody mapping such as Prosodic Vacuity (Kandybowicz 2015) and the Lexical Category Condition (Truckenbrodt 1999).
The article is structured as follows. In section 2, we outline a phase-based theory of default prosodic mapping proposed by Dobashi (2003). In sections 3–5, we demonstrate that (1) not only captures the core observations regarding the that-t effect, but also allows for a unified explanation of various mitigating circumstances (e.g., adverbs, parentheticals, resumption, and right node raising) that otherwise appear unrelated. In so doing, we compare our analysis with Kandybowicz’s (2006, 2009) alternative PF interface account of the that-t effect. Section 6 is the conclusion.
2 A Phase-Based Theory of Default Prosodic Phrasing
A growing body of work within phase theory (Chomsky 2000) has argued for a minimal theory of the syntax-prosody mapping according to which the spelled-out domain of a phase head (C or v)—TP and VP—is marked as a potential prosodic domain; see Adger 2007, Kahnemuyipour 2004, Kratzer and Selkirk 2007, and Sato 2012 for different formulations of the Nuclear Sentence Stress Rule within this version of phase theory. For the purposes of this article, we outline Dobashi’s (2003) approach to phonological phrasing within thi s minimal theory of the syntaxprosody mapping. Dobashi first assumes that Spell-Out defines a linear order among lexical items when it sends the syntactic derivation to the phonological component. Consider the phase-theoretic derivation of the structure [CP [TP Subj T [vP v [VP V Obj]]]]. When Spell-Out applies to the VP domain, it yields the V >> Obj order (where >> stands for precedence). When Spell-Out applies to the TP domain, it yields the Subj >> T >> v order. The next step in linearization then is to fix a linear order between the strings derived by the two successive applications of Spell-Out: V >> Obj and Subj >> T >> v. It is not immediately obvious, however, how the two strings can be ordered with respect to each other, because there is no a priori reason why the string created later in the syntactic derivation must precede the string created earlier. Dobashi calls this the assembly problem. Dobashi points out that to solve this linearization problem, it is necessary for the two surface strings to have a shared member. Consider the schematic strings in (3).
It is important to note here that the two strings in (3a) and (3b) can be ordered with respect to each other by virtue of the shared member p so that the resulting final string in (3c) is consistent with both (3a) and (3b). On the basis of this consideration, Dobashi proposes that when Spell-Out sends a linearly ordered string to the phonological component, the initial element in the string stays behind so that that element can be accessible to the next application of Spell-Out and serve as a shared element to fix the order between two strings that otherwise would be unlinearizable. Under this system, then, when Spell-Out applies to the VP level, it yields the phonological phrase (Φ-phrase) in (4a), because V escapes the mapping to remain accessible to the next application of Spell-Out for the purposes of linearization. Similarly, the subject escapes the mapping at the TP level, adding another Φ-phrase as shown in (4b). Finally, when the rest of the derivation is spelled out, the final Φ-phrase results, as shown in (4c).
(T v V)Φ (Obj)Φ
(C Subj)Φ (T v V)Φ (Obj)Φ
In this article, we assume, following Kratzer and Selkirk (2007), that Spell-Out yields only a skeletal prosodic representation (call it a pf ) for the purposes of prosodic phrasing. In other words, cumulative results of the successive applications of Spell-Out may be subject to further domain-specific adjustments to give rise to the complete phonological form (call it a PF) before it is externalized for use by the articulatory-perceptual system. This model of the syntax-phonology interface is schematically illustrated in (5).
(5) Phasal Spell-Out ⇒ Pfs ⇒ PF ⇒ Externalization
In the next section, we show that the that-t effect and its well-known “exceptions” receive a unified account under this hybrid model, which combines key insights from phase theory and independently motivated postsyntactic prosodic adjustment rules.
3 Prosodic Phrasing and the That-t Effect
Whoi do you think t′i (*that) ti wrote the book?
Whati do you think t′i (that) Bill wrote ti?
Wheni do you think t′i (that) Bill wrote the book ti?
Since the advent of Government-Binding Theory, this subject-nonsubject asymmetry has standardly been assumed to arise from the Empty Category Principle (ECP), which states that all traces must be properly governed. Leaving details aside (see Lasnik and Saito 1984, 1992), object traces and adjunct traces are always properly governed, whereas subject traces are not owing to the intervention of the overt complementizer that. The ECP-based analysis cannot be formulated within the Minimalist framework, which attempts to dispense with the notion of government. Our PF account correctly predicts the asymmetry in question without relying on government as its central analytical premise. Within our phase-based theory of prosodic mapping, the examples in (6a–c) are assigned the default phonological phrasing shown in (7a–c), respectively.
*Whoi do you think (thatti)Φ (wrote)Φ (the book)Φ?
Whati do you think (that Bill)Φ (wrote ti)Φ?
Wheni do you think (that Bill)Φ (wrote)Φ (the book)Φti?
(7a) is ill-formed because the Φ-phrase, (that ti)Φ, violates (1). This violation does not occur in (7b–c), where that forms a Φ-phrase in construction with the lexical item Bill, which immediately follows it. Our account also correctly predicts that the that-t effect remains even when T or Neg heads are overtly filled, as shown in (8a–b), because both heads are grouped within the Φ-phrase that contains the verb in the embedded clause under our current system.
*Whoi do you think (thatti)Φ (will/should/might/could write)Φ (a book)Φ?
*Whoi do you think (thatti)Φ (didn’t write)Φ (a book)Φ?
There is an important piece of independent evidence showing that the that-t effect results from violating a PF principle such as (1), rather than exclusively from violating purely syntactic principles such as the ECP. Merchant (2001) (see also Ross 1969) proposes that sluicing, illustrated in (9), is derived through regular wh-movement, followed by TP-ellipsis at PF.
(9) Someone just left. – Guess whoi
[TPti just left].
Our analysis thus predicts that the that-t effect should be mitigated once the offending Φ-phrase in (7a) is removed by PF ellipsis. This prediction is indeed borne out (Merchant 2001, Perlmutter 1971), as shown by the contrast in (10a–b).
a. *It’s probable that a certain senator will resign, but [DP which senator]i
[TP it’s probable that ti will resign] is still a secret.
b. It’s probable that a certain senator will resign, but [DP which senator]i
[TP it’s probable that ti will resign]is still a secret.
If the that-t effect were syntactic in nature, we would wrongly predict no contrast in (10a–b) because an ECP-like violation is a strictly syntactic violation that cannot be repaired. The contrast therefore suggests that the relevant effect is a prosodic phenomenon. See also Bruening 2009 and Postal 2004 for further evidence from as-parentheticals showing that the effect in question is not due to subject extraction per se.
4 Prosodic Restructuring and the That-t Effect
It is well-known that the that-t effect is mitigated in several contexts besides ellipsis. In this section, we show that our analysis provides a simple account of these facts.
4.1 Adverbs, Parenthetical Expressions, and Resumptive Pronouns
The best-known mitigating environment for the that-t effect is the so-called adverb effect (Bresnan 1977, Culicover 1993): the violation is somehow suspended when a sentential adverb is inserted between that and the subject trace, as in (11).
Whoi do you think [CP that after years and years of cheating death [TPti finally died]]?
This effect is a straightforward consequence of (1). As Sobin (2000:537) notes, the adverb allows that to be prosodically incorporated into the intonational phrase (I-phrase) to its right, as illustrated in (12), a pattern that is clearly indicated by the comma intonation superimposed on the first I-phrase.
(12) Who do you think (that after years and years of cheating death)I (finally died)Φ?
The prosodic incorporation of sentential adverbs suggested here is independently supported by coordination. Sobin (2000, 2002) observes that the that-adverb sequence can be coordinated, as illustrated in (13).
The lawyers claimed (that on July 4)I—and (that on July 5)I—(Bill was in Rhyl)I.
One might suspect that this example could be dealt with by saying that the string Bill was in Rhyl undergoes right node raising (RNR). Sobin shows that this analysis is incorrect. RNR may apply to a sentence when the conjoined sequences bear contrastive focus, as shown by the contrast between (14a) and (14b).
The lawyers claimed that on July 4, and Mary testified that on July 5, Bill was in Rhyl.
*The lawyers claimed that on July 4, and the lawyers claimed that on July 5, Bill was in Rhyl.
(14b) shows that the contrast between the two temporal PPs is not sufficient to license the application of RNR. The grammaticality of (13), however, naturally follows from our proposal if the coordination is at the level of the I-phrase, which arises through the prosodic incorporation of the C.
Our analysis makes another important prediction. Sobin (2000) observes that the adverb effect looks just like a parentheticalization in that a that-adverb sequence is prosodically set off as if it were not present in the syntactic derivation of the host clause. Our analysis then predicts that the that-t effect should also be lessened by parentheticals. As Ackema (2011) notes, example (15) bears out this prediction.
Whoi do you think that, according to the latest rumors, ti is quitting politics?
Our native-speaker consultants indeed report that this example sounds best with parenthetical intonation, namely, phrase boundaries at the commas, and pitch lowering over the parenthetical phrase marked by a pitch accent on the first syllable of rumors and an L-H% rising boundary tone on its second syllable. Note that this observation provides further evidence against syntactic explanations of the that-t effect. Parentheticals are known to be invisible to the syntactic wellformedness of the host structure (Espinal 1991, Haegeman 1988, Potts 2002). For example, asparentheticals do not disrupt the adjacency required between a transitive verb and its direct object, as illustrated in (16a). This example directly contrasts with (16b), where the intervening VP-level adverb passionately disrupts the required adjacency.
Susan loves, as we are all aware, silly books.
*Susan loves passionately silly books.
Given the invisibility of as-parentheticals to the syntactic derivation of the host clause, a purely syntactic approach to the that-t effect would leave unexplained why the amelioration of the effect could ever be triggered by the parenthetical expression.
Our analysis also predicts that the that-t effect should be lessened if the offending Φ-phrase is replaced by material inserted between C and the subject trace. Kandybowicz (2006, 2009) observes that in Nupe, a Niger-Congo language of West Africa, the relevant violation can be circumvented by resumption, as illustrated in (17a–b).
*Zěi Gana gàn [gànán ti du nakàn] o?
who Gana say C cook meat FOC
‘*Who did Gana say that cooked the meat?’
Zěi Gana gàn [gànán u:i du nakàn] o?
who Gana say C 3SG cook meat FOC
‘Who did Gana say that cooked the meat?’
[DP Which author]i is everyone saying that the publisher predicts that hei would be adored? (he intended to be a resumptive pronoun)
[DP Which author]i is everyone saying that the publisher predicts that [DP the guy]i would be adored? (the guy intended to be an epithet DP)
4.2 Focus Restructuring and the That-t Effect
Drury (1999) and Kandybowicz (2006) observe that, as illustrated in (19a), a that-t violation is mitigated by contrastive focus stress on the subordinate verb. Note that, as shown in (19b–c), stress on any other element does not cause any improvement.
(?)Whoi do you think that ti WROTE Barriers? (as opposed to, say, edited it)
*Whoi do you THINK that ti wrote Barriers? (as opposed to, say, know)
*Whoi do you think that ti wrote Barriers YESTERDAY? (as opposed to, say, a year ago)
Under our system, this effect follows from an independently motivated prosodic adjustment rule (Frascarelli 1997, 2002, Kenesei and Vogel 1995). Kenesei and Vogel (1995) argue that English has the Leftward Focus Restructuring Rule (LFR) in (20), which applies at the Φ-phrase level.
Leftward Focus Restructuring Rule: English
If some word in a sentence bears focus, place a Φ-phrase boundary at its right edge, and join the word to the Φ-phrase on its left.
This rule receives empirical support from the curious effect on the Rhythm Rule triggered by contrastive focus. Consider (21a–b).
It’s hard to outcláss Délaware’s football team.
It’s hard to óutclass DÉLAWARE’S football team.
The Rhythm Rule is responsible for familiar stress shifts that avoid the clash of two adjacent primary word stresses. The application of this rule has been standardly assumed to be governed by the Φ-phrase boundary (Nespor and Vogel 1986, Selkirk 1978). The rule does not apply in (21a) because the verb and its object are contained in two different Φ-phrases (recall (4c)), as shown in (22a). However, when focus is placed on the first word of the direct object, the Rhythm Rule may apply, as shown in (21b). This pattern is correctly captured by the LFR in (20), which states that the left boundary of a Φ-phrase that contains a focused word is removed so that the word is restructured into another Φ-phrase to its left. The result of applying the LFR to (21b) is shown in (22b).
(It’s hard)Φ (to outcláss)Φ (Délaware’s)Φ (football team)Φ.
(It’s hard)Φ (to óutclass DÉLAWARE’S)Φ (football team)Φ.
(adopted from Kenesei and Vogel 1995:31, with a slight modification)
(?)Whoi do you think (that ti WROTE)Φ (Barriers)Φ?
*Whoi do you THINK (that ti)Φ (wrote)Φ (Barriers)Φ?
*Whoi do you think (that ti)Φ (wrote)Φ (Barriers YESTERDAY)Φ?
In (23a), that forms a Φ-phrase with the subordinate verb thanks to the LFR, thereby evading the violation of the condition in (1). Placing focus on the matrix verb or the adverb does not change the phrasing and hence does not avoid this violation. Therefore, (23b–c) are both ungrammatical.
Our analysis coupled with the LFR leads us to predict further that the that-t effect should also be mitigated in any other syntactic environments where focus necessarily creates a derived prosodic structure for that to evade the violation of (1). This prediction is borne out by the absence of such effects in focalization and RNR. First, the that-t effect is known to be ameliorated by embedded focalization, as shown in (24).
Robin met the man whoi Leslie said (that to KIM ti)Φ (had given)Φ (the money)Φ.
This observation receives the same account as the one we offered for (19a): the focused PP is prosodically integrated into the same Φ-phrase that contains that. Second, the elements that undergo RNR are obligatorily parsed as separate I-phrases: I-phrase boundaries occur between the first and second conjuncts and between the second conjunct and the target of RNR, as the intonational phrasing in (25) illustrates.
(25) (John could have planned)I (and Mary could have hosted)I (a huge party)I.
As we showed in section 4.1, the expression in the first conjunct, which immediately precedes the target of RNR, and its correspondent in the second conjunct must bear contrastive focus. Given this observation, we predict that the that-t effect should once again be suppressed when the C occurs at the right edge of the second I-phrase as the result of RNR. De Chene (1995, 2000, 2001) observes that the I-phrase boundary created by RNR indeed ameliorates the that-t violation, as example (26) illustrates. See section 4.3 for further discussion of auxiliary reduction and its effect on phonological phrasing.
That’s the guyi (( Jim’s)Φ (been wondering if )Φ)I (and (Tom’s)Φ (been saying that)Φ)I ((really)Φ (likes)Φ (Sue)Φ)I.
(adopted from De Chene 2000:4, with prosodic phrasing added)
Here, the two Cs stand in a relation of contrastive focus. Given the LFR, this means that the second C is prosodically restructured into the Φ-phrase that contains the verb saying.
4.3 Auxiliary Reduction and the That-t Effect
(?)Who do you suppose that’ll leave early?
This pattern also follows from our system given the independently documented prosodic characteristics of reduced auxiliaries, namely, that they are morphosyntactically proclitic, but phonologically enclitic. On the one hand, an auxiliary cannot undergo reduction when it is immediately followed by a gap created by transformations or deletion (Bresnan 1971, King 1970, Lakoff 1970). (28a–b) illustrate this blocking effect.
Bresnan (1971) argues that this blocking effect is accounted for if contracted auxiliaries are proclitic. On the other hand, Lakoff (1972) and Wood (1979) argue, on the basis of voicing assimilation, that phonologically reduced auxiliaries show enclitic behavior. The examples in (29) show that the reduced auxiliary -s assimilates in voicing to the preceding segment, not the following one, just as the phonologically equivalent suffix -s does in possessive and plural contexts.
We analyze (27) as follows. The reduced auxiliary first forms a Φ-phrase with the verb it precedes because of its morphosyntactically proclitic nature, as shown in (30a). Subsequently, the auxiliary’s enclitic nature requires that the auxiliary-verb sequence attach to the preceding C, as shown in (30b). Consequently, the C ends up in the same i-phrase that contains the auxiliary-verb sequence, yielding the amelioration pattern, as desired.
4.4 Section Summary
It is clear from the above discussion that all the apparently disparate “exceptions” to the that-t effect can receive a unified explanation under our system as the result of independently motivated prosodic readjustments applied to the phase-based default prosodic phrasing template—readjustments that happen to evade the violation of the interface condition in (1). Adverbs, parentheticals, and resumptive pronouns work to prosodically incorporate the C into a newly created I-phrase to its right. Contrastive focus and RNR create a prosodic configuration where the C can form a Φ-phrase with other lexical material as a result of the LFR. Finally, auxiliary reduction forces contracted auxiliaries to integrate prosodically into the same Φ-phrase with the C, because of their enclitic character. In other words, the various environments discussed here are not “exceptions” to the that-t effect, as they have been deemed in the exclusively syntactic approach that has dominated generative grammar over the last 30 years or so. The fact that our interface-oriented approach provides a simple, unified explanation for such diverse environments that alleviate the that-t effect speaks strongly in favor of this approach.
PF Antiadjacency Filter on Cs and Traces
*〈C0, t〉 if and only if
a. C0 and t are adjacent within a prosodic phrase and
b. C0 is aligned with a prosodic phrase boundary.
The condition in (31a) accounts for the facts discussed in section 4. First, (6a) is ruled out because that and the trace are adjacent within a single prosodic phrase. Second, (11) and (26) are fine because the two items are no longer within the same prosodic phrase because of an intervening adverbial and an application of RNR, respectively. Third, the that-t violation is suppressed in (19a) since focus on the subordinate verb disrupts the C-t adjacency within the same prosodic phrase. Finally, (27) is fine because auxiliary reduction makes the wh-trace internal to C so that that plus the trace do not count as adjacent.
[the butler that___ murdered the maid]I
Our analysis also correctly predicts this result. Kayne (1994) proposes that the relative head originates from within the relative clause TP and moves into Spec,CP. Under this head-raising analysis, the relevant part of the derivation for (32) would be as shown in (33).
Within Dobashi’s (2003) system of the syntax-prosody mapping, this derivation yields the three -phrases shown in (33): ((the butler that)Φ (murdered)Φ (the maid)Φ)I. This particular phrasing is independently supported by Selkirk’s (2005) observation that a substantial pause is not necessary between the head and the relative pronoun in restrictive relative clauses; see (34a). This is in direct contrast with nonrestrictive relative clauses, which require a comma intonation; see (34b).
The Romans who arrived before one hundred AD found a land of wooded hills.
The Romans, who arrived before one hundred AD, found a land of wooded hills.
Below, we present two arguments showing that our proposed analysis constitutes a further refinement of Kandybowicz’s filter-based PF analysis. Theoretically, as Kandybowicz (2006:224) himself notes, the filter in (31) is “a descriptive generalization” that one ultimately wants to derive from independent principles of the syntax-prosody mapping. As developed thus far, our system allows us to achieve precisely that. Let us assume Truckenbrodt’s (1999) Lexical Category Condition, defined in (35) (see also Selkirk’s (1984, 1986, 1996) Principle of the Categorial Invisibility of Function Words), which states that the syntax-prosody mapping can refer to lexical categories, but not to functional categories, including the C that. See also Borsley and Tallerman 1996, Nespor and Scorretti 1985, and Nespor and Vogel 1986 for further evidence in favor of the view that empty categories, including traces/copies, have no effect on the application of prosodic rules.
Lexical Category Condition
Constraints relating syntactic and prosodic categories apply to lexical syntactic elements and their projections, but not to functional elements and their projections, or to empty syntactic elements and their projections.
Our condition in (1) directly follows from an independent condition imposed on the syntaxprosody mapping. Our phase-based prosodic mapping yields the default Φ-phrase (that t)Φ. Given the Lexical Category Condition, this phrase contains no phonetic content that is visible to prosodic phrase formation. We suggest, following Kandybowicz’s (2015) recent proposal, that the relevant -phrase violates Prosodic Vacuity. Kandybowicz presents a wide variety of observations on verbal resumption in Asante Twi to show that this resumption is obligatory whenever AspP, the complement domain of the phase-defining head v, would otherwise be phonetically empty at PF. He suggests that Prosodic Vacuity, in turn, is a corollary of Selkirk’s (2011) Match Theory of phonosyntactic alignment. This theory maintains that the left and right edges of a syntactic constituent must correspond to the left and right edges of its corresponding prosodic constituent in the phonological representation. Now, if a syntactic constituent cannot be mapped to a prosodic domain because it lacks any phonetic content, then the Match requirement is necessarily violated at the syntax-prosody interface (see also An 2007a,b for a related proposal that highlights the importance of phonetic emptiness in the syntax-prosody alignment). Adapting the Prosodic Vacuity hypothesis to the present case, we can now maintain that the Φ-phrase (that t)Φ is an ill formed phonological representation because it is phonetically vacuous. In this way, our system provides a deeper explanation for the particular form taken by Kandybowicz’s (2006) PF filter (31), deriving from the interaction of independently motivated general principles of the syntaxprosody mapping.4
There is also an empirical problem with Kandybowicz’s analysis, based on the whether-t construction illustrated in (36a). Sobin (1987) conducted an experiment with 42 undergraduate students, all native speakers of Standard English. The results of his experiment showed that whether-t constructions structurally identical to that in (36a) “had an average rejection rate of 97.6%” ( p. 45).
One might think that (36a) is degraded independently because of the Wh-Island Constraint, but there are three reasons to suspect that this cannot be the whole story. First, Sobin’s experimental result (1987:58) shows that the rejection rate is critically different between (36a) and (36b); whereas (36a) had only a 4.8% passive acceptance rate with 95.2% rejection, (36b) had a 35.7% acceptance rate (4.8% for active acceptance and 30.9% for passive acceptance) with 64.3% rejection. Second, Bley-Vroman, Felix, and Ioup (1988) observe that many native speakers accept whether-islands even though other wh-phrases uniformly disallow similar extraction patterns (see also Chomsky 1981). Third, there are languages such as Icelandic (Maling 1978:84, Sobin 1987:39) that do not prohibit extraction from wh-islands in some contexts but nonetheless exhibit the whether-t violation. These considerations thus indicate that (36a) involves the violation of some other constraint in addition to the Wh-Island Constraint.
Keeping this observation in mind, let us now consider the prosodic representation of (36a). Ha (2010:125) points out that as a function of its inherent lexical properties, whether creates a prosodic break left-aligned to it and tends to be separated from the prosodic phrase to its right, so that the C (whether) itself constitutes an I-phrase, as shown in (37) for (36a).
(37) *Who did you ask (whether)I (kissed Harriet)I?
In (37), the C and the trace are not adjacent within the same prosodic phrase (see (36a)); recall that the trace is not contained within the first I-phrase, for Kandybowicz (2006:223) assumes that a prosodic phrase boundary can only occur between two prosodic words (see also Nespor and Vogel 1986, Schütze 1994). Accordingly, his analysis would predict that there should be no appreciable contrast in grammaticality between (36a) and (36b) because both examples would violate just the Wh-Island Constraint, a prediction not borne out by Sobin’s (1987) experimental result reported above. Our current analysis, on the other hand, correctly predicts this pattern because the prosodic representation in (37) straightforwardly violates our condition in (1). Note in this connection that like that-t violations, whether-t violations are ameliorated by resumptive subject pronouns as well, as illustrated in (38a–b) (cf. (18a–b)), a pattern that further supports our prosodic analysis.
In this article, we have presented an interface-oriented analysis of the that-t effect, which we take to be a further refinement of Kandybowicz’s (2006, 2009) filter-based analysis. Adopting the phase-based model of prosodic mapping (Dobashi 2003, Kratzer and Selkirk 2007) coupled with independently motivated prosodic adjustment rules such as focus restructuring and prosodic incorporation (Kenesei and Vogel 1995, Nespor and Vogel 1986, Sobin 2000, 2002), we have demonstrated that the core facts and exceptions regarding the that-t effect can be straightforwardly derived if function words cannot form a prosodic phrase on their own. We have further shown that this interface condition, in turn, is derived from the interaction of two independently established constraints governing the syntax-prosody mapping: Truckenbrodt’s (1999) Lexical Category Condition and Kandybowicz’s (2015) Prosodic Vacuity. We hope that our work serves as a case study proving the fruitfulness of an interface approach to what has long been deemed the exclusive territory of generative syntactic research.
Earlier versions of this article were presented at GLOW in Asia IX and at the Keio Linguistic Colloquium. We thank Takamichi Aki, Zhiming Bao, Mike Barrie, Ben Bruening, Qizhong Chang, Peter Culicover, Cati Fortin, Heidi Harley, Scott Jackson, István Kenesei, Shin-Ichi Kitada, Hisa Kitahara, David Medeiros, Masaru Nakamura, Koichi Otaki, Jeff Punske, Zechy Wong, Jianrong Yu, Dwi Hesti Yuliani, and the audience members at the aforementioned meetings for comments and/or grammaticality judgments. Special thanks to the LI reviewers for important suggestions and criticisms on an earlier draft. The research reported here was supported by the faculty start-up grant from the Faculty of Arts and Social Sciences of the National University of Singapore (Sato) and by JSPS KAKENHI grant number 25370545.
1U = utterance, I = intonational phrase, Φ_ = phonological phrase, C = clitic group, and ω = phonological word.
2 As a reviewer points out, and as (ia–b) illustrate, subject resumption/copy Spell-Out does not always avoid a that-t violation. This means that resumption, as well as other prosodic/semantic circumstances that we discuss below, does not have the teleological purpose of avoiding the violation.
*Whoi do you think that hei wrote the book?
*Whoi do you think that whoi wrote the book?
We can only suggest that a certain strategy can avoid the relevant violation only when it is independently available. For example, it is well-known that resumptive pronouns may occur when the distance between the filler and the gap is sufficiently complex or crosses island boundaries (Chomsky 1982, Sells 1984). The example in (ia) is independently excluded because the distance between the filler and the gap in this example does not meet these structural conditions for resumptive pronouns. A similar analysis might hold for the copy Spell-Out case shown in (ib), though we leave this issue open for reasons of space.
3Bošković (2011) presents another PF-based analysis of the that-t effect. Assuming that the effect involves a locality violation, he proposes that this violation is marked as * on the C when a subject wh-phrase moves to the embedded Spec,CP. (i) is then ruled out because the * survives into the PF representation.
(i) Who do you think [CP
whothat* whowrote the book]?
This analysis faces numerous problems; for reasons of space, we mention only two here. First, as Stepanov (2012:684) points out, the star-marking involved in this analysis not only violates the Inclusiveness Condition (Chomsky 1995), which prohibits introducing into the syntactic derivation any features that do not appear in the numeration, but also raises the question of what decides the exact position of the marking. Second, and more important, this analysis crucially assumes, in line with the vast majority of the generative work on the topic thus far (see section 2), that the that-t effect involves some violation internal to the syntactic derivation. Accordingly, the analysis cannot account for many other prosodic circumstances discussed in this article that, as we showed, ameliorate the relevant violation.
4 We thank an LI reviewer for bringing Kandybowicz’s work on prosodic vacuity to our attention.