The PF-deletion approach

Under the PF-deletion approach, non-movement is argued for solving the puzzles of RNR which would arise under the traditional ATB-movement approach. It assumes that the two copies of the identical RNR target are base-generated within each conjunct. Under the syntactic and semantic licensing condition, the one in the first conjunct is phonologically elided (Wexler and Culicover 1980, Hartmann 2000, Bošković 2004, Féry and Hartmann 2005, An 2007, Ha 2008a, b, c).

(64) a. John loves but Bill hates syntax.

b. John loves syntax but Bill hates syntax.

For example, (64a) is analyzed as (64b): the RNR target syntax is elided in the first conjunct at the level of PF while overtly appearing at the rightmost of the second conjunct. The most crucial distinction between the ATB-movement approach and the PF-deletion approach is that the latter does not dislocate the RNR target from their base-generated position (Bošković

31

2004). That is, the RNR target is in the conjunct internal position, not in the conjunct external position.

We review two works under the PF-deletion approach, namely Hartmann (2000 and her subsequent work with Féry 2005) and Ha (2008a, b, c), and focus on their licensing

conditions for RNR. Basically, Hartmann proposes that (i) syntactically, the conjuncts must be in an identical structure, (ii) the pre-RNR elements in the first and second conjuncts must be contrastively focused, creating a set of alternatives that are identical for both conjuncts semantically,⁹ and (iii) phonologically, the deletion takes place right after the contrastively focused pre-RNR elements in the first conjunct. See (65).

(65) Hans [SCHȀLT]_F den Kȕrbus und Maria [KOCHT]_F den Kȕrbis. (Hartmann 2000) Hans peels the squash and Maria cooks the squash

‗Hans peels and Maria cooks the squash.‘

Exemplified in (65), both conjuncts involve a transitive structure, i.e. being syntactically parallel. The pre-RNR elements SCHȀLT and KOCHT show a contrastive focus. At LF, the two conjuncts mutually entail each other since they create the same set of alternatives. Finally,

den Kȕrbus ‗the squash‘in the first conjunct is elided to satisfy the prosodic alignment

condition. Nevertheless, the problems of Hartmann‘s proposal are pointed out in Ha (2008b) as follows. First, the identical syntactic structure wrongly rules out vehicle change effects.

Secondly, it is unclear how to prohibit the RNR target in the second conjunct to be deleted under the prosodic alignment condition. As argued, the nuclear accent received on the RNR target has to be aligned at the rightmost in each Intonational Phrase, meaning that both RNR targets would be deleted. However, only the deletion of the RNR target in the first conjunct

9 The contrastive focus would be absent if these two parts are identical, too similar or not of a same semantic kind.

32

forms the structure of RNR.

Alternatively, Ha (2008a, b, c) revises Hartmann and argues for syntactic, phonological, and semantic licensing conditions in RNR. Adopting Merchant‘s (2001) ellipsis feature of sluicing (ES), ERNR feature has to (i) enter the derivation with contrastively focused pre-RNR elements in the first conjunct, (ii) instruct its sister to be unpronounced, and (iii) meet the requirement that the RNR target (in the first conjunct) and its antecedent (in the second conjunct) are mutually entailed at LF. Also note for the E_RNR feature to be licensed, it has to check its feature with the closest head of C.

(66) [CP F [&P JOHN MADE[ERNR] <the spaghetti>, and Bill ATEthe spaghetti.]]

As illustrated in (66), JOHN MADE are the contrastively focused elements bearing ERNR. Outside the coordination, the feature F triggers agreement relation with its closest E_RNR feature. Next, the ERNR feature tells the PF not to spell out its complement the spaghetti;

semantically, the RNR (clause) John made the spaghetti turns to variables at LF and is existentially closed by F-closure, yielding

Ǝ x Ǝ R [x R-ed the spaghetti]

; the same applies to

the antecedent (clause) Bill ate the spaghetti, resulting in

Ǝ x Ǝ R [x R-ed the spaghetti].

¹⁰ Taken together, the F-closure of the antecedent entails the RNR, and that of the RNR entails the antecedent, meaning that the two conjuncts mutually entail each other, satisfying the e-GIVENESS condition.

2.3.1 Advantages

This approach assumes that RNR involves a PF-process and no overt movement takes place; therefore, it directly solves some problems of the traditional ATB-movement approach:

10 Adopting Merchant (2001), the LF condition of mutual entailment is called e-GIVEN.

33

those related to movement constraints (i.e. non-constituency, CSC, Right Roof Constraint violation, island insensitivity and P-stranding). In particular, Ha (2008b) accounts for sloppy reading, asymmetric vehicle change, and asymmetric NPI licensing. Also note that the RER is captured independently from the licensing requirements of ERNR. That is, even if an ERNR

meets the syntactic, phonological, and semantic requirements, it would be ill-formed if the RER is not obeyed. Regarding the apparent non-constituency, Ha argues that it is the outcome of the deleted elements as a whole, and RNR is in fact targets the constituents. More

specifically, Ha allows multiple ERNR features and each of them instructs the PF to

unpronounce its sisters, i.e. the constituents. Even though the elided RNR target as a whole does not seem to be a constituent, each deletion at PF is indeed a constituent. He uses a German example to illustrate this notion because there seems to be no direct account for such non-constituency.

(67) a. [CP Ich habe einen Mann, [CP der DREI <Katzen besitzt>, <gekannt>], und I have a man who three cats owns knows and [CP Sie hat eine Frau, [CP die VIER Katzen besitzt], gekannt.]

she has a woman who four cats owns knows

‗I have known a man who owns three cats, and she has known a woman who owns

four cats.‘ (German, Wilder 1997)

34

b.

(Ha 2008b)

In (67a), the RNR target Katzen besitzt gekannt is a non-constituent sequence of NP katzen

„cats‟, embedded verb besitzt ‗owns‘ and the main verb gekannt ‗knows‘. Each of them is

contrastively focused and bears the E_RNR feature. The D E_RNR feature projects and deletes its sister, vP gekannt ‗knows‘; similarly, the NP ERNR feature projects and licenses the deletion of its sister Katzen besitzt „cats owns‟. Note that the RNR target contains focused three, which cannot be elided.The AP[ERNR] is an adjunct and cannot project further, so it only deletes its sister NP Katzen „cats‟. Finally, the sequence of Katzen besitzt gekannt is elided. Taken together, Ha claims that since each ERNR feature only deletes its sister, the elided elements are indeed a consitituent.

It is worth mentioning here that the most crucial difference between Hartmann (2000 and her subsequent work) and Ha (2008a, b, c) is that regarding the RNR target, the former assumes a syntactic identity whereas the latter assumes a semantic identity which can nicely solve the problematic sloppy reading, asymmetric vehicle change, and asymmetric NPI licensing. Sloppy reading can be captured by assuming the RNRed pronoun is a bound variable. Recall that the RNRed pronoun is assumed to be a bound variable and the semantic licensing condition allows different reference for pronominals.

35

(68) JOHNi LIKES_[ERNR]<hisi father>, but BILLj HATES hisj father. (Ha 2008b)

In (68), at LF, the antecedent turns to bound variables and is existentially closed as Bill λx [ x

hates x‟s father] and the RNR clause as John λy [ y likes y‟s father]. The outcome of

F-closure of the antecedent and the RNR clause is still the same,

Ǝ x Ǝ R [x R x‟s father],

satisfying the semantic condition. As a result, sloppy reading is captured.

Asymmetric vehicle change effects can be captured as well. Ha assumes the semantic identity is relatively flexible than the syntactic identity; therefore, the

co-reference between the R-expression and the pronoun is available as long as they have the same interpretation.

(69) JOHNi hopes that Susanj WON‘T <fire himi

(at the end of this year)>, but THE

SECRETARY knows that shej WILL fire Johni at the end of this year.

In (69), the semantic licensing condition can be met when the deleted RNR target and the antecedent mutually entail each, meaning that him refers to John. Another type of vehicle change, the asymmetric NPI licensing, can be accounted for in a similar way. Ha assumes that an NPI and a positive polarity item (i.e. PPI) share a binary feature [± polarity indefinite]: the former has the feature [+polarity] and the latter [-polarity]. Since vehicle change is insensitive to the feature value, an NPI can be treated as a PPI under reconstruction. Consider (70).

(70) a. John read, but he hasn‘t understood any of my books. (Ha 2008b) b. *John read any of my books, but he hasn‘t understood any of my books.

c. John read some of my books, but he hasn‘t understood any of my books.

36

Example (70a) has the reconstructed representation in (70b), whose ungrammaticality can be avoided by applying vehicle change. In (70c), any is realized as some in the first conjunct (i.e.

the ellipsis site), giving rise to grammaticality.

Additionally, Ha claims that the licensing of relational modifiers and scope ambiguity can be accounted for. After the derivation of RNR (i.e. only one target on the surface), the single instance of the relational modifier in the second conjunct can undergo covert movement out of the coordinate structure, and thus scopes over the whole coordination at LF.¹¹ Note that this single entity of the RNR target in the second conjunct induces an internal reading, different from the cases where the two entities of the RNR target yield an external reading.

(71) a. Peter sings and Mary whistles a similar tune. (Ha 2008b) b. [A similar tune] [Peter sings and Mary whistles] (Ha 2008b)

Seen in (71), syntactically, a similar tune can covertly move out of the coordination, where it has an inverse scope, i.e. wide scope. Therefore, the internal reading can be obtained: the tune Peter sings and the tune Mary whistles is similar.

Just as the cases of relational modifiers, quantifiers can exhibit inverse scope by undergoing covert movement at LF after the derivation of RNR. Specifically, after the deletion of the RNR target in the first conjunct, the RNRed quantifier phrase in the second conjunct is available for covert movement (i.e. Quantifier Raising) out of coordination at LF, inducing an inverse scope.

(72) Every philosopher read and every linguist reviewed some paper. (Ha 2008b)

11 Ha notes that if the relational modifiers occur overtly in the first conjunct, covert movement of the target in the second conjunct is disallowed. This is because the CSC would be violated.

37

(every >some; some > every)

(73) [TP Some paper [&P [TP every philosopher read <some paper>] and [TP t‘ every linguist

reviewed t] (Ha 2008b)

As seen in (72), some paper in the first conjunct is elided after valued with the head of C, deriving the only one instance of some paper in the second conjunct. Next, the target is assumed to undergo Quantifier Raising at LF, making the inverse scope (i.e. some > every) possible as in (73).

2.3.2 Disadvantages

Summative agreement effects are problematic under the PF-deletion approach. With a copy of the RNR target contained in each conjunct, the approach would wrongly predict (74a) to be ill-formed just as its undeleted counterpart in (74b), contra fact. It is unclear how the plural verb have can be licensed even though the embedded subject in each conjunct is singular.

(74) a. [Mary is proud John ______ ] and [Sue is glad that Bill _____ ] have traveled

t

_John/Bill to Cameroon. (Grosz 2014)

b.* [Mary is proud that John have traveled tJohn to Cameroon ] and [Sue is glad that

Bill have traveled t

_Bill to Cameroon.] (Grosz 2014)

In addition, Kluck and Zwart (2009) points out that Ha (2008a, b, c) fails regarding the cases of more than two conjuncts. This is because Ha claims that only the closest and

uninterpretable ERNR need to check its feature with the head of C during syntactic derivation.

In other words, if there are three conjuncts, only the target in the first conjunct can have its value checked and then be deleted, leaving the target in the second and the third conjunct to

38

occur on the surface structure.

2.3.3 Summary

The PF-deletion approach directly accounts for the puzzles which would arise under the traditional ATB-movement approach, namely violations of movement conditions, sloppy reading, asymmetric vehicle change, and asymmetric NPI licensing. Moreover, the licensing of relational modifiers and the scope ambiguity are captured by assuming the covert

movement of the single RNR target at LF. However, it is not clear how the deletion approach captures the summative agreement effects.

Table 3 Properties explained by the PF-deletion approach

1. Right Edge Restriction (RER) 

2. Non-constituency 

3. Coordinate Structure Constraint (CSC)  4. Right Roof Constraint (RRC) violation 

5. Island insensitivity 

6. P-stranding 

7. Strict/sloppy readings 

8. Asymmetric vehicle change 

9. Asymmetric licensing of NPI 

10. Licensing of relational modifiers 

11. Scope ambiguity 

12. Summative agreement ?

在文檔中 漢語右部節點提升結構 (頁 37-45)