Seeking an optimal design of Search and Merge: its consequences and challenges

4.1.1 The freezing effect

The MS-free Merge Hypothesis can provide a principled explanation for the freezing effect that movement is not possible out of a moved element (see, among many others, Bosković 2018; Wexler and Culicover 1980 and references cited therein). One of the most typical examples of the freezing effect is the subject island effect as shown in (13) (Chomsky 1973):

In (13), who is extracted out of the subject DP moved from SPEC-v* to SPEC-T. We can account for the ungrammaticality of (13) as a Binarity violation on Search Σ for Merge. Consider the WS before who undergoes IM to SPEC-C:

In (14), the DP₁ containing who ₁ occupies SPEC-v*, and the DP₂ containing who ₂ SPEC-T. These are required from θ-Theory (Chomsky 2021) and Labeling Theory (Chomsky 2013, 2015), respectively. The former dictates that language must provide argument structure at the Conceptual-Intentional (CI) interface, and the latter that language must provide labeled SOs at the CI interface. DP₁ is assigned a θ-role by the R(please)-v* amalgam, and the SPEC-T construction is labeled as <φ, φ> through φ-agreement between DP₂ and T. Note that since the SPEC-T construction is of symmetric {XP, YP} type, its label cannot be determined by MS. Its label is rather determined by prominent feature sharing via agreement. To generate (13), Search Σ for IM(C, who) must apply to (14). There are, however, two copies of who, i.e., who₂ and who₁. If Search Σ were to select C and who, the elements provided to IM would be ternary, i.e., C, who₂, and who₁. This would violate Binarity on Search Σ. Since (13) cannot be generated by Merge, it is ungrammatical.^[10]

Notice in this analysis that if Search Σ for Merge selected C and one of the two copies of who, i.e., either who₂ or who₁, it would satisfy Binarity on Search Σ. This search process, however, would not provide an appropriate input to Merge. This is because Merge is defined as an operation that combines two particular SOs (see Section 2.1). Phase-level memory dictates that who₂ and who₁, which are within the same Transfer domain (i.e., TP), are the occurrences of the same SO who. In other words, the SO who counts as a discontinuous SO in WS (14). Hence, there is no way of selecting C and part of a particular SO, i.e., either who₂ or who₁, so as to provide an appropriate input to Merge of C and who.^[11] It should also be recalled that to satisfy Binarity, Search Σ for Merge here can in principle select an element other than who along with C, but the resulting output would be ruled out by independent principles (e.g., by a labeling failure of <Q, Q>).

Our analysis predicts that the subject island effect is obviated in an environment where a subject DP containing a wh-phrase stays in-situ. This prediction is borne out by the following example (see Lasnik and Park 2003 for more examples):

In (15), there occupies SPEC-T, and accordingly, the DP containing who stays in SPEC-v. The grammaticality of (15) can be accounted for by considering the following WS:

To generate (15), Search Σ for IM(C, who) applies to WS (16). There is only one copy of who in (16). When Search Σ is to select who along with C, the elements provided to IM are binary, i.e., C and who. This satisfies Binarity. Hence, (15), which can be generated by Merge, is grammatical.

To see how our theory deals with extraction from object position, consider (17) as an example:

In (17), who is extracted out of the object DP, and the sentence is acceptable. Consider the WS before who undergoes IM to SPEC-C:

(18)

To generate (17), Search Σ for IM(C, who) must apply to WS (18). Note that who₁ in DP within R-COMP is inaccessible by the PIC/Transfer. When Search Σ is to select who along with C, the elements provided to IM are binary, i.e., C and who₂. This satisfies Binarity; (17) is grammatical. The fact that it is possible to extract a wh-phrase from an ECM subject (Chomsky 2008: 153) can be treated on a par with the extraction from object position, given the raising-to-object analysis of ECM, which claims that the ECM subject is raised to SPEC-R of the matrix clause.

Another example of the freezing effect is concerned with verb-particle constructions in English:

Johnson (1991) proposes that V(erb)-Part(icle) is a single item in the underlying structure, arguing that the basic word order of the verb particle construction is the V-Part-DP order, and the V-DP-Part order is derived, as shown in (21):

Johnson (1991: 590–591) presents evidence for V-Part as a single verb. The first piece of evidence comes from morphological processes. As shown in (22), -ing to form a noun and -ed to form an adjective, which only apply to verbs, can be attached to V-Part as a whole:

These facts indicate that the particle up should be treated as part of the single verb looked up.

The second piece of evidence is obtained from Gapping (Johnson 1991: 591):

Gapping treats the V-Part looked up as a single verb (23a), but it cannot apply to the verb looked alone, with the particle up being stranded (23b). Since only verbs, either a single or in series, can be gapped, (23b) shows that the particle up should be treated as part of the single verb looked up.

Johnson (1991: 607) reports that when DP precedes Part, extraction out of the DP becomes degraded (see also Lasnik 2001); the freezing effect is observed in the V-DP-Part order:

Within the framework assumed here, if V-Part forms a single verb and if R universally raises to v* for root-categorization (Chomsky 2015: 15; see also Section 3.3), it then follows that [_R V-Part] raises to v*, whether it is in the V-Part-DP order or the V-DP-Part order, as represented in (25):

In (25), [_R V-Part] is with v* for root-categorization. Given feature inheritance from a phase head to the head of its complement, the copy of R (t_i) inherits ϕ-features from v*. The object DP in SPEC-R agrees with the copy of R (t_i) for {ϕ, ϕ} labeling. This derives the V-Part-DP order. To derive the V-DP-Part order from this structure, V has to raise further, leaving Part in v*, and DP also needs to move to around SPEC-v*, as shown in (26):

Given that V-Part is a single item in the underlying structure, wherever the exact landing sites of the moved V and the moved DP are, derivation (26) should be involved in the V-DP-Part order.^[12]

Importantly, if (24b) has the derivation as in (26), then the degraded status can be accounted for as a violation of Binarity on Search Σ for Merge. Consider the following relevant WS:

(27)

In (27), there are three copies of DP, i.e., DP₃ in SPEC-v*, DP₂ in SPEC-R, and DP₁ in COMP-R. Note that DP₁ is inaccessible by the PIC/Transfer, but DP₃ and DP₂ are accessible. Thus, if Search Σ were to select C and who, the elements provided to IM would be ternary, i.e., C, what₃, and what₂. This would violate Binarity; (24b) is ungrammatical.

4.1.2 The that-trace effect

The MS-free Merge Hypothesis can also range over the that-trace effect exemplified by (28) (Bošković 2016; Chomsky 1986; Ishii 2004; Kayne 1984; Lasnik and Saito 1992; Perlmutter 1971):

In (28), who is moved to the matrix SPEC-C from the embedded SPEC-T over the embedded SPEC-C with the overt COMP-that. The unacceptability of (28) can be accounted for as a Binarity violation. Consider the WS before who undergoes IM to SPEC-C in the embedded clause:

To generate (28), Search Σ for IM(C, who) must apply to WS (29). There are, however, two copies of who, i.e., who₂ and who₁. If Search Σ were to select C and who, the elements provided to IM would be ternary, i.e., C, who₂, and who₁. This would violate Binarity; (28) is ungrammatical.

It is well-known that the that-trace effect disappears when C is deleted, as shown in (30):

We assume with Chomsky (2015) that finite clauses without overt COMP-that (that-less clauses) are TPs, T is a phase head through the transfer of phasehood from C to T, and T-COMP (v*P) is inaccessible by the PIC/Transfer. Evidence for the view that that-less clauses are TPs comes from the fact that topicalization is impossible in that-less clauses (see Bošković 1997; Doherty 2000):

Assuming that topicalization targets the CP field, this fact suggests that that-less clauses are TPs: topicalization in that-less clauses is impossible because there is no C head (see also, among many others, Ishii 2004; Weisler 1980, for the TP approach to that-less clauses). If there is no CP in the embedded clause of (30), the matrix RP is the next relevant phase level for further wh-movement:

(32)

To generate (30), Search Σ for IM(R, who) must apply to (32) under successive-cyclic movement. Note here that who₁ in the embedded T-COMP (the shaded v*P) is inaccessible by the PIC/Transfer. As a result, when Search Σ is to select who along with R, the elements provided to IM are binary, i.e., R and who₂. This satisfies Binarity on Search Σ for Merge. (30) is grammatical.^[13]

We can obtain the same prediction as in the analysis of obviation of the subject island effect: the that-trace effect does not occur in an environment where a wh-phrase stays in-situ. This prediction is borne out by the following contrast (Rizzi and Shlonsky 2007: 126):

The grammaticality of (33b) can be accounted for in the same way as the fact that the subject island effect is obviated when there occupies SPEC-T (see (15) and (16)).

4.1.3 Why Japanese has neither the that-trace effect nor the subject island effect

Japanese exhibits neither the that-trace effect (Ishii 2004) nor the subject island effect (Ishii 1997, 2011; Kayne 1984; Lasnik and Saito 1992; Saito and Fukui 1998):

In (34) (Ishii 2011: 408), the wh-element dare-ni ‘who-Dat’ is scrambled out of the nominative phrase that occupies the subject position in the embedded clause, and the sentence is acceptable. This shows that there is no subject island effect. In (35), the subject basu-ga san-dai ‘three busses’ is extracted out of the embedded clause with the COMP-to ‘that’, and the sentence is acceptable. This indicates that there is no that-trace effect. The question is why English has both the subject island effect and the that-trace effect, whereas Japanese has neither.

If we assume with Fukui (1986) and Kuroda (1988) that the subject in Japanese stays in SPEC-v throughout a derivation, we can give a principled explanation to this fact.^[14] One of the arguments for the claim that the subject stays in SPEC-v in Japanese is presented by Kasai (2018). His argument is based upon how coordination structures like (36) are derived, where the verb tabe ‘eat’ in the first conjunct is bare and lacks a tense morpheme:

Kasai (2018: 9) argues that if the subject stays in SPEC-v, (36) is supposed to be derived from coordination of the embedded vPs, as illustrated in (37a), and on the other hand, if the subject moves to SPEC-T, (36) should be derived from coordination of the matrix TPs by applying ellipsis in the first conjunct, as illustrated in (37b) (strike-through here indicates ellipsis):

Kasai argues that the non-ellipsis analysis (37a) is more plausible than the ellipsis analysis (37b) based on the ungrammaticality of (38) below, where Taroo-ga, Hanako-ni, nattoo-o and tabe do not make a constituent:

Kasai argues that under the ellipsis analysis, it is not clear why (38) cannot be derived for the same reason as (37b) (by applying ellipsis in the first conjunct), but under the non-ellipsis analysis, the ungrammaticality of (38) is easily captured by a ban against coordination of non-constituents. He concludes that (36) should be analyzed as in (37a), and the subject stays in SPEC-v in Japanese.^[15]

A similar argument appears in Chomsky (2021: 34). (39) shows that the verb in the first conjunct appears in the present tense, and the verb in the second conjunct in the past tense:

Based on the fact that a tense feature is not necessarily shared by two conjuncts, Chomsky (2021: 33–34) argues that tense is a feature of v. He further argues that such cases as (40) are supposed to be derived from coordination of the embedded vPs not with a single T head, as shown in (40a), but with a single INFL that only has φ-features but not a tense feature, as shown in (40b):

It is plausible to analyze (39) as (40b). This is because if T has a tense feature, as in (40a), not only it is unclear why the two conjuncts do not share the same tense (although they are c-commanded by the same single T head), but it is not clear at all how the different tenses are realized by the two conjuncts. If, as Chomsky argues, tense is a feature of v and there is an INFL only with φ-features without tense, however, it naturally follows that a tense feature does not have to be shared by two conjuncts like (39). Given that tense is a feature of v, since there is no φ-feature agreement in Japanese (Fukui 1986; Kuroda 1988; Saito 1985), it is natural to claim that in Japanese there is no INFL whose SPEC is the target of subject raising, and subjects remain SPEC-v in the language.^[16]

Given that the subject in Japanese stays in SPEC-v throughout a derivation, the relevant WS of (34) and (35) before the wh-element dare-ni ‘who-Dat’ and the subject basu-ga san-dai ‘three busses’ undergo IM to SPEC-C in the embedded clause is as follows (where English glosses are used for ease of exposition):

To generate (34) and (35), Search Σ for IM(C, who/three busses) applies to WS (41). There is only one copy of who/three busses in WS (41). When Search Σ is to select who/three busses with C, the elements provided to IM are binary, i.e., C and who/three busses. (34) and (35), which are Merge-generable, are grammatical.^[17]

4.1.4 The anti-locality effect

The anti-locality effect that a subject cannot undergo “short” topicalization (Fiengo et al. 1988; Lasnik and Saito 1992) also receives a principled account under the MS-free Merge Hypothesis. As shown by the contrast between (42a) and (42b), “short” topicalization of a subject from SPEC-T to SPEC-C is impossible, but clause-internal topicalization of an object is possible (we are assuming that an element to be topicalized is merged with C; see the discussion around (31)):

Lasnik and Saito (1992: 110–111) present two pieces of evidence for the non-existence of “short” topicalization of a subject. The first evidence is concerned with the effects of topicalization on anaphor binding. Consider the following examples:

(43a) and (43b) show that if the embedded object anaphor himself does not undergo topicalization, it cannot be bound by John, but if himself undergoes topicalization, it can be bound by John. The ungrammaticality of (43c) illustrates that if the embedded subject anaphor himself does not undergo topicalization, it cannot be bound by John, in the same way as (43a). These facts lead us to predict that (43d), where the embedded subject anaphor himself undergoes topicalization, is grammatical for the same reason as (43b). This prediction, however, is not borne out. Lasnik and Saito (1992) argue that the ungrammaticality of (43d) supports the claim that a subject cannot undergo “short” topicalization.

The second evidence has to do with the following paradigm involving the subject island effect (Lasnik and Saito 1992: 111):

(44a), where the wh-phrase is extracted out of the subject DP that is moved from SPEC-v* to SPEC-T in the embedded clause, is a case of the subject island effect. On the other hand, (44b) and (44c), where the wh-phrase is extracted out of the subject/object DP that undergoes wh-movement to SPEC-C in the embedded clause, indicate that wh-extraction out of DPs in an A′-position is reasonably acceptable, irrespectively of whether the DPs are subjects or objects. Lasnik and Saito (1992: 111) observe that (44d), where the wh-phrase is extracted out of the object DP that undergoes topicalization in the embedded clause, is marginally allowed, conforming to the generalization that wh-extraction out of DPs in an A′-position is acceptable. These facts lead us to predict (44e), where a wh-phrase is extracted out of the subject DP that undergoes topicalization in the embedded clause, is as acceptable as (44b), (44c), and (44d), since the DP is in an A′-position. However, this is not the case. Lasnik and Saito (1992) argue that this indicates that a subject cannot undergo “short” topicalization.

Turning back to (42a), we can account for the anti-locality effect as a Binarity violation. Consider the following WS of (42a) before John undergoes IM to SPEC-C for topicalization:

To generate (42a), Search Σ for IM(C, John) must apply to WS (45). There are, however, two copies of John, i.e., John₂ and John₁. If Search Σ were to select John along with C, the elements provided to IM would be ternary, i.e., C, John₂, and John₁. This would violate Binarity. Since (42a) is not Merge-generable, it is ungrammatical. Topicalization of an object as in (42b), on the other hand, is possible. This fact can be accounted for in the same way as the extraction from an object and an ECM subject seen above.

4.1.5 The vacuous movement hypothesis

Our theory provides a principled explanation for the Vacuous Movement Hypothesis (VMH) that a wh-subject does not move locally to SPEC-C (Chomsky 1986; George 1980; Ishii 2004):

According to the VMH, the derivation of (46a) and (46b), where who stays in SPEC-T, is chosen over that of (46b), where who undergoes IM to SPEC-C from SPEC-T. In other words, string vacuous wh-movement of a subject is not allowed.

Chomsky (1986: 50) argues that the difference in acceptability between (47a) and (47b) below constitutes evidence for the VMH:

According to Chomsky (1986), (47a), which includes relativization from within the wh-island in which who functions as a subject, is more acceptable than (47b), which involves relativization from within the wh-island in which who functions as an object. Chomsky (1986) argues that if the subject who in (47a) does not occupy the embedded SPEC-C, then the embedded SPEC-C serves as an escape hatch for successive-cyclic movement of to whom; to whom moves out of the CP complement of wonder without violating Subjacency. Since the object who in (47b) occupies the embedded SPEC-C, on the other hand, the embedded SPEC-C does not serve as an escape hatch for successive-cyclic movement of to whom; movement of to whom out of the CP complement of wonder violates Subjacency. Hence, the contrast between (47a) and (47b) follows from the VMH.

Chomsky (2021: 35) also presents parasitic gap (PG) facts as evidence in favor of the VMH:

The structures of (48a) and (48b) are represented below:

As shown by the contrast between (48a) and (48b), the object wh-phrase licenses a PG, while the subject wh-phrase does not. Following the well-established generalization that the antecedent of a PG must be in an A′-position, and a PG can only be licensed by overt movement (see, among others, Chomsky 1982, 1986; Engdahl 1983), Chomsky argues that this contrast follows from the VMH coupled with the Form Copy analysis of PG and the ban against improper Copy Pair, the Copy analog of improper movement.^[18] In (49a), since what₄ moves to SPEC-C, an A′-position, the identical inscriptions <what₄, what₂>, which are in an IM configuration, are assigned a Copy relation by Form Copy; what ₄ and what ₂ are interpreted as in a Copy relation. This forms the Copy pair <what₄, what₂>. This Copy pair is legitimate, since both what ₄ in SPEC-C and what ₂ are in an A′-position. The dependency of the parasitic chain on the licenser’s chain is guaranteed by this Copy pair; the PG is licensed. In (49b), on the other hand, the VMH requires what₄ to remain in SPEC-T, an A-position. The identical inscriptions <what₄, what₂> in an IM configuration are assigned a Copy relation by Form Copy. This results in the improper Copy pair <what₄, what₂>, however, since what₄ in SPEC-T is in an A-position whereas what₂ is in an A′-position. The PG cannot be licensed in (49b). It should be noted that if what₄ moved further from SPEC-T to SPEC-C as represented in (50), the identical inscriptions <what₅, what₂> could be assigned a Copy relation by Form Copy, which is an optional operation. This Copy pair would license PG in (48b), contrary to fact. Hence, the contrast between (48a) and (48b) presents evidence for the VMH:

Chomsky (2021: 35) argues that this analysis also rules out (51), whose structure is (52), due to the improper Copy pair <what₄, what₂>. It should be noted that since Form Copy only applies within a phase, what₅ in the matrix SPEC-C cannot form a Copy relation with what₂ due to the intervening phase boundaries. Hence, PG is not licensed in (51):

The VMH follows from the MS-free Merge Hypothesis. Consider the relevant WS of (46a, b) below, where who still stays in SPEC-T:

In WS (53), there are two copies of who, i.e., who₂ and who₁. If who were to undergo IM to SPEC-C vacuously as in (46b), Search Σ for IM(C, who) must apply. In so doing, however, there would be three accessible elements for Search Σ, i.e., C, who₂, and who₁. This would violate Binarity.^[19] In order not to violate Binarity, Search Σ for IM(C, who) should not be applied, as shown in (46a). This is exactly what the VMH requires, which follows from the MS-free Merge Hypothesis.^[20]

4.1.6 No superfluous steps

The MS-free Merge Hypothesis also provides us with an important insight to understand the last resort nature of successive-cyclic A′-movement that avoids superfluous steps. Let us compare two possible derivations of (54), which are represented in (54a) and (54b):

In (54a), who undergoes IM from SPEC-R to SPEC-C successive-cyclically without stopping over any intermediate positions. In (54b), on the other hand, who undergoes IM from SPEC-R to the TP-adjoined position before moving to SPEC-C. Within the framework of Chomsky’s (1986) Barriers system, Chomsky (1986: 5, 32) and Lasnik and Saito (1992: 72–73) present evidence to show that a wh-phrase may not adjoin to TP, as in (55b). We present their argument against TP-adjunction of a wh-phrase within the Barriers system for an expository purpose. They argue that if TPs were possible landing sites for wh-movement, then (55) would not violate Subjacency at all, as shown in (55a):

In the Barriers system, any maximal projection is a potential barrier. A potential barrier, however, is exempted from barrierhood if it is θ-marked.^[21] In (55a), where who adjoins to the embedded TP, there are no barriers between the embedded TP-adjoined position and the matrix RP-adjoined position (i.e., VP-adjoined position in Chomsky’s 1986 system). The embedded CP is devoid of its barrierhood for Subjacency, since it is θ-marked by the matrix predicate wonder. The embedded TP is not a barrier for who in the TP-adjoined position, either. We would incorrectly predict that (55) is perfect. In (55b), where what does not adjoin to the embedded TP, the embedded CP, though θ-marked by the matrix predicate wonder, functions as a barrier for Subjacency due to the inheritance of barrierhood from the embedded TP. Since the movement of what from the embedded RP-adjoined position to the matrix RP-adjoined position crosses one barrier, i.e., the embedded CP, conforming to 1-subjacency, the derivation (55b) can correctly capture the marginality of (55).^[22]

Then, the question is, why (54b) with the TP-adjunction is prohibited, and (54a) with phase-by-phase cyclicity is chosen over (54b) with a superfluous step. If Merge is free, even such “superfluous” steps should be a freely available option. In Chomsky (1995), derivations such as (54b) with superfluous steps were excluded by the principle of Economy of Derivation, such that “shorter derivations are always chosen over longer ones” (Chomsky 1995: 139). Significantly, this economy principle follows from the MS-free Merge Hypothesis. Consider the following relevant WSs of (54a) and (54b) before who undergoes IM to SPEC-C (where who in R-COMP is inaccessible by the PIC/Transfer):

To generate the sentence (54), Search Σ for IM(C, who) must apply. If it applies to (56a), with phase-by-phase cyclicity, it satisfies Binarity, as the elements provided to IM are binary, i.e., C and who. But, if it applied to (56b), with a superfluous step, it would violate Binarity, as the elements provided to IM would be ternary, i.e., C, who₂ and who₁. Hence, the preference for a derivation with phase-by-phase cyclicity follows from our MS-free Merge Hypothesis.

4.2.1 Eliminating redundancy

In addition to the empirical consequences, our MS-free Merge Hypothesis has theoretical consequenses. Our hypothesis clears up some theoretical and conceptual uncertainties about the relation between IM and Minimal Search (MS) in the system advocated by Chomsky (2021). The claim that IM obeys MS induces a redundancy problem. Let us consider the configuration [X₂ … [PH ]], where there are one copy of X above the phase head (PH) and another copy of X below PH. In this configuration, MS requires that the higher X₂ is accessible but the lower X₁ is not accessible (here the strike-through line stands for inaccessibility). This (in)accessibility is also ensured by the PIC. The PIC dictates that items within the complement of PH cannot be accessed due to Transfer at a phase level. The PIC/Transfer makes the higher X₂ accessible but the lower X₁ inaccessible as it is contained in the complement of PH, as indicated by the gray shade. Since eliminating redundancies has been a working hypothesis in the minimalist program (Chomsky 1995: 152), such a redundancy between MS and the PIC should be eliminated. Our MS-free Merge Hypothesis, where MS is dissociated from Merge, eliminates this redundancy.^[23]

4.2.2 Our analysis of the illegitimate IM

Since Chomsky’s argument that IM is constrained by MS is based on the legitimacy of IM, we need to show that our hypothesis can explain the legitimacy of IM even without assuming MS.

Let us first reconsider the illegitimate IM (4) (repeated here as (57)). (58) is the relevant WS of (57) before IM is applied:

To generate (57), Search Σ for IM(C, who) must apply to WS (58). There are, however, two copies of who, i.e., who₂ and who₁. If Search Σ were to select C and who, the elements provided to IM would be ternary, i.e., C, who₂, and who₁. This would violate Binarity on Search Σ. Since (57) cannot be generated by Merge, it is ungrammatical. Note that who₁ in R-COMP is accessible if the complement of the passive v phase head is not blocked by the PIC/Transfer, unlike the transitive v* phase head.^[24] Thus, our theory can correctly rule out illegitimate IM without assuming MS.

4.2.3 Our analysis of the legitimate IM

Turning now to legitimate applications of IM, let us reconsider (7) (repeated here as (59)):

In WS₂ (59b), IM applies to c. Recall that in Chomsky’s analysis, the lower c is protected/blocked by the higher c because of MS, thereby being no longer accessible; there is no violation of Minimal Yield (MY).

How can the MS-free Merge Hypothesis rule in the legitimate IM without assuming MS while capturing the effect of MY? In order to derive (59b) from (59a), Search Σ for IM needs to select {a, {b, c}} and c in it in accord with Optimal Computation. This search process satisfies Binarity, so our theory rules in IM without assuming MS. It is true that Merge adds two new accessible elements from (59a) to (59b), i.e., one copy of c and {c, {a, {b, c}}}. Hence, one might say that it is a violation of MY. It is important to notice, however, that the main purpose of MY is not to restrict the application of Merge. In fact, Chomsky (2021: 19) argues that “Merge should construct the fewest possible new items that are accessible to further operations, thereby limiting Σ [emphasis ours]”. That is, the ultimate effect to be exhibited by MY is to limit Search at a later stage of a derivation, not to restrict application of Merge at the present stage of a derivation. Reinterpreting MY in this way, there is nothing wrong with IM producing (59b) with two new elements added at the present stage of a derivation. If further Search Σ for IM applies to c in (59b), it results in a violation of Binarity due to the two copies of c, thereby limiting Search at a later stage of a derivation. Hence, our theory can capture the insight of MY without positing it as an independent constraint on Merge. Note that if further Search Σ for IM is not applied to (59b), then the derivation is maintained without any problems. Hence, our theory can correctly rule in legitimate applications of IM without assuming MS.

4.2.4 Our analysis of the extensions of Merge

One significant consequence of Chomsky’s (2021) MY is that the so-called extensions of Merge, such as Late Merge (Ishii 1997; Lebeaux 1991), Parallel Merge (Citko 2005), and Sideward Movement (Nunes 1995), are ruled out.

Let us consider (60), the relevant WSs of Parallel Merge/Sideward Movement, where (60b) is yielded by applying Parallel Merge/Sideward Movement to a and c in {b, c} in (60a):

In WS₁ (60a), the number of accessible elements is four (i.e., a, b, c, {b, c}), while in WS₂ (60b), it is six (i.e., a, c, b, c, {a, c}, {b, c}). Note here that c in {b, c} remains accessible, since it is not c-commanded and thus not protected/blocked by c in {a, c}. From WS₁ (60a) to WS₂ (60b), Merge adds two new accessible elements to WS; this violates MY, according to Chomsky (2021).^[25]

Under the MS-free Merge Hypothesis, the extensions of Merge can be ruled out as a violation of Binarity on Search Σ for Merge without recourse to MY. Let us consider (60) again, now with particular attention to WS₁ (60a) before Parallel Merge/Sideward Movement applies. In order to derive WS₂ (60b) from WS₁ (60a) in one step, Search Σ for Merge needs to select three items, i.e., a, {b, c}, and c in {b, c}. This violates Binarity on Search Σ for Merge; hence it follows that the extensions of Merge can be ruled out without recourse to MY. Recall that according to Optimal Computation, Search Σ cannot directly select c, which is not a member of WS₁. In order to select c in {b, c}, Search Σ must first select {b, c}, a member of WS₁.