On wh and subject positions, the EPP, and contextuality of syntax

1 Introduction

I will start by enumerating puzzles and issues that this paper will discuss. While at first sight it may appear that at least some of these issues are unrelated to each other, the discussion will show that they are in fact very much related. The first, and the main puzzle, that this paper will address concerns what I will refer to as the who left effect, regarding subject wh-questions like (1).

There is a considerable debate regarding how (1) should be analyzed, the main question being whether who is in SpecCP or SpecTP. The answer will be neither. In fact, we will see, based on broad crosslinguistic data, that the derivation that is most often assumed for (1), given in (2), clearly cannot be right, which will be shown to have numerous consequences, especially regarding the status of the EPP.

A number of related cases crosslinguistically will be examined regarding what looks like local subject A’- movements, where it is not clear whether something is in left periphery or SpecIP, e.g. the controversy whether subject V-2 clauses in Germanic are CPs with the subject in SpecCP, or IPs with the subject in SpecIP (cf. (3)), locative inversion, clausal subjects, imperative subjects, various non-agreeing subjects (including subjects in languages without agreement) …, which will all be unified with the who left effect.

Another question that will be addressed, and shown to be related, is: Were we right to unsplit INFL? The answer will be no (quirky subjects will be one of the issues addressed during that discussion, with a new generalization proposed regarding their availability crosslinguistically). The answer will in turn be shown to have a bearing on another issue, delineating left periphery and the inflectional/A-field. Examination of left periphery and subject positions (so SpecIP and higher) will show that this cannot be done, which will in turn indicate that left periphery has no theoretical status (given that it cannot even be delineated).

The overarching concerns during the discussion will be what all this tells us about wh and subject positions, the EPP, and the syntax more generally. The answers to these concerns will lead to fine-tuning the position of various subject and wh-phrases, split IP, and a new conception of the EPP that situates it within a broader theoretical move toward contextuality in syntax more generally. In this respect, the paper argues that the locus of the EPP effect is not a fixed, unique position, but varies depending on the syntactic context, the key to seeing this being who left, i.e. local subject A’-movement.

I will start the discussion in the next section with the who left effect. Section 3 compares examples in (4) and argues that who in (4b) is not as high as who in (4a) but is higher than Amy in (4c).

This is then extended to a number of phenomena crosslinguistically (e.g. only-subjects, subject-restricted relativizations, aux-contraction, Defaka focus-movement, Norwegian C-marking, Hong Kong Sign Language rightward wh-movement, imperative subjects …), providing us with a new tool to address a number of well-known puzzles. An important concern during the discussion will be the status of the EPP. Section 4 examines quirky subjects. Section 5 focuses on certain asymmetries in Germanic V-2. Sections 6–7 discuss a number of non-nominal/non-agreeing subject constructions, including locative inversion, clausal subjects, and non-agreeing subjects in Singlish. Section 8 discusses subjects in languages without agreement (Japanese and Chinese). Section 9 examines constructions where objects have been argued to occur in the subject position. Sections 10–11 examine the consequences of the current discussion for the nature of the EPP, situating the main proposal this paper makes regarding the EPP within a broader move toward contextuality in syntax, which also argues against regarding syntax as fully Markovian in nature.

2 Starting point: the who left puzzle

Let us consider subject wh-questions like (5) in more detail.

It is sometimes suggested that who is in SpecIP in (5) (e.g. Carstens et al. 2016; Chomsky 1986; see also fn 2).^[1] There is, however, strong evidence that this is not the case (see Bošković 2016a, 2021b; Messick 2020 and references therein). Thus, (6) shows that hell phrases, which modify only wh-moved DPs, can modify subject wh-phrases (Pesetsky 1987). Also, the possibility of inverse scope in (7b) shows that object everyone can scope over a quantifier in SpecIP. The impossibility of inverse scope in (7a) then indicates that who is not located in SpecIP in (7a) (Mizuguchi 2014). Furthermore, if sluicing involves wh-movement followed by IP deletion, as standardly assumed, the wh-phrase in (8) cannot be in SpecIP (Agbayani 2000). (6)–(8) thus provide evidence that who does not stay in SpecIP in (5).

There is also a lot of evidence that examples like (5) crosslinguistically do not involve subject wh-movement through SpecIP. Thus, in a number of languages with both SV and VS order where in the latter the subject doesn’t move to SpecIP, the two orders are associated with different subject-agreement morphology. What we get in who left is the morphology associated with the VS word order (e.g. Fiorentino). This shows wh-movement to SpecCP cannot proceed via SpecIP or we would get the SV order morphology. The same point can be made with languages where the agreement morphology associated with subjects in SpecIP must be dropped in who left (Kinande, Kaqchikel; see e.g. Bošković 2016a; Erlewine 2016; Schneider-Zioga 1995). This is illustrated for Kinande in (9), where instead of the usual agreement morphology we get what is referred to as anti-agreement in subject questions (compare (9a-b) and (9c)).

This also holds for quirky subjects, as in Icelandic (10). (10a) shows experiencers block agreement with lower nominatives. However, an NP-trace does not induce a blocking effect, as shown by (10b), where the experiencer is a quirky subject, i.e. it moves to SpecIP. On the other hand, (10c) shows a wh-trace does induce a blocking effect. Now, if the experiencer in (10c) were to move to SpecIP before wh-movement, the intervener would be an NP-trace, and (10c) should pattern with (10b), which it doesn’t. The wh-phrase apparently does not, in fact cannot, move to SpecIP in (10c); it must move directly to SpecCP.

Another argument comes from auxiliary(aux)-contraction (Bošković 2021b). Aux-contraction is not possible when the aux is followed by a wh-trace (Bresnan 1971; Kaisse 1983; Bošković in preparation argues that this holds when the auxiliary and the wh-trace are located in the same phase).

That aux-contraction is allowed in (12) then indicates that wh-movement in (12) does not proceed via SpecIP, leaving a wh-trace in that position (note that I assume that the aux is in C here^[2]).

Additionally, Bošković (2021a) provides a number of arguments that null C induces Comp-trace effects, as would be expected under a syntactic account of the effect (this means there is no CP in the embedded clause in who do you think left, see also Bošković 1997; Chomsky 2015; Rizzi 2006; Rizzi and Shlonsky 2007). One argument comes from (13), which parallels the familiar Comp-trace effect paradigm: object and adjunct movement are allowed, but subject movement isn’t. Given the parallelism, Bošković (2021a) gives a null-C-inducing-a-Comp-trace-effect account of (13), and a number of additional paradigms.

Given that null C induces a Comp-trace effect and that who in (5) does undergo wh-movement, a question arises: why doesn’t a Comp-trace effect arise with subject wh-questions like (5), which should involve movement crossing C (the C is null but that doesn’t matter)? The answer is straightforward if subject wh-movement does not pass through SpecIP. What is subject to the Comp-trace effect is the trace in SpecIP. Thus, it is well-known that in languages where subjects can stay postverbally, without raising to SpecIP, the Comp-trace effect is inoperative. The same holds for (14), where there fills the subject position.

If subject wh-movement does not pass through SpecIP, (5) in fact parallels (14), rather than (13c), in the relevant respect (notice that the claim here is that only subject wh-movement to interrogative SpecCP does not pass through SpecIP in English, for reasons to be discussed below).

Conclusive evidence that local subject questions do not involve wh-movement via SpecIP is provided by the following West Ulster English (WUE) data, noted by McCloskey (2000).

Unlike standard English, WUE allows Q-float under wh-movement (16). Such Q-float is also possible in (15a). Still, just like standard English, WUE disallows (15b): a subject in SpecIP cannot float a quantifier in the postverbal position in passives. This rules out the derivation where who in (15a) moves to SpecCP via SpecIP since all would then float under movement to SpecIP. This is disallowed, as (15b) shows. Note that the who-in-SpecIP option is also ruled out by this paradigm.

I conclude therefore that who does not pass through SpecIP in who left/(5). How is the EPP satisfied in (5) then? In the following I address the issue by arguing that there are two wh-positions, a higher and a lower one, where the lower wh-position is occupied by wh-moved subjects. I will then show that this resolves the EPP issue in a way that also has important consequences for our understanding of the EPP.

3 Evidence for two wh-positions

4 Expanding the EPP domain: quirky subjects

So far we have (68) for different subjects:

I will now explore the possibility of additional subject positions. Consider first the argument for a return to split IP from Bošković (2020): Given that bar-level coordination is disallowed, (69), where the subject is outside and the modal inside the coordination, provides evidence that the subject and the modal are not in the same phrase, the modal being lower than the phrase whose Spec the subject occupies, which means IP should be split (see Bošković 2022b and references there for additional arguments to this effect).

Split IP leaves room for additional subject positions (for relevant discussion, see also Cardinaletti 2004). In this respect, Bošković (2019) argues that non-agreeing quirky subjects (see Icelandic (71)) are lower than agreeing subjects (in early minimalist structure, which split IP into AgrsP and TP, XP would be AgrsP and YP TP – non-agreeing subjects are naturally not located in SpecAgrsP).

There is a poorly understood variation regarding quirky subjects (English disallows them). One of the reasons is that, as Poole (2015) notes, quirky subjects crosslinguistically do not behave uniformly regarding Zaenen et al’s (1985) tests. I will take the possibility of binding subject-oriented anaphors as a diagnostic for true quirky subjects. This enables us to finally shed light on what is behind the relevant crosslinguistic variation. More precisely, a new generalization then emerges regarding the availability of quirky subject constructions. A typological survey of the literature reveals that quirky subjects are allowed in Icelandic, Faroese, Laz, Kannada, Korean, Malayalam, Spanish, Telugu, Japanese, Tamil, Imbabura, Polish, Georgian, Russian, Basque, Old French, Marathi, Guajarati, and Hindi. What these languages have in common is that they all allow pro-drop (full or partial). This then leads to the new generalization in (72) (with binding of subject-oriented anaphors taken as the relevant diagnostic):

(72) indicates that pro-drop is required for quirky subjects. Why is that the case? This can be captured if quirky subjects are not located in the same position as regular agreeing subjects (i.e. SpecAgrsP). Pro would then be needed for the regular subject position to satisfy the EPP (it would be an expletive pro), hence only pro-drop languages would allow quirky subjects.

There is an alternative (where quirky subjects themselves satisfy the EPP) if in pro-drop languages there is no separate AgrsP; rather, T has φ-features in such languages (this would essentially be T that is strong enough to label in Chomsky 2015). The suggestion is that quirky subjects cannot go to SpecAgrsP (for relevant but different discussion see Citko et al. 2018), so only languages without AgrsP can have quirky subjects and only pro-drop languages are like that. The suggestion that T can have φ-features only in pro-drop languages, which are typically morphologically rich,^[16] in a way recasts Chomsky (2015); it is not a weakness for labeling that is the result of morphological “deficiency” (as in Chomsky 2015) though but the need for AgrsP. In English, agreement is not morphologically rich, so AgrsP is needed; in Italian (corresponding to Chomsky’s strong T), this is not the case, so there is no need for AgrsP.

Regarding the two ways of treating quirky subjects noted above, an issue arises under the pro-in-SpecAgrP analysis. Quirky subjects do undergo movement, which would now be to SpecTP. If there is a pro in AgrsP that satisfies the EPP, why would they need to move to SpecTP? It appears that under this particular analysis we need to appeal to accounts where quirky subjects have a structural case in addition to the morphologically overt case, with that structural case requiring movement, as in Bošković’s (2007) system. On the other hand, under the analysis where T has φ-features in the relevant languages, and AgrsP is lacking, movement of quirky subjects can be taken to take place for EPP reasons.

Let us compare more closely the latter, T-with/without-φ-features account with Chomsky (2015). In Chomsky (2015), T has φ-features in all languages but there are languages with weak and strong T, where weak T needs a Spec. The current analysis also has a two-way split, the main difference being that while Chomsky’s weak T (i.e. its projection) simply needs to undergo another merger after it enters the structure, what corresponds to it in the current analysis more specifically needs to merge with Agr. T which has φ-features need not merge with Agr, T which does not have φ-features needs to merge with Agr. In other words, one way or another, φ-features need to be represented in the structure.

An alternative way of looking at all this is the following: finite T has φ-features in all languages (as Chomsky assumes); the relevant difference is whether or not these φ-features need to be checked against an Agr head. If they do, an Agr head needs to be present in the structure.^[17] There is another option where T still always has φ-features. Saito (2012) argues heads can be base-generated in an adjunction structure, in which case one of them excorporates, undergoing head-movement. Somewhat similar is a proposal in Martinović (2015), where a head X is generated with features F1, F2 … Some of these features can move out to project a phrase on their own. Applying a version of these approaches to the current proposal, T would always start with φ-features, in some languages (those that don’t have pro), φ-features would excorporate, projecting AgrsP. In those languages Agrs is the locus of the EPP effect (putting aside A/A’P). So, under the T-always-with-φ-features account, there is still a similarity with Chomsky (2015) in that languages differ in the properties of T, but the account is still rather different from Chomsky (2015).

Above, I suggested agreeing preverbal subjects like Mary are in SpecAgrsP (70). It appears that under what was said in the preceding passage, this would not be the case in pro-drop languages like Spanish. However, all we need to keep this suggestion for Spanish is to optionally allow φ-excorporation from T in such languages. The idea here is that in languages like English, due to their morphological properties, such excorporation must take place. In languages like Spanish, the morphology does not require it, but nothing prevents it from taking place. Quirky subjects are incompatible with AgrsP, they are located in SpecTP. I suggest that agreeing nominative subjects are incompatible with SpecTP, if they move preverbally they must then be located in SpecAgrsP. Agreeing preverbal subjects are then always in SpecAgrsP. Quirky subjects, on the other hand, are in SpecTP. In languages like English, there is always SpecAgrsP. In languages like Spanish, it is there with agreeing preverbal subjects, otherwise it is not.

Some Spanish-style languages show a rather dramatic agreement difference between pre and postverbal subjects. In Trentino and Fiorentino only preverbal subjects agree; postverbal subjects don’t agree, resulting in default V-morphology. The difference is not surprising under the current approach: AgrsP must be present for agreement to be morphologically manifested in Trentino and Fiorentino (cf. also fn 26).

The suggestion that in languages like Spanish and Icelandic (see Section 8 for languages like Japanese, which lack agreement), AgrsP is present when there is an agreeing preverbal nominative subject, but not otherwise, i.e. not with quirky subjects, provides a new perspective on the well-known Icelandic subject anaphor paradigm in (73). Icelandic in principle allows subject anaphors in subjunctive clauses, but not always. This is possible in (73b), with a quirky subject, but not in (73a), with a nominative subject.

There is a line of research that treats subject anaphors of this sort in terms of an anti-agreement effect (Rizzi 1990b). The above suggestion regarding when AgrsP is present provides a new way of implementing an analysis along these lines for (73), where the agreement effect in (73) is about the presence versus absence of AgrsP, with AgrsP necessarily lacking with subject anaphors. Given the above discussion, AgrsP is present in (73a), but not (73b). A subject anaphor is then only possible in (73b). Note also that there is a nominative element in (74). However, given the above discussion, there is still no AgrsP in (74) (it is there only with preverbal agreeing subjects), hence a subject anaphor is also possible here.^[18]

More straightforward candidates for quirky subjects in SpecTP are cases of reduced non-agreeing clauses with non-nominative subjects. One such case concerns Russian infinitives with dative subjects:^[19]

(75) involves regular infinitives, involving no agreement (so it is nothing like inflected infinitives in languages like Portuguese). What is ‘irregular’ about it is that it has an overt quirky-like subject in dative case instead of PRO. A plausible treatment of such constructions is to treat the infinitive as a TP (with AgrsP missing), with the dative subject in SpecTP (see Franks 1995 for a suggestion along these lines).

Icelandic ECM infinitives can also have quirky subjects (see e.g. Zaenen et al. 1985); if ECM infinitives are TPs such cases would also have a quirky subject in SpecTP.

5 V-2 subject/non-subject asymmetries in Germanic V-2 languages

All this can be extended to a lot of constructions/issues, e.g. V-2 asymmetries. There is a controversy regarding whether subject V-2 clauses in Germanic are CPs or IPs: subject V-2 clauses (76)/(79a) in several respects differ from non-subject V-2 clauses (77)/(79b), which was used to argue that they are IPs (Travis 1991; Zwart 1993), but they are also not exactly the same as regular IPs in (78) and their subject has different properties from both the subject in (78a) and (78b), where the subject is clearly in SpecIP, which was used to argue that subject-initial V-2 clauses are CPs (Schwartz and Vikner 1996).

We will see that the apparently contradictory data actually indicate that the subject in subject V-2 clauses is lower than SpecCP but higher than SpecIP, which can be captured if it is located in SpecA/A’P.

Subject/non-subject V-2 asymmetries have been used to argue that the subject in subject V-2 clauses is in SpecIP. Consider C-agreement in Dutch. In (80), the subject and the verb are clearly not in CP.

In non-subject V-2 clauses the verb in C bears C-inflection, not the usual V-inflection (81a). However, in subject V-2 clauses, the verb shows V-inflection, not C-inflection (81b), which suggests the verb and the preceding phrase are not in the same position in subject and non-subject V-2 clauses, i.e. (76) and (77).

A number of phonologically weak elements also show non-subject/subject V-2 assymetries. Thus, object pronouns in Dutch cannot cliticize to V in the former, but can in the latter (Zwart 1993). Additionally, subject and non-subject V-2 clauses differ in that the initial pronominal element can be reduced in the former but not the latter, which was argued to indicate a structural difference between subject and non-subject V-2 clauses (the initial element is not in the same position in (82a) & (82b), i.e. (76) & (77)).

A similar point can be illustrated with the weak pronoun es in German (the same form for subject/object).

The above patterns have been used to argue that subject V-2 clauses (see (76)) are structurally different from non-subject V-2 clauses (see (77)), in particular, that the former are IPs, with the subject in SpecIP, while the latter are CPs, as standardly assumed.

Schwartz and Vikner (1996), however, give cases where subjects in subject V-2 clauses (76) do not behave like subjects that are clearly in SpecIP (78a-b), which are problematic for the IP analysis. Thus, the two differ regarding adverb placement. In particular, adverbs that can precede the subject in (78a-b), which is clearly located in SpecIP (see (84a-b)), cannot precede the subject in (76) (see (84c)), the adverb also cannot precede the object in (84b), suggesting that the latter is higher than the former.

Also, the non-subject and the subject in (77) and (76) block extraction while the subject in (78a-b) does not, another case where the subject in subject V-2 clauses patterns with the non-subject in non-subject V-2 clauses, and differently from subjects that are clearly in SpecIP. (This can be seen when the relevant clauses are embedded, see Schwartz and Vikner 1996 for the data; note that the blocking effect is a bit weaker with the context in (76) than (77), as expected given the above discussion).

Furthermore, while (82)–(83) show that there are cases where a weak subject in subject V-2 clauses behaves differently from a similar non-subject in non-subject V-2 clauses, which suggests these elements are not located in the same position, there are also cases where a weak subject in a subject V-2 clause behaves differently from a weak subject that is clearly in SpecIP, suggesting these two are not in the same position either. Thus, unstressed expletive pronouns in German, Icelandic and Yiddish can occur in subject V-2 clauses (85), but they cannot occur in SpecIP (86) ((86) is acceptable without the expletive).

This three-way contrast in the distribution of weak elements suggests that the subject in subject V-2 clauses is not located in either SpecCP or SpecIP, which is indeed the case under the current analysis. In particular, the reasoning employed regarding the contrast in (82a) and (82b) which was used to argue that the subject in the subject initial V-2 clause is not in SpecCP, given that the initial element in (82b) is clearly in SpecCP, when extended to (85) and (86) should be taken to indicate that the subject in the subject initial V-2 clause is not in SpecIP either, given that the subject in (86) is clearly in SpecIP.

Putting it all together, subject V-2 clauses are not like non-subject V-2 clauses, but the subject of subject V-2 clauses is also not like subjects that are clearly located in SpecIP. The required three-way distinction is straightforwardly captured under the current account: the subject in subject V-2 clauses is in a position that is lower than SpecCP but higher than SpecIP – SpecA/A’P. (The initial element is still in a different position in (81b) & (81a) as well as (82a) & (82b), and (83a) & (83b). However, the subject in (84c) & (85) is also in a different position from the subject in (84a-b) & (86).) The current analysis thus reconciles the data which have been used to argue against the CP treatment of subject V-2 clauses and the data which have been used to argue against the IP treatment of subject V-2 clauses. The seemingly contradictory data are all rather straightforwardly captured, and reconciled, under the current analysis.

Another otherwise puzzling paradigm from Vikner (1995) falls into place, in a way that also confirms the mixed A/A’-status of the relevant subject position. Icelandic and Yiddish allow CP recursion. Although CP recursion is in principle allowed, wh-movement across a V-2 topic in (87) is not acceptable. Vikner argues this is due to relativized minimality (RM): (87) involves A’-movement across an A’-Spec.

Interestingly, subject wh-movement across a V-2 topic is allowed.

Vikner leaves this as an open issue, noting that (88) should also be ruled out via RM. The current system provides a new perspective on this contrast. In both cases we are dealing with traditional CP recursion (89). YP is clearly an A’-Spec. With object wh-movement, XP is also clearly an A’-Spec, resulting in A’-movement across an A’-Spec. However, under the current analysis, with subject wh-movement XP is actually a mixed A/A’-position, which may enable that movement to evade the RM effect.^[20]

Interestingly, Vikner notes that (87) also contrasts with (90), where the element in the lower Spec is an expletive (this also holds for Yiddish). While Vikner (1995) treats the expletive in (90) as being in SpecCP (taking (86) to indicate that it cannot be in SpecIP), noting the problematic nature of (90) under that assumption, I have suggested above that this expletive is actually located in SpecA/A’P. That can then help us explain the contrast between (87) and (90): while (87) involves A’-movement across an A’-Spec, this is not the case in (90), where the crossed Spec is also an A-Spec.^[21]

6 PolP rescuing effect

It is well-known that adverbs, which have been argued to be located in SpecPolP in such cases, can rescue that-trace effect violations, as in (91a). Topics, on the other hand, cannot do that, as (91b) shows.

PolP and topicalization thus differ regarding the that-trace effect. They differ in another respect: There is inversion with PolP, not with topicalization. I suggested above that inversion can make low left periphery part of the EPP domain. This may be exactly what is happening in (91a). Inversion (cf. under no circumstances would Ann leave) makes PolP part of the EPP field, which means the adverb satisfies the EPP. The subject then need not go through the Spec of the EPP field to satisfy it in (91a).^[23] Such movement is still needed in (91b) (see Section 5.1). The underlying assumption is that the that-trace effect arises if the subject moves to the EPP-satisfying position before moving to the Spec of that. Under the analysis suggested above, this is the case in (91b) as well as the standard that-trace effect (*who do we say that left), but not (91a). This suggestion may also provide an account of the contrast in (21) since it provides a rather different treatment of the two examples. In (21a), there is a subject in SpecAgrsP – I assume that whenever this is the case the subject satisfies the EPP (this is in fact the case in all examples discussed in the paper), which means PolP is not part of the EPP domain here (the suggestion is that inversion can make low left periphery part of the EPP domain only when there is nothing in SpecAgrsP, see also fn 23). In (21b), on the other hand, PolP can be part of the EPP domain. (21a) can then be ruled out either because it involves A’-movement across a pure A’-Spec or, alternatively, if wh-C cannot take A’-PolP as its complement. Neither issue arises in (21b): there is no crossing of a pure A’-Spec and there is no C taking an A’-PolP complement (PolP is mixed A/A’ here). Where is who then? It may exceptionally move to SpecCP (the wh-C would still not be taking an A’-PolP complement), or we may be dealing with multiple Specs of the same head, PolP, given that PolP is in fact A/A’P here (so object wh-movement couldn’t target it). Alternatively, the adverb may simply be pushed lower here, with no PolP present. Note in this respect Amy under no circumstances should ever hire Jon, where Amy is in SpecAgrsP, hence the adverb must be lower. All this also extends to (22): regarding (22b), note that Amy should under no circumstances ever hire Jon is also acceptable. As for (22a), the accounts of (21a) extend to (22a), with another option available: it is possible that inversion away from PolP is not allowed when PolP is a pure A’-projection.

7 Locative inversion and related constructions

I now turn to locative inversion (LI). As is well-known, LI PPs show a number of subject properties, e.g. subject-raising (92a) and the lack of WCO effects (LI (92b) contrasts with topicalization in (92c)).

They also show some non-subject properties: they block the extraction of lower elements (93a-b) and disallow inversion to C (93c).

They also disallow inverse scope (Kuno 1971), like A’-moved subjects and in contrast to regular subjects.

These properties are all captured under (95), where LI subjects move to a higher subject position than the regular subject position, where this higher position has mixed A/A’ properties, hence it also blocks A’-movement.^[24] Furthermore, since LI subjects are higher than regular subjects inverse scope is not allowed. As for inversion, if it is Agrs+T that undergoes it the intervening A/A’-head would block it.^[25]

Note that the higher position cannot be the Topic position; it must be a distinct position, as indicated by the WCO contrast in (92b-c). Also, local subject topicalization is disallowed in English, as Lasnik and Saito (1992) show. Thus, (96a) shows that a topicalized anaphor can get an antecedent from a higher clause; if short-subject topicalization were allowed the subject in (96c) should be able to undergo topicalization and get an antecedent from the higher clause, on a par with the topic in (96a).

It should be noted that there is a debate in the literature regarding whether LIs are subjects or topics, due to the conflicting data regarding this issue (e.g. (92) versus (93)–(94)). As in the case of subject V-2 clauses, the conflicting data are reconciled under the current analysis, in a way that also does not require positing a null expletive (which topic analyses need to) in a non-pro-drop language like English.

Regarding the domain approach to the EPP, all the projections from (95) belong to the EPP domain: the EPP requirement is satisfied in the highest projection present in Split IP. Furthermore, as discussed above, under the current approach to Split IP, non-nominative subjects do not move to SpecAgrsP. Since the EPP is satisfied in the final position of LI, the LI need not pass through SpecAgrsP.

Furthermore, it has been argued that there is a QP above wh-DPs undergoing wh-movement (Cable 2010) and that there is a similar projection above all phrases undergoing traditional A’-movement (see Bošković 2023b; Yoo 2018, so there is a topic/focus counterpart of QP with DPs undergoing topic/focus movement). This means that elements undergoing traditional A’-movement are in fact not DPs/nominals. In light of this, the right generalization may then be that non-DP/nominal subjects cannot move to SpecAgrsP, which is in fact what Bošković (2023b) argues for based on the system developed here.^[26]

This easily extends to clausal subjects ([that he left] is likely), which also show mixed subject properties, with a similar debate based on conflicting data (e.g. (97) versus (98)) as with LI (e.g., Stowell 1981 treats them as topics (this requires a null ex given the impossibility of subject topicalization) and Bošković 1995 as subjects). The conflicting data can be reconciled if they are treated as SpecA/A’P subjects.

The discussion of LI and clausal subjects can be extended to (at least some) there+V constructions.

For most speakers, there+V constructions differ rather significantly from there+be constructions. There is a locality effect ((100); Hartmann 2011 shows it’s the same as with LI) and inversion is degraded (101).

There in the there+be construction does not show these effects ((102)–(103), cf. also (14)).

This can be captured if there in (99) is in SpecA/A’P, on a par with LI. There is a controversy about whether subject there is nominative (compare e.g. Chomsky 1995 with Bošković 1997; Lasnik 1995a). From the current perspective, where non-nominative subjects cannot be in SpecAgrsP, both sides of the debate may be correct, with the different patterns in there+be and there+V attributed to the different properties of there (there not being nominative in the latter; for another perspective, see Hartmann 2011).

There are other phenomena where LI, clausal subjects, and there+V pattern together (and differ from there+be), which further argue for a uniform treatment of these constructions. One is ECM.

For-infinitives provide another relevant case. There in there+V and there+be constructions behaves differently regarding for-infinitives, with LI and clausal subjects patterning with the former.^[27]

To sum up the discussion so far, agreeing/nominative subjects (like Amy) are in SpecAgrsP, non-nominal and traditional locally A’-moved (they are also non-nominal) subjects are in the Spec of a higher phrase, and quirky subjects are in SpecTP. The EPP is satisfied in the highest projection present in this domain.

Under this analysis, a question arises concerning the status of wh-in-situ subjects, as in who thinks that who left, in particular, whether they are located in SpecA/A’P or not. The hell-test suggests they are not located in SpecA/A’P. (111) indicates that both wh-phrases in SpecCP and wh-phrases in SpecA/A’P are compatible with the-hell modification, while wh-phrases that are clearly in-situ (i.e. located in a position that is not associated with any A’-features) are not compatible with it. Given this, (112) suggests that wh-subjects-in-situ, at least in English, are not located in SpecA/A’P. (This means the non-nominal projection above wh-DPs discussed above is present only under wh-movement, as in Yoo 2018).^[28]

A note is now in order about quirky subjects regarding languages where they are somewhat but not fully ‘subjects’. One possibility is that those quirky elements are actually in SpecA/A’P (see Bošković 2023b, who argues (regarding Serbo-Croatian) that such quirky subjects are in fact PPs, fitting the claim that SpecA/A’P is a position for non-nominal subjects). Recall quirky elements crosslinguistically do not behave uniformly regarding the Zaenen et al. (1985) tests, which means a simple subject/non-subject cut is not enough. The above analysis provides a tool for capturing the variation in passing quirky subject tests, given that it provides options for the location of actual quirky subjects, SpecTP and SpecA/A’P.

There are many other candidates for SpecA/A’P, e.g. quotative inversion and subject topics of Singlish no-agreement constructions (113) (see Lee 2022, 2023 on the latter). As Lee shows, the subject of Singlish no-agreement constructions is incompatible with topicalization (114); just like topicalization, it is subject to a definiteness restriction (115) and cannot occur under verbs like regret, which disallow topicalization (116). This topic status of the subject in (113) makes it a natural candidate for SpecA/A’P.

Recall also the scope paradigm discussed above, repeated below, where (118)–(119) indicate that subjects in SpecA/A’P disallow inverse scope, in contrast to subjects in the regular subject position (117). Importantly, Singlish non-agreeing subjects also disallow inverse scope (120) (see Lee 2022).^[29]

The A/A’P account of non-agreeing Singlish subjects may also be appropriate for Chinese (the Singlish construction in question in fact represents the influence of the Chinese languages, see Lee 2023), given that Chinese subjects are also subject to a topic-like discourse effect, namely a specificity requirement, and disallow inverse scope.^[30] The issue is discussed below, in a way where the lack of agreement in the Singlish construction in question is not an accident but crucial to its A/A’P treatment.

At this point it should be noted that there is a potential alternative where our SpecA/A’P would actually be a somewhat different SpecTP, i.e. SpecTP with an A’-feature (perhaps getting that feature from C via Chomsky’s 2008 C-T association). Another option could be that the A/A’P cases involve CP/TP conflation (see in this respect Aldridge 2022). However, due to the lack of a height difference that the current proposal provides, such analyses could not account for scope contrasts noted above, like the contrast regarding inverse scope between (61) & (62) (see also (63), (94)). The same holds for Q-float in (15), i.e. the contrast between (15a) and (15b), given that WUE allows Q-float under A’-movement. The blocking effect on inversion is also an issue for the SpecTP analysis ((93c), (98), (101); this is less clear for the C-T confla- tion analysis, though it is hard to make this analysis sensitive to different A/A’ constructions behaving differently regarding inversion, cf. (56)). Another issue for the SpecTP analysis concerns the C-marking data where a ‘C’ follows the subject, as in e.g. (31b). Furthermore, Icelandic/Yiddish examples like (88) are problematic for both analyses since they would require a TP, or a conflated TP, to precede a CP.

The strongest evidence against these alternative analyses may come from patterns like those discussed regarding Germanic V-2 in sec 5, where we have seen that there is a three-way height distinction between non-subjects undergoing A’-movement to SpecCP, subjects undergoing local A’-movement, and subjects in SpecIP, where subjects undergoing local A’-movement are lower than non-subjects undergoing A’-movement to SpecCP but higher than subjects in SpecIP (more precisely, the subject in subject V-2 clauses is in a position that is lower than SpecCP but higher than SpecIP). This can be captured under the current analysis, which provides a three-way height distinction, but not on the alternative analyses considered here, which maintain the two-way distinction of a simple SpecCP versus SpecIP analysis.

8 Lack of agreement

In this section, I will explore some possibilities regarding languages that lack agreement, like Japanese. There is a long-standing tradition, going back to Kuroda (1988) (for more recent work see e.g. Saito 2007), that treats such languages as structurally/syntactically different from languages like English – the lack of agreement then is not a superficial PF difference. The possibility to be explored is that in such languages, due to the lack of agreement there is no AgrsP or AgrsP is defective – it cannot have a Spec so the regular subject needs to go somewhere else, in particular SpecA/A’P.^[31] Interestingly, Saito (2010, 2011) argues that the subject in Japanese is higher than in English, one of his arguments involving constructions like (121), which is ambiguous in English, but not in Japanese, where the subject has to scope over negation (122), suggesting that the Japanese subject is higher than the English subject.^[32]

To account for this, Saito argues the Japanese subject is higher than TP (the English subject is in SpecTP). More precisely, he argues that the EPP satisfying position in Japanese is a low head in the left periphery (he calls it Pred). This nicely fits the above suggestion (and the current system), where the relevant subject position would be SpecA/A’P. Even more interestingly, Saito suggests the reason why the EPP satisfying position in Japanese is higher than in English is the lack of agreement in Japanese, which is exactly the suggestion made above. This can be tied to another peculiarity of Japanese: multiple subjects. They may be possible in Japanese since the subject is not in SpecAgrsP, which would make the subject unique. More generally, in this language type we may expect subjects to exhibit quirks (which would not be necessarily the same in all languages of this type) that are not found in languages where regular subjects are in SpecAgrsP. One quirk was already mentioned, multiple subject constructions in Japanese.

Another relevant quirk may be the definiteness/specificity effect displayed by Chinese subjects, which is essentially a discourse effect (recall that A/A’P is a mixed inflectional field/left periphery projection).

Another property, shared by Japanese and Chinese, can be straightforwardly explained then: the languages allow subject anaphors, which is often considered to be a property of languages that lack agreement. This can be easily captured under the current analysis even if these languages do have AgrsP: Suppose that the binding domain for anaphors is AgrsP. This seems quite plausible given that even in English, anaphors that are higher than AgrsP can be bound outside of their clause, as in John thinks that himself, Mary likes. Japanese and Chinese would then allow subject anaphors for essentially the same reason. Their subject would be outside of AgrsP, like himself in John thinks that himself, Mary likes and unlike himself in *John thinks that himself likes Mary.

Another property of Chinese and Japanese can be captured under the above suggestion. They are known to have radical pro-drop, i.e. pro-drop that is not licensed by agreement. K. Zanon (p.c.) suggests that radical pro-drop in languages like Chinese and Japanese may be connectable to the lack of AgrsP/defective AgrsP. The intuition here is that if there is no AgrsP (or it is defective) it is not possible to impose the agreement requirement on pro. It is worth noting in this respect that Saito (2007) argues that radical pro-drop is in fact found only in languages without agreement (like Japanese and Chinese).^[33]

Another property of Japanese and Chinese can be captured. In Spanish-style agreement-licensed pro languages, overt pronominal subjects are used only when they are focalized, otherwise pro must be used. Montalbetti (1984) shows that these overt pronominal subjects cannot function as bound pronouns in Spanish. Thus, pro rather than an overt pronoun must be used on the bound variable reading in examples like Everyone _i claims that he _i is smart in Spanish. Since overt pronouns in Spanish are focalized, it seems natural to place them in SpecA/A’P under the current analysis. The effect in question may then be connectable to overt pronominal subjects being located in SpecA/A’P (this would not be the case with pro). Interestingly, although overt pronouns are not used only when focalized in Japanese, Japanese patterns with Spanish regarding the Montalbetti effect – an overt pronoun cannot be used in the example given above (see Jovović in press; Montalbetti 1984; the same holds for Chinese). This immediately follows if the effect in question is not focus-related but concerns pronouns located in SpecA/A’P, given that Japanese subjects are quite generally located in SpecA/A’P.

9 Non-subjects in SpecA/A’P?

A question now arises whether there are cases where elements other than subjects occupy SpecA/A’P. In this section I tentatively discuss some candidates for such treatment. It has occasionally been suggested for some languages that SpecIP is a mixed A/A’ position. Such cases seem to be candidates for a SpecA/A’P treatment, in fact non subject-dedicated SpecA/A’P given that the relevant cases in the literature were generally assumed to involve non subject-dedicated SpecIP. One nice consequence of the current treatment, which may favor it over previous alternatives, is that the proto-typical subject position in the inflectional field can still be considered subject-only (i.e. AgrsP would still be confined to subjects), which was not the case under the previous analyses (for proposals of this kind for various languages, see e.g. Bailyn 2004; Diesing 1990; Finer 1991; Goodall 1991; Lavine and Freidin 2002; Rögnvaldsson and Thráinsson 1990; for relevant discussion see also van Urk and Richards 2015).

One candidate for non-subjects in SpecA/A’P concerns Japanese scrambling. Scrambled O in the OSV order shows both A and A’-properties. This can be captured if it is located in SpecA/A’P. Miyagawa (2003) and Saito (2011) in fact argue that this scrambled O can satisfy the EPP in Japanese. In particular, this is the case in (123). Recall the Japanese subject must scope over negation (122), in contrast to English (121), which I interpreted as indicating that the Japanese subject is higher than the English one (following Saito 2011), in particular it is in SpecA/A’P. Importantly, the subject can scope under negation in (123). Following Saito, I suggest that this is because the fronted object in this case actually moves to the EPP-satisfying position, SpecA/A’P, satisfying the EPP (and leaving the subject lower). Japanese may then present a case of a non subject-dedicated EPP-satisfying position, which corresponds to SpecA/A’P.^[34]

As noted above, a number of authors have argued for various languages that SpecIP is a mixed A/A’ or not a simple A-position, and that it is not confined to subjects. In the current system, this SpecIP with mixed properties can quite generally be SpecA/A’P (with SpecAgrsP, the ‘real’ subject position, still being subject-dedicated). In fact, it is possible that if for independent reasons, the regular subject position is pushed higher up, into SpecA/A’P, as in the case of Japanese, it will no longer be subject-dedicated.^{[35] , [36]}

10 Contextuality in general

Above, I argued for a contextual approach to the EPP, where the locus of the EPP is not a single position, but varies depending on the syntactic context. The goal of this section is to place this approach to the EPP within a more general perspective regarding contextuality. Consider first the history of research on locality of movement. In the early bounding approach (Chomsky 1973), the trouble-makers for movement were defined rigidly: NP and IP were bounding nodes regardless of their syntactic context. Barriers (Chomsky 1986) were very different but the importance of one difference largely went unnoticed – the contextuality of Barriers. One cannot even ask whether e.g. CP in general is a barrier. Its status in this respect depends on the syntactic context that the CP occurs in; in Barriers, trouble-makers for movement were defined contextually. Chomsky’s (2000) early phase approach went back to the bounding approach in that it defined the trouble-makers, i.e. phases, rigidly: CP and vP are phases regardless of their structural position. This was soon followed by various contextual approaches, where whether XP is a phase depends on the syntactic context in which it occurs (on a par with Barriers, and in contrast to the bounding/early phase approach), see e.g. Bošković (2005), (2014), (2015), Bobaljik and Wurmbrand (2005), den Dikken (2007), Despić (2011), Gallego and Uriagereka (2007), Takahashi (2011).^[37]

Further, Bošković (2016b) argues that just like the phase status of a phrase is affected by the syntactic context in which it occurs, the concept of phasal edge, i.e. the status of a Spec regarding the PIC, is affected by the syntactic context in which it occurs: the highest phrase in a phasal domain functions as a phase, and the highest edge in multiple-edge contexts (α in [_XP α [_XP β [_XP γ) functions as the phasal edge.

There has thus been a consistent move toward contextuality in the locality of movement. The contextual approach to the EPP gains theoretical significance within this broader picture: It shows broader relevance of contextuality, contextuality now also being relevant in defining the EPP (in fact in the same way as for phases and phasal edges – there is a domain for phases/phasal edges/EPP, with the highest phrase in the relevant domain functioning as a phase, phasal edge, locus of the EPP effect).^[38]

The scope of the contextuality of syntax is even broader. Chomsky (2013) argues that labeling is also contextual. The same element behaves differently for labeling in different contexts – a phrase behaves differently in phrase-phrase and head-phrase mergers as well as in different phrase-phrase mergers, and its labeling status changes during the derivation, e.g. α in Who do you think [ _α who’ that she fired who’’] changes from an unlabelled object (before who’ moves away) to a CP (after who’ moves away).

Bare phrase structure is also very contextual: whether β is a head, phrase, or an intermediate projection depends on its syntactic context – its status also changes during the derivation: what is a maximal projection after a head and a phrase merge becomes an intermediate projection with further merger.

There is a parallelism here with the locality of movement: In the bounding node/rigid phasehood approach, one can look at a node itself, without paying attention to anything around it, and determine whether it is a bounding node/phase or not. This is not possible with Barriers/contextual phasehood. In GB phrase structure one only needs to look at a node to determine its phrase structure status, whether it is a phrase or a head (124a). In bare phrase structure, looking at any of the like-s in (124b) does not help in determining whether it is a head or a phrase, its status being determined contextually (an element that does not project (like) is a phrase and an element that is not a projection (like) is a head).

(124)

The A/A’ distinction is also now (i.e. in the phase system) contextual. To illustrate, movement out of vP must stop by SpecvP. The A/A’ status of a SpecvP depends on the nature of movement that stops by SpecvP: if we are dealing with A-movement (i.e. the position below and above SpecvP in the relevant chain is an A-position), the SpecvP counts as an A-position (also if SpecvP is the landing site of object shift); if we are dealing with A’-movement (as with wh-movement of the adjunct in (125) or long-distance movement of objects out of vP), the SpecvP counts as an A’-position; we need to look at the larger syntactic context to determine the status of a particular SpecvP regarding the A/A’-distinction.

Furthermore, Bošković (2015, 2016a, 2018, 2020) provides a uniform account of all island/locality-of-movement effects based on a contextual approach to phases and the labeling theory, which is also heavily contextual, where there are in fact no islands as this notion has been traditionally understood – there are no phrases that by their nature, independently of their syntactic context, disallow extraction (extraction is possible from all islands in well-defined contexts).

Additionally, Bošković (2023c) argues for the contextuality of spell-out. In particular, based on contextual phasehood and the labeling theory Bošković (2023c) argues that phase XP with YP in its Spec does not always behave in the same way with respect to spell-out: if YP is in a successive-cyclic movement position, the sister of YP is sent to spell-out, otherwise the full XP is sent to spell-out.

There has thus been a constant broader move toward contextuality/context-sensitivity of syntax which permeates many domains, including structure-building and labeling, the A/A’ distinction, formulation of locality domains (traditional islandhood as well as the status of phasal projections and their edges), spell-out, and now also the EPP (the contextuality of the EPP is essentially the same as the contextuality of phases and phasal edges, being defined in the-highest-phrase-in-the-relevant-domain terms).

11 More on the nature of the EPP

Above, I adopted the term EPP essentially for ease of exposition, what I really meant is the EPP effect. I have not provided an actual account of the EPP effect, which is still largely mysterious. In this section I look more closely into what the current discussion can tell us about the nature of the EPP.

First, the above discussion indicates that things like the EPP and the A/A’ distinction are not simple properties of the Merge operation itself, which would leave no room for contextuality. More precisely, they are not properties of the Merge operation in the sense that they can no longer be defined just through the Merge operation that creates the relevant position. The EPP used to be defined that way – it was assumed to be satisfied through the merger with T’/I’ in GB structure building, and TP/IP in bare phrase structure. Similarly with the A/A’ distinction: a position created by a merger with a case/θ-assigning head (or its projection) is an A-position. In that sense, the EPP and the A/A’ distinction were simple properties of the Merge operation. That, however, leaves no room for the required contextuality. They can then no longer be considered to be simple properties of the Merge operation itself.

There is a larger point to be made: they can no longer be considered simple Markovian properties. In fact, while put in very different terms, the above discussion confirms a conclusion from Chomsky (1957) regarding finite-state devices (Markov processes): they are insufficient to capture language. Finite-state devices are characterized by: an initial and final state (finite in number), a specification of transition from one state to another (not necessarily distinct), and a specification of a symbol printed when a transition obtains. (126) is a simple finite state device grammar that generates the man comes and the men come.

(126)

The transition from state 1 to state 2 prints the, from 2 to 3 man, from 2 to 5 men … A word restricts the choice of the next one, so *the man come is underivable. The system captures infinity. Inserting a loop and comes at state 4 would give the man comes and comes, the man comes and comes and comes …

A serious limitation of the system concerns non-adjacent relations. Once a finite-state device is in a certain state, it knows that state, where it can go from there, and what symbol to print. It does not know where it was, i.e. how it got to the current state, and how many times it was in it. It has no memory; as a result, it cannot deal with non-adjacent relations. This is clearly problematic when it comes to language, where non-adjacent dependencies that need memory are quite common. Consider e.g. (128), which derives missiles, anti-missile missiles (a device which neutralizes missiles), anti-anti-missile missile missiles … The best a finite-state device can do is (127) (anti ⁿ missile ⁿ ), which does not equal (128).

(127)

Either x or Y and the man who said we left is calling also illustrate non-adjacent dependencies. So, it is clear that a finite-state device cannot fully capture the nature of language. But is there anything in language that looks like a finite-state device? There actually is: compositional semantics. Consider (129).

A tenet of compositional semantics is that when computing the meaning we get by combining A and B, we do not know how we got to A or B; we only know A and B and what we get by combining them. In this respect, compositional semantics has no memory, and no look-ahead power either – it is like a finite-state device. Many PF processes, like affix hopping, require adjacency, which is in its nature Markovian.

While there are components of the grammar which are Markovian, a Markovian device is insufficient for syntax. Probably the most important things the theory of syntax needs to capture are infinity and non-adjacent/long-distance relations. A Markovian device can capture the former, but not the latter. Returning to the discussion of contextuality, contextuality is by its nature non-Markovian. The current discussion thus confirms the non-Markovian nature of syntax, considerably broadening its scope. As noted above, things like the EPP and the A/A’-distinction were treated as simple properties of the Merge operation, which essentially means they were Markovian properties. The contextuality of these properties the above discussion has revealed shows that a simple Markovian perspective on them cannot be maintained.

There is, however, one component of Markovian processes that has been endorsed in minimalism (see also Chomsky 2021). Recall that they are incapable of look-ahead. The lack of a look-ahead power is one of the tenets of minimalism. Consider the notion of local economy (Collins 1997). Under local economy, when evaluating the best course of action, all we know is where we are and what we can do next, nothing else, we cannot look-ahead to the steps after the next step, which are in fact Markovian properties.

A question then arises, do the contextual processes noted above require look-ahead? At first sight, it seems they do. But at closer scrutiny, this turns out not to be the case. Bošković (2014) in fact shows this for the contextual approach to phases discussed above. Consider in this respect the contextual approach to the EPP. Agrs merges into the structure, looking for a nominal to attract (only nominal elements undergo internal merge with it). If there is no nominal to attract, no internal Merge can take place – Agrs is done driving the derivation. The A/A’ head then merges with it. This actually tells us something about the nature of the EPP requirement. If a head has an EPP property, once it enters the structure, the EPP property is satisfied by merging the projection of that head either with a phrase that can satisfy the EPP or with a head that has the EPP property (the latter captures the propagation of the EPP property, see below). There is no disjunction here; informally, both cases involve an EPP merger.^[39] To implement this, I assume that phrases that can satisfy the EPP have a valued EPP feature. Heads like Agrs and A/A’ have an unvalued EPP feature. Once Agrs enters the structure, if there is a nominal, like Al in (130), to attract, Agrs attracts it – merger with the nominal satisfies the EPP property (I follow the standard assumption that pure Agree cannot satisfy the EPP, which requires movement – this is the strength property of early minimalism). If there isn’t, as in (131) (recall who is not a nominal), the A/A’ head merges into the structure.

The A/A’ head and Agrs undergo feature sharing in the sense of Frampton and Gutmann (2000), where two instances of an unvalued feature become one – when one of them gets valued they both do. The A/A’ head is capable of attracting who (what matters here is the matching A’-feature, not nominality), who then merges with it, valuing its EPP property, which automatically takes care of the EPP property of Agrs.

Note that there is no look-ahead here (I have only discussed convergent derivations; as usual, there are derivations that don’t converge – ‘bad’ numeration choices lead to crashing but crucially no look-ahead is needed). Furthermore, the percolation of the EPP requirement is captured without real percolation. The above simply implements the conclusions about the EPP reached in the prior discussion. Accordingly, two different heads, Agrs and A/A’, were involved, where the former (not surprisingly given its nature) can only merge with nominal elements capable of agreement (I mean traditional agreement, not Agree).

12 Conclusions

I have argued for a split IP and a contextual approach to the EPP, where the locus of the EPP effect is not a fixed, unique position, but varies depending on the actual syntactic context, the key to seeing this being the who left effect. In this respect, I have argued that wh-subjects are lower than wh-non-subjects but still higher than regular subjects, as illustrated in (132):

The account was extended to numerous cases, including Germanic V-2 subjects, languages with subject only relativization strategies, only-subjects, Bùlì wh-movement, Defaka focus movement, Belfast English and Norwegian C-marking, BP nunca ellipsis, HKSL rightward wh-movement, do-ellipsis, imperative subjects, clausal subjects, locative inversion, there-V constructions, Singlish non-agreeing subjects, Japanese and Chinese subjects. Regarding the EPP, I have argued for a contextual approach to the EPP where, on a par with the contextual approach to phases, there is an EPP domain, with the highest projection in the EPP domain being the locus of the EPP effect (on a par with the highest projection in a phasal domain being a phase). The hierarchy of the subject positions discussed is given in (133), with a more detailed representation of the split IP/EPP domain in (134): agreeing/nominative subjects are in SpecAgrsP, traditional locally A’-moved subjects and non-nominal subjects are in the Spec of a higher projection, and quirky subjects in SpecTP. The EPP is satisfied in the highest projection in this domain.

A number of conclusions were reached concerning the nature of the EPP, left periphery, split IP, and most importantly, regarding a more general, broader theoretical move in syntax toward contextuality, which is by its nature non-Markovian, hence syntax cannot be characterized as fully Markovian by nature (though it does have some Markovian properties, like the lack of a look-ahead power).