2.1 Two issues for the default theory

The first claim to be demonstrated is that a theory with non-decomposed cases allows for any case to be the default. To have a concrete case to work with, consider again left-dislocation in English, see (12).

Schütze (2001) argues that me is the realisation of a case-less pronoun, an analysis which I adopted here as well. Assuming non-decomposable case values, one can model this fact as in (13). Here, me happens to be lexically specified for person and number only, but underspecified for case features, see (13a).

Assuming the Subset Principle of Distributed Morphology, the lack of case specification in (13a) makes me compatible with all 1.sg environments, regardless of case. This correctly entails that me can appear in the acc and in default (i.e., caseless) environments. However, the rule also allows me to be the realisation of nom. The reason why me does not actually surface in nom is because I is specified for this feature, see (13b). I and me therefore compete for insertion in the nom environment, and I wins because it is a better match. Me therefore only appears in the acc and in caseless environments, since it has no competitor there.

It turns out that this approach places no restrictions on the range of available defaults. It turns out that this is impossible to achieve as long as cases are considered non-divisible primes such as nom, acc, etc. To see this, consider the invented paradigm in (14).

Suppose now that in syntax, each case corresponds to a non-decomposable feature like [nom] or [acc]. In this setting, the rules that generate the paradigm (14) may look as in (15).

Each rule in (15a–c) perfectly matches a particular case. The rule inserting hin in (15d) is not specified for case. In this setting, hin can in principle appear in any case, but it loses in competition to other lexical items in all other cases but the dative. What is most interesting to us is that the form hin would also realise a caseless 3.sg form, leading to a system with a default expressed by the dative form.

Of course, it is not necessary to set up the rules like this; the nominative could be easily modelled as the default, if the form he was unspecified for case, while the other forms fully specified, as in (16). Such a system would lead to a nominative default.

In fact, any of the forms in the invented paradigm (14) could be inserted by a rule that is not specified for case (with the other rules fully specified). However, if any of the rules can be construed as unspecified for case, any case can be the default. This clashes with the observations in Footnote 1 that the literature only found default nom and to some extent acc, but no other case (like genitive or dative). Ideally, this state of affairs should somehow follow from the theory (see McFadden 2007 for the same desideratum); but it does not follow, as things stand.

Another issue for an approach with non-decomposable case values is undergenaration. This problem arises in paradigms with certain types of syncretism. To see that, consider the paradigm in (17) from German, discussed from the perspective of default case in Schütze (1997: 51–2).

Recalling that the default case in German is the nominative, the 3.sg.fem paradigm is unproblematic. To model it, we specify ihr as [dat], and sie is underspecified, see (18a–b).

This automatically leads to the consequence that sie is also the default, inserted as the best candidate in environments where case features are absent.

However, adopting a theory with non-decomposable case values, the 1.pl pronoun brings some issues. In order to model the fact that the form uns is found in both acc and dat, the form must be underspecified for case, as in (18d). But if uns is underspecified, then to model (17), we must specify wir as nom, see (18c). However, this wrongly predicts that uns would be the default case (appearing in the absence of case features), which is not the case.

To conclude, a default approach to default case does not fare well if non-decomposable case values are assumed. First, it fails to provide any restrictions on the range of possible defaults. Second, it fails to account for paradigms where a syncretic (elsewhere) form is not the default. The next section shows that this is not an inherent property of the default theory. Rather, it is the property of the specific case representations (non-decomposable case values) used in the reasoning.

2.2 Cumulative decomposition

This section introduces the cumulative decomposition (Caha 2009; 2013). It argues that if the decomposition is adopted, then the German facts can be easily modelled within the default theory. At the same time, the decomposition is also restrictive, and it rules out default genitives, datives and instrumentals, predicting that out of all the cases, only the nominative can be the default.

The cumulative decomposition has been proposed to capture a *ABA constraint on case syncretism. *ABA is a term devised by Bobaljik (2012), conveying the impossibility (hence the asterisk) of a particular pattern of syncretism, where the first and last term of a particular sequence are identical, while the middle term is different. For case paradigms, it has been established that there is a *ABA constraint on syncretism in the sequence nom–acc–obl (Baerman et al. 2005: 56, among others). The statement in (19) expresses this constraint:

Table (20) illustrates which patterns of syncretisms are allowed and which are not allowed under the constraint in (19). In the table, each row represents a particular pattern of syncretism, where identical letters indicate the identity of form. The first three rows present patterns consistent with (19), the last row is incompatible with it. obl is used as a placeholder for the relevant oblique case.

The current literature on *ABA in case agrees that the generalisation (19) puts some constraints on the decomposition of cases into features (McCreight & Chvany 1991; Johnston 1996; Caha 2009; 2013; Harðarson 2016; Starke 2017; McFadden 2018; Bergsma 2019; Smith et al. 2019; Zompì 2019; Davis 2021). To be able to derive the constraint (19), the literature relies on a type of decomposition where case features are privative, and their number monotonically grows as we go left to right in the *ABA sequence nom–acc–obl. This is shown in (21), where the nominative has just one case feature, and the number of features grows as we move to the accusative and to the oblique cases. Caha (2009) calls this cumulative decomposition.⁵

The cumulative decomposition (21) derives the *ABA constraint because the decomposition makes it impossible to write realisation rules that would generate an ABA paradigm. To see that, let us indeed try to write the rules needed to generate the ABA paradigm in (22).

In (22), ey occupies multiple cells, while bee is fixed to the accusative. This means that ey needs to be underspecified, and we need to specify bee for the features of the accusative, i.e., K1 and K2; see (23a). Any other specification for bee is out of question, given the paradigm (22).

Before we start discussing ey, note that if there were no other entries than (23a), bee would be expected to appear in the obl cell. This expectation arises due to The Subset Principle: since bee is specified for a subset of the features of obl, it is a candidate for insertion in obl. To prevent it from appearing in obl in the paradigm (22), we need to write a lexical entry for ey such that it blocks bee from surfacing in obl. The only way this can be achieved is as shown in (23b).

However, this forced specification makes it impossible for ey to appear in nom. The two lexical entries in (23a,b) therefore fail to generate the paradigm (22).

It is impossible to modify any of the entries in (23) in a way that the paradigm (22) is generated. To begin with, changing bee is a non-starter. Similarly, we could posit the following entry so that ey is able to realise nom:

This rule is applicable in all cases and it therefore correctly inserts ey in nom. In acc, the rule (24) loses with the rule inserting bee (see (23a)), which is a better match. However, the rule (24) also loses to bee in obl. Therefore, the combination of the rules (23a) and (24) generates and ABB pattern. In sum, when cumulative decomposition is adopted, there is no way to set up lexical entries in a way that an ABA pattern is generated.⁶

Let me now demonstrate that when the cumulative decomposition is combined with the default theory of default case, it only allows that the nom form is used as the default. This is shown in Table (25). Here the def(ault) case has been added as a special case, so that the cases ore ordered top-down in the order def-nom-acc-obl. Following the definition of def in (1), the case is added as a form with no case features.

Once the default case is added, it is easy to see that it becomes a part of the cumulative decomposition. Specifically, the bare ϕ of the default is incrementally augmented by one feature at a time as we go down the paradigm. Therefore, it is predicted that a *ABA constraint should hold over this paradigm, and the only case that can be syncretic with the default is therefore nom (see the first column). Paradigms where the default is the same as the accusative or an oblique – across a distinct nominative – are ruled out because they represent the ABA pattern, which is underivable.⁷

This is a good result because in most languages, the default is indeed nom (see Footnote 1). However, this is the point where English (and other languages like that) become problematic. The logic of the cumulative decomposition makes it impossible to have one form in def and in acc, across a distinct nom because this is a *ABA violation. Therefore, the discussion of English and similar cases will preoccupy me in the remainder of this paper starting from Section 3.

The final point of this section is to demonstrate that the cumulative decomposition also allows us to generate the (originally problematic) German paradigm (17). The paradigm is repeated for convenience in (26), with the row for default added on top.

Assuming the Subset Principle, the problematic paradigm of the 1.pl can now be captured as follows. Uns, even though it appears in multiple cases, does not need to be radically underspecified: due to the cumulative decomposition, the environments where it appears (acc and dat) share the features K1 and K2. This allows us to specify uns as in (27a).

Wir is then stored without any case features, see (27b). This specification makes wir applicable in any case, including the default. The surface paradigm is then derived by competition: wir appears in nom and def, where it has no competitor. Uns is used in acc, dat, where it wins over wir because it is more specific. Generalising, decomposing case allows us to capture the fact that a form that has a wide distribution in the paradigm (uns) does not necessarily have to be the default.

Before we leave this section, it should be noted that since syncretism is never necessary, the current theory also allows for the default to have a unique form distinct from nom. There are reasons to think that Japanese is such a language (Moritake 2008). In Japanese, hanging topic left-dislocations require a bare DP at the left edge of the clause, which is distinct from nom -ga or acc -o, see (28).

Within the current approach, this can be analysed as the default simply not being syncretic with any case, which is an option in the proposed system.⁸

Finally, note that even though default case has no case features, is not necessary that a noun in a default case would always be bare. Consider, for instance, the following Icelandic paradigm, where the accusative is Ø marked, while the nominative has an actual suffix.

Since in Icelandic, the default is syncretic with nom, the default inherits the quirk of the language, and shows an overt suffix. The reason why an affix appears even in the absence of any case features is analogous to pronouns. Pronouns are often portmanteau markers for ϕ features and case; so even in the absence of case, there are still ϕ features to lexicalise (see, e.g., (27)). Under the plausible assumption that Icelandic nominal suffixes are portmanteau markers for number and case, the suffix is needed even in the absence of case (see, e.g., Caha 2021 for an analysis of case-number portmanteau morphology compatible with this idea).

2.3 Conclusions

This section combined the idea that the default case is the absence of case features with the cumulative case decomposition. Relying on the default theory, we predict that out of all cases, the default case can only be identical to nom/abs. In most languages, this prediction is borne out, and we now have a principled explanation for why that should be so. However, we have to say something languages, where the default appears to be the accusative. The discussion of such languages occupies the three following sections.

3.1 Danish

This section details how the current proposal achieves the correct distribution of pronouns in apparent default-acc languages. The abstract paradigm proposed for such languages is in (30). The important thing about (30) is that there are two series of pronouns, deficient and strong, and that the default form β is a part of a licit AAA paradigm, namely that of a strong pronoun.

The deficient default cell is empty, reflecting the claim that the only true default environment – namely left dislocation – is a strong environment. The weak default (whichever form that would be) therefore never surfaces.

The empirical discussion looks at Danish (see Ørsnes 2002; Parrott 2009; 2021; Jensen 2019). In presenting the facts, I focus on the environments identified by Schütze (2001) as default in English (coordination, pronoun modification, gapping, fragment answers, left dislocation). However, in presenting the facts, I do not assume that all of these are default-case environments. As outlined in Section 1, the idea is that languages like Danish have the same distribution of nom, acc and def as German, and that the only special thing about them is the paradigm shape as given in (30).

I start from the fact that in some sentence types, Danish distinguishes deficient/weak object pronouns from strong ones by their placement (see Vikner 2005 for an overview). Weak object pronouns undergo the so-called object shift, and they are obligatorily found to the left of negation, see (31a). Strong (contrastive) objects are found after negation, see (31b). The morphological shape of the object pronoun remains constant across the two environments, suggesting a syncretism of these two cells, see the bottom row in (30).

It is relevant to note that object shift is impossible with modified or coordinated pronouns in Danish, see (32a,b) respectively.

This gives us reasons to think that coordinated and modified pronouns are strong in Danish. This is relevant since coordination and pronoun modification are two of the environments where Schütze (2001) assumes a default-case analysis.

Let us now turn to subject pronouns. (33a) shows a subject coordination. Coordination requires strong pronouns (recall (32b)). Therefore, the form that we see here (dig) is analysed as a strong nom. As expected (based on (30)), this is the same form as the strong object; recall dig in (32b).

The weak subject form du ‘you’ is in (33b). As expected, it is different from the strong subject dig.⁹

Let us now move to modified pronouns. Recall from (32a) that modified pronouns pattern with strong pronouns in Danish in not allowing object shift. They are therefore analysed in (34a) as strong nom. The form found here is the same as the object form (dem ‘them’), but different from the weak subject form, which is shown in (34b).¹⁰

Continuing with strong subjects, let us move to fragment answers. Recall from Section 1 that Merchant (2005) noted a strong tendency for languages to use DPs with matching case in this construction. In conformity with this universal, I analyse the form mig in (35) as a nom form, albeit the nom of a strong pronoun.

Once again, the point is that the strong-subject form mig is identical to the form of the direct object in conformity with (30).¹¹

As the final example of a strong nominative, consider the case of non-locally extracted subjects (Ørsnes 2002). The example in (36a) serves as a base-line example, showing a nominative (weak) subject han in the embedded clause. Some speakers allow for such subjects to be extracted and placed on the left periphery of the matrix clause. In such examples, the subject must have the form ham (identical to the direct object); han is impossible, see (36b). (e indicates the original position of the subject, the verb tror ‘think’ moves across the matrix subject Peter due to the V2 requirement.)

Ørsnes (2002: 338) argues that (36b) is not a default-case environment because the extracted subject moves from a case position, designated by e in (36b). The difficulty of analysing (36b) as a default case leads me to suggest that it is a strong nom form (predictably identical to the direct-object form).

Let us now move to the default. Recall from Section 1 example (2) that hanging-topic left dislocation requires def in German. In Danish (just like in English), we find here a form identical to the strong/weak object form. (37) provides some relevant examples.

This completes the overview of the different functions of the pronominal forms such as mig ‘me,’ dig ‘you’ or ham ‘him’ in Danish. In addition to being used as weak object pronouns (31a), they can also be used as strong object pronouns ((31b), (32)), as strong nominative pronouns ((33a), (34a), (35), (36b)) and strong default pronouns (37). On the basis of these facts, the paradigm (38) can be proposed, where one and the same form occupies all the relevant cells. The important point is that once this paradigm is in place, the same grammar of case can be proposed for Danish (an apparent default-acc language) as for German (a default-nom language).

3.2 Deriving the Danish paradigm

In order to generate the paradigm (38), I adopt two proposals from the literature. First, it is necessary to encode the distinction between strong and deficient pronouns. The approach I adopt is one where strong and deficient pronouns differ not only prosodically and in their position, but also in their internal morphosyntactic structure and their semantics. For example, Cardinaletti & Starke (1999) point out that strong pronouns require animate referents while deficient pronouns are compatible with inanimate referents, strong pronouns allow contrastive interpretations unavailable for deficient pronouns, etc.

Such differences suggest that strong pronouns have at least one additional morphosyntactic feature compared to deficient pronouns. This idea is supported by the fact that in some languages, strong pronouns morphologically contain deficient pronouns. For instance, the Czech deficient pronouns ho and mu ‘him, acc/dat’ are contained in the corresponding strong pronouns je-ho and je-mu ‘him, acc/dat.’

In the literature, there are at least two proposals implementing the idea that strong pronouns have more features than deficient pronouns. Cardinaletti (1994) proposes that weak pronouns lack the lowest projection of the extended NP, i.e., the NP itself and realise only higher functional projections. Cardinaletti & Starke (1999) propose that deficient pronouns lack a high functional projection, C_N. Abstracting away from the specific implementation, this paper encodes the distinction by using the feature str for strong pronouns. Deficient pronouns lack this feature.

The second assumption I make is that lexical insertion is governed by the so-called Superset Principle. The Superset Principle is for the current intent largely equivalent to the better-known Subset Principle, with the difference that insertion takes place when the lexical entry specifies a superset (rather than a subset) of the features in the node. I give the specific formulation to be used here in (39).

The reason why I adopt the Superset Principle is that the paradigm in (38) has one particular property, which is impossible to model by the Subset Principle. The property is that a special form (jeg) is found in an environment with the smallest number of features. I shall not discuss this issue for reasons of space, but see Christopoulos & Zompì (2023); Caha (2024) for a discussion.

With the two assumptions in place, let us turn to the question what lexical entries are needed to generate the paradigm (38). Consider first the lexical entry for mig. The Superset Principle says that if an item like mig is to be inserted in a cell (corresponding to a node in the tree), its entry must contain all the features of the given cell. The cell in (38) with the most features is the strong acc form. Since mig appears here, it must be specified for all the features of that cell, see (40a).

This specification makes mig a candidate in every cell of the paradigm in (41), since it contains the features defining each cell.

The reason why it does not appear in the weak nom is because of the competing entry in (40b). This entry does not spell out the str feature, and it therefore cannot appear in any of the strong-pronoun cells. Similarly, it fails to spell out K2, and it therefore cannot appear in the weak acc. However, we do find jeg in the weak nom, where it is a perfect match, and it is therefore a candidate for insertion. Since it is a better match than mig, it wins in competition.

To summarise, what has just been shown is that the paradigm (41) can be generated by the rules in (40), using the Superset Principle (39). The paradigm (41) provides the correct forms needed in all the examples in Section 3.1, assuming that the only default environment is left dislocation. There is no need to assume a default accusative for any of the other sentence types.

3.3 Oslo Norwegian

This section shows how the approach sketched above handles languages like German, where a nom default has been proposed. As an example of such a system, I discuss Oslo Norwegian. Table (42) compares Oslo Norwegian (on the right) with the Danish system on the left. Oslo Norwegian differs in that its strong nom form is identical to the weak nom, rather than to the strong acc. Predictably, in such a language, the default copies the shape of the strong nom. (It cannot copy the strong acc, as this would be a *ABA violation.)

Let me now show how the forms in the Oslo Norwegian paradigm are distributed in concrete sentence types. Like in Danish, object shift in Oslo Norwegian distinguishes strong and weak object pronouns, despite their formal identity. (43a) shows that weak (non-contrastive) pronouns precede the negation, while strong (contrastive) pronouns follow it. This is captured by assuming identity between weak and strong objects in both languages, see the bottom row in (42).

Like in Danish, object shift becomes impossible with modified or coordinated pronouns, see (44a,b) respectively.

The crucial difference between the languages is that Oslo Norwegian strong subject pronouns do not look like object pronouns. The object form ham ‘him’ (seen in (43)) is ungrammatical in (45a). Instead, one has to use han ‘he,’ which is the same as the regular subject pronoun, see (45b).

This contrast between Danish and Oslo Norwegian is captured in the paradigm (42) by placing into the strong nom cell a form different from the strong acc, but identical to the weak nom.

The same contrast emerges for pronouns in coordination. (46a) shows that the object form deg ‘you’ is impossible in subject coordination in Oslo Norwegian. Instead, we see a form identical to the weak subject pronoun (shown in (46b)).

Predictably, these differences between Oslo Norwegian and Danish also show up in fragment answers. While Danish uses a pronoun identical to the object form in subject fragments (recall (35)), Oslo Norwegian uses jeg ‘I,’ a form identical to the weak nom, see (47a).

(47b) shows that the nom in (47a) is not a default, but a form that faithfully reflects the case assigned in a fully sentential counterpart.¹³

Another contrast between Danish and Norwegian arises in the case of non-locally extracted subjects (Taraldsen 1981). The example in (48a) serves as a base-line, showing a (weak) nom subject han in the embedded clause. When extracted, the subject keeps the form han, see (48b). This contrast with the Danish (36b), which requires ham.

Let’s now turn to the hanging-topic left dislocation, which is the only construction that requires def case in German. In Oslo Norwegian, we find here forms identical to weak subjects. (49) provides two relevant examples.

This completes the overview of pronoun forms in Oslo Norwegian. In this dialect, we find one form used ambiguously as a weak object pronoun (43a), and a strong object pronoun ((43b), (44), (47b)). Differently from Danish, strong subject pronouns ((45a), (46a), (47a), (48b)) are identical to weak subject pronouns. As a result, the default form in (49) cannot be identical to the strong accusative, since that would be a *ABA violation. Instead, the default is predictably the same as the strong nominative, see (50), repeated from (42).

The lexical items that generate the Oslo Norwegian paradigm are in (51). Meg is specified for all the relevant features, i.e., as a strong acc. This makes it a candidate for insertion in all the cells.

The difference between Oslo Norwegian and Danish is only in the specification of jeg, see (51b). While in Danish, the relevant item lacks the str feature (recall (40b)), the Oslo Norwegian form jeg has it. As a result, Norwegian jeg can be used in the strong nom cell of the paradigm and, as a consequence of Elsewhere, it wins over meg both there and in the cell corresponding to the strong default. This way, the correct distribution of jeg and meg emerges.

3.4 Conclusions: macroparameters in a microparametric approach

The comparison of the two systems shows that on the current account, the traditional ‘parameter’ distinguishing languages with default nominative from languages with default ‘accusative’ can be reduced to a rather simple difference in the specification of post-syntactic lexical entries. On the current account, the only difference between the languages is whether subject pronouns like jeg are listed in the lexicon as strong nom (Norwegain) or weak nom (Danish). The rest follows automatically from this simple lexical difference. This type of explanation for morphosyntactic differences has been known as the Borer(-Chomsky) conjecture and it constitutes one of the desiderata of theoretical research up to this day (see Borer 1984; Fukui 1988; Baker 2008).

It is to be admitted that on this account, there is no single parameter that controls the difference between a ‘default acc’ language like Danish and a ‘default nom’ language like Oslo Norwegian. Rather, this macroparametric difference between ‘default acc’ and ‘default nom’ arises as a cumulative effect of several microparameters, represented by the specifications of individual subject pronouns. In this approach, a system with a language-wide ‘default acc’ or ‘default nom’ arises if the specifications of all subject pronouns are aligned the same.

There are reasons to think that the more nuanced, microparametric approach may be correct. This conclusion is suggested by the corpus study of Danish pronouns in Jensen et al. (2023). Their results indicate that different persons behave differently when it comes to the selection of the strong subject form, and this in turn suggests that there is no macroparameter after all.

To see this, consider first the fact that Danish strong subject pronouns exhibit much variation. Jensen et al. (2023) investigate this variation in constructions where subject pronouns are modified by a relative clause. For English, Parrott (2009: 288) reports that these constructions “are rarely used in modern spoken English,” but in Danish, they exist and fall under the general umbrella of modified pronouns. In this construction, the 3rd person singular pronoun has the form ham in 83% of cases, as in (52).

Contrasting with this, Jensen et al. (2023: 194) report that when the 1.pl subject pronoun is used in this construction, it strongly prefers to have the same form as the weak subject, i.e., vi ‘we,’ which is found in 89% of cases attested in the LANCHART corpus. An example is provided in (53).

An asymmetry of the same kind is found in coordinations. In this construction, there is a lot of normative pressure to use the weak-subject form (e.g., han ‘he’). Despite this pressure, Jensen et al. (2023: 204) found that in a corpus of the spoken language, 3rd person pronouns prefer the weak object form (i.e., ham ‘him’), which is used in 60% of the cases. On the other hand, the 1.pl pronoun has the weak-subject form vi ‘we’ in 100% of the cases; i.e., the weak-object form oss ‘us’ is not attested.

Summarising, the default theory of default case explored here relies on a microparametric approach: the pattern observed in the language as a whole is a consequence of an independent setting of multiple smaller ‘parameters’ (lexical entries). This allows for a lot of nuance in the patterning of individual pronouns. This may be seen as a drawback from a macroparamteric perspective, but it seems to be vindicated by the fact that individual pronouns may not pattern in the same way.

5.1 All default-case environments are strong-pronoun environments

This section provides evidence for the claim that the five environments identified as default in Schütze (2001) (repeated in (63)) are strong-pronoun environments.

Before showing this, recall that under the current analysis, only left dislocation is analysed as a true default environment. The remaining environments are analysed as featuring a strong nominative pronoun.

The first piece of evidence for the strong-pronoun status of (63a–e) comes from cross-linguistic comparison: all of the environments in (63) are strong-pronoun environments in languages that morphologically distinguish strong and deficient pronouns. (64a–e) lists the relevant French examples. In all of these environments, a strong pronoun is required.

There is also some English-internal evidence for the strong status of these environments. The evidence is of two different types. The first type concerns the phonological reduction of pronouns, the second type concerns the behaviour of it.

The phonological evidence is based on the observation that English deficient pronouns are able to undergo phonological reduction (see Quinn 2005: 74–5 and the references cited there). This is illustrated in (65a,b), where the pronouns drop their initial [h].

However, such phonologically reduced (deficient) forms cannot be used in any of the environments given in (63), suggesting that these environments are strong.

(66) shows that the reduced forms are unavailable in fragment answers (Vos & Veselovská 1999: 945). This indicates that fragment answers are a strong-pronoun environment.

Similarly, reduced pronouns cannot be modified, see (67a,b) (from Quinn 2005: 75), or coordinated, see (67c) (from Cardinaletti & Starke 1995: 7). Again, this indicates that these environments are strong.

Colley & Bassi (2022) argue that weak (reduced) pronouns are also unavailable in gapping. They use the contrast between the sentence (68a,b) to illustrate this. (68a) is a sentence that contains a verb and a weak non-contrastive pronoun. When the verb is elided in (68b), the weak pronoun is no longer acceptable; only strong (contrastive) pronouns survive, see (68c). When the object is a non-contrastive (full) noun, as in (68d), gapping is possible (though the sentence requires a context, for which see Colley and Bassi’s work). Colley and Bassi thus conclude that gapping is in general compatible with non-contrastive remnants, but not with weak pronouns.

Finally, reduction is also impossible in left-peripheral positions, illustrated in (69) for topicalisation.

Summarising, the examples (65)–(69) show that phonologically reduced forms are unavailable in the five environments identified in Schütze (2001) as default. If we take the impossibility of reduction to be a hallmark of a strong environment, this provides evidence that these environments are treated as strong not only cross-linguistically, but also internally to English.

The second type of evidence for the strong-pronoun status of the relevant environments comes from the behaviour of the pronoun it. In Cardinaletti (1999), it is treated as a weak pronoun. This is because it may refer to inanimates, which is a hallmark of weak pronouns. If it is weak, we expect it to be excluded in Schütze’s “default” environments. This seems to be the case, as the following examples show.

(70) is an example of left-dislocation. (70a) shows that it (a weak pronoun) cannot occupy the left-dislocated position, while a full NP can (70b).

Similarly, Cardinaletti (1999) notes that it is unavailable in fragment answers, see (71).

In the literature, it is also possible to find examples indicating that for at least some speakers, it is degraded when coordinated, modified and when the verb is elided; see (72a–c) respectively.

These data support the conclusion that coordinations, pronoun modifications and gapping are strong-pronoun environments.¹⁶

Concluding, this section discussed three types of evidence for the conclusion that all constructions identified in Schütze (2001) as default are strong-pronoun environments. If this is so, the paradigm (73) correctly predicts pronoun distribution in these constructions without the need to assume that the default cell is expressed by acc (it is expressed by strong nom, which happens to be syncretic with weak acc).

A prediction of the account is that other pronoun uses outside of this core paradigm are compatible with this analysis. In Sections 5.2 and 5.3, several facts will be discussed that may seem problematic for the current account, but turn out to be compatible with it under scrutiny.

5.2 Some strong-pronoun environments are not default-case environments

This section focuses on strong environments that fail to exhibit the expected case-invariant strong pronouns me, him, etc. These environments are initially problematic because a theory based on the paradigm (73) cannot deal with such cases. The section provides two examples of such environments and accommodates them by proposing a bifurcation within the class of strong pronouns.

The first strong environment that fails to exhibit the expected forms such as me, him, etc. are pronouns modified by only. To see the issue, consider first the fact that in French, this is an environment that requires strong pronouns, see (74) (from Cardinaletti & Starke 1999: 152).

There are reasons to think that only requires strong pronouns also in English. The first reason is that pronouns cannot undergo phonological reduction after only, see (75a). The second reason is that the weak it also cannot be modified by only, see (75b).

The data in (74) and (75) suggests that pronouns after only are strong. Yet, as shown in (76), the supposed case-invariant strong pronoun me cannot be used after only in subject position.

The second case of a strong environment where I, he etc. are the only option are contrastive uses of pronouns, as in (77).

In French, such environments disallow deficient pronouns, see (78a–b) (adapted from Cardinaletti & Starke 1999: 219).

In a similar vein, Cardinaletti (1999: 60) notes that it cannot be contrastive, see (79).

On the basis of these facts, I conclude that sentences such as (77) represent another environment where we see a strong pronoun different from me, him, etc.¹⁷ The overall picture of pronoun distribution in English can then be depicted as in (80).

In (80), the rectangle represents all kinds of environments where pronouns occur. A subset of these environments can be characterised as strong. These environments are represented in (80) by the larger ellipsis. This set of environments corresponds to the contexts where we find strong pronouns in French, which is so far the same set of environments where English pronouns cannot be phonologically reduced, and where it is unavailable for at least some speakers.

Section 5.1 argued that all environments that Schütze recognised as default-case environments are within this set. The current section argued that there are also strong contexts that disallow the case-invariant me/him/etc. The five “default” environments are thus properly contained in the set of strong environments, and they are represented by the smaller of the two ellipses in (80).

The bifurcation within strong environments (me vs. I environments) shows the need to recognise two different kinds of strong pronouns. This has been also concluded in Quinn (2005: 152f), who calls the two subclasses gracile and robust. Schütze’s environments are robust, and require the case-invariant me. Environments that are strong but do not show the case-invariant pronoun are gracile: these are contrastive pronouns and pronouns modified by only.¹⁸

This (final) analysis is depicted in (81). The new feature in the table are the gracile pronouns, which have been introduced in this section. The newly postulated class has no effect on the analysis of the original data set; the only new thing here is that once (81) is adopted, the existence of strong pronouns with the shape I does not invalidate the results that have been achieved.

In terms of the technical implementation, this update necessitates the introduction of an additional feature for robust pronouns, rob. Once this feature is introduced, the feature specifications of the three types of pronouns are as in (82).

As before, all pronouns have ϕ-features. All strong pronouns (now split into two sub-classes) have str. Robust pronouns have the feature rob in addition. Gracile pronouns are strong pronouns lacking rob. This feature distribution reflects the type of containment relations seen in (80), where strong pronouns are a special type of pronouns, and robust are a special type of strong. With the features in place, it is possible to model the paradigm in (81) using the rules in (83).

Me is specified for all features (K1, K2, str, rob); it can therefore appear in any cell. I is excluded in acc environments (for the lack of K2) and it is also excluded in robust environments (for the lack of rob). With the entries in (83), the distribution of the pronouns is as in (81).

To conclude: faced with the fact that some strong-pronoun environments require pronouns like I/he/etc., this section argued that strong-pronoun environments in English are of two types, gracile and robust. Robust pronouns appear in Schütze’s default-case environments and they are case-invariant, while gracile pronouns appear in the remainder of strong environments and have the same paradigm as deficient pronouns. This analysis makes the existence of a case-invariant strong pronoun (on which the analysis of default case rests) compatible with the observation that some strong environments require I/he/etc.

5.3 Variation and other puzzles

The current section continues the discussion of potentially problematic cases. The first issue to consider is variation. Specifically, some of the relevant constructions only allow me (see (84a–c)), while others show variation (see (84d–f)). The presence/absence of variation requires splitting pronoun modification into two different cases, namely pre-modification and post-modification, see (84c,d).¹⁹

In Schütze (2001), an attempt is made to attribute this variation to the variation in case assignment. When case assignment succeeds, forms such as I arise. When assignment fails, me is used as a default. Clearly, this account cannot be used in the present case, and an alternative explanation of the variation must be invoked.

Two alternative proposals have been pursued in the literature, both compatible with the current account. The first one is a ‘register-variation’ approach. This approach postulates, within a special prestigious register, different lexical entries for the forms I, we, etc., treating them, in essence, as robust (within that register). The idea of relying on different registers stems from the observation that where variation exists, forms such as I are consistently associated with high-prestige use, while the use of me is variably called ‘normal usage’ (Emonds 1986) or ‘non-prestige’ (Sobin 1997). See also Grano (2006); Parrott (2009; 2021); Lemon (2017).

In the current theory, the approach can be implemented by saying that the prestige register lists I as a robust pronoun, see (85b), while the non-prestige register treats it as gracile, see (85c). The lexical entry for me is the same in both registers, see (85a).

According to this proposal, speakers activate either (85b) or (85c) depending on the relevant social variables. If (85b) is used, its interaction with (85a) yields a paradigm typical of a nom-default language. If (85c) is used, we get the non-prestige paradigm discussed in (81).

It is relevant to note that in many cells, it is immaterial which of these rules speakers choose, because they yield the same results. The difference arises only in two cells of the paradigm, which are given in grey in (86).

This is an interesting result, because these cells correspond precisely to the five environments circumscribed as default environments by Schütze (2001). In Schütze (2001: 220), variation is linked directly to the fact that the prestigious forms can only override default case specifications, but not ‘real’ cases. The same result is replicated here (the set of environments with variation is restricted), but there is no need to circumscribe the relevant set of environments by the notion of a default.

This approach works well where there is free variation conditioned by register. However, such variation is not found in all cases. For instance, left dislocations and fragments require me and disallow I, recall (84a), (84b). One could argue that there is no variation in left dislocations and fragment answers because these constructions are only found in non-prestige registers. For instance, Gregory & Michaelis (2001: 1679) note, summarising the results of previous research, that “detachment constructions [by which they mean left dislocation] are inappropriate in formal register.” If this is so, the rule (85b) would be inactive in these environments, yielding me as the only option.

Another option that has been explored is to attribute the variation to morphosyntax. For instance, Quinn (2005) proposes that some of the variation can be analysed as a morphosyntactic variation in pronoun type (gracile vs. robust). As an example, consider the fact that in coordinations, the first conjunct and the second conjunct behave differently. As Quinn (2005: 121) points out, if we take the 1.sg pronoun out of the picture, then there is an asymmetry between the conjuncts in that the initial conjunct accepts gracile pronouns much more readily than the second conjunct:²⁰

Quinn (2005) suggests that this asymmetry can be understood if the second conjunct must be always stronger than the first. It is thus possible that the first conjunct is gracile, while the second conjunct is robust (she and him), but the reverse is much worse (*him and she). What is crucial for us is that this kind of an account allows for a degree of structural variation without invoking an alternation in case (i.e., between default acc and an assigned nom).

The idea of variation between gracile and robust forms may provide some interesting insights into the interaction between various strong environments (namely, when we embed one strong environment within another). For example, we already know that left-dislocations require robust pronouns (88a), while post-modified pronouns show variation (88b). It is therefore interesting to note that when a post-modified pronoun is left dislocated, one still gets variation, see (88c).

In Schütze (2001: 215–6), it has been suggested that the variation between we/us in (88b) may arise due to how case assignment interacts with different structures. He proposed that when the modified pronoun is in the topmost D projection, nominative assignment succeeds, and we appears. When it is not in the topmost projection (i.e., not in D but perhaps in Num), assignment fails, and us surfaces.

However, the example (88c) shows that the variation between we/us is also found in left-dislocation constructions, where no nominative case is assigned to the left-dislocated pronoun, so it is impossible to explain the form we in (88c) by recourse to the mechanism proposed in Schütze (2001: 215–6).

On Quinn’s approach, one could posit that phrases in left-dislocated positions must be always robust. This requirement is fulfilled by placing there a full DP (we/us Californians), and since this requirement is satisfied by the full phrase, the pronoun embedded inside it is free to show its regular variation between gracile and robust, as found in the regular case of post-modified pronouns.

Moving away from variation, the final puzzle to consider is the ungrammaticality of the sentence (89).

The issue with (89) is the following: if me is a (robust) nominative pronoun, and subjects are nominative pronouns of variable strength, the sentence (89) is expected to be grammatical. Since it is ungrammatical, it seems that me cannot be a nominative pronoun (contrary to the current analysis).

However, as observed in Kayne (2000: 167), French strong nominative pronouns cannot appear in such sentences either, see (90a).

Kayne proposes that in order to be felicitously used, such strong pronouns must be doubled by a weak pronoun, see (90b). Note at the same time that the doubling requirement observed in (90) only holds in sentences where the bare pronoun serves as a subject of a non-elliptical clause; none of the other strong-pronoun environments in (64) requires doubling in French (just like in English).

The parallel between the English me and the French moi suggests that (89) is quite likely ruled out for a different reason than me being accusative, since moi is routinely analysed as a strong pronoun invariant for case. As a speculation, let me pursue here Kayne’s idea that such sentences are ruled out not because of case, but because the robust pronoun must be doubled, as in French, see (91b).

The question is naturally how to define the doubling requirement so that it only applies in (91), but not in the other sentence types (where no doubling is observed). This can be achieved if one stipulates that doubling only pertains to bare pronouns. This exempts coordinated and both pre- and post-modified pronouns from the doubling requirement.

It leaves us with the question why fragment answers and gapping do not need doubling either. One option (considered in Kayne 2000) is that in these contexts, such doubling could be present in the underlying structure, but eliminated by ellipsis, see (92).

While the actual account is admittedly speculative, what matters is the parallel between French and English, such that subject-related me mirrors the behaviour of moi. This parallel includes apparent quirks such as the impossibility of using either me or moi as the sole subject of a non-elliptical sentence. While the ultimate explanation for this is left for future research, the conclusion I end up with is that the English-French parallel suggests that the issues are not a artefact of the analysis, but a genuine puzzle in the syntax of strong pronouns.