The first piece of evidence which I bring to bear upon this issue is the fact that agreement-drop constructions, unlike their fully-agreeing counterparts, are resistant to object topicalisation.

Examples (5a) and (5b) form a minimal pair, diverging only in the presence/absence of the standard third-person present tense agreement morpheme -s. That there is a contrast in acceptability arising from this minimal difference therefore demonstrates that the absence of the agreement morpheme bleeds the syntactic operation of object topicalisation.

Note that the ungrammaticality reported here cannot be attributed to any particular prosodic requirement, e.g. a requirement that there be a prosodic gap immediately following the topicalised object, such as is usually required in standard English, since the insertion of an equivalent intonational break, as indicated by the comma following the topicalised object in (5b), is insufficient to rescue the sentence from ungrammaticality.

Nor can the ungrammaticality be a result of some phonological constraint that rules out the realisation of a phonologically null element at the right edge of a prosodic phrase: the insertion of additional phonological material at the right edge does not eliminate the observed contrast between object topicalisation in fully-agreeing versus agreement-drop constructions (6a vs. 6b).

The incompatibility of object topicalisation with agreement-drop further extends to biclausal contexts as well. Where object topicalisation is usually licensed in the embedded clause (7a), it is again illicit when the agreement morpheme is unrealised on the embedded predicate (7b).

This alternation crucially cannot be attributed to the mismatch between the presence/absence of agreement on the matrix and embedded clauses respectively, as demonstrated by the acceptability of both (8) and (9).3

This state of affairs is incompatible with a PF-based account of agreement-drop, since the treatment of agreement-drop as a PF phenomenon would amount to the assertion that agreement-drop applies post-syntactically, and as such, cannot have any syntactic reflexes under standard models of syntax and its interactions with the interfaces. That is to say, under such a view of agreement-drop, the syntax should not be privy to how the agreement-morpheme is ultimately pronounced (or in this case, not pronounced). Since agreement-drop constructions and their fully-agreeing counterparts have the same underlying syntax, they are not expected to diverge in their syntactic behaviour.

Contrary to this prediction, what we have seen is that the loss of the agreement morpheme bleeds the syntactic operation of object topicalisation in the above examples. The incompatibility of agreement-drop with object topicalisation therefore poses a serious empirical challenge to the treatment this phenomenon as a PF one, and consequently, serves as evidence motivating a syntactic account of the alternation under study.

The next piece of evidence which I present to demonstrate that agreement-drop cannot be treated as a post-syntactic alternation appears in multiply-embedded clauses, in the form of a contrast involving adjunct extraction possibilities.

Consider first the standard English example (10). The embedded indirect object to John can be dislocated to the left edge of the embedded clause, yielding the perfectly grammatical (11).

Notably, it has been observed that this operation of embedded topicalisation blocks the extraction of adjuncts from any clause that is lower in the structure (Lasnik & Saito 1992). To illustrate, example (12) is ungrammatical under the indicated reading, where the extracted adjunct how associates with the most deeply embedded clause.

That is not to say that the linear string corresponding to (12) is itself ungrammatical, however. If construed differently, the linear string is perfectly acceptable; in particular, the sentence is grammatical if the adjunct is instead interpreted as an associate of either of the two higher predicates. In other words, the linear string corresponding to (12) can be interpreted, and is in fact two-way ambiguous between a question about the manner of saying (13) and the manner of thinking (14), but crucially, the interpretation where it is the manner of fixing that is being questioned cannot be accessed; ergo, the topic ‘to her friends’ blocks wh-extraction from a more deeply embedded clause, but crucially does not block local wh-extraction.

This alternation can be isolated and directly attributed to the embedded topicalisation structure, when we consider the otherwise-equivalent construction (15). (15) forms a minimal pair with (12), distinguished only by the fact that it does not have the same embedded topicalisation structure; unlike (12), (15) does indeed allow for the extracted adjunct to be interpreted as an associate of the lowest predicate, i.e. (15) is three-way ambiguous.

The contrast in accessible interpretations that arises between (12) and (15) therefore demonstrates that the syntactic operation of topicalisation can bleed the extraction of a lower adjunct.

With this as background, we now consider the case of agreement-drop constructions: if, as I have claimed, agreement-drop corresponds to the presence of a topicalisation structure within the syntax, we predict that agreement-drop should block the extraction of a more deeply embedded adjunct in exactly the same fashion as the topicalisation structures discussed above.

We should therefore expect to find that the contrast between (15) and (12) is replicated in the minimal pair comprising (15) and its agreement-drop counterpart.4 This is indeed what we find:

Crucially, the available interpretations for the extracted adjunct in (16) are exactly the same as those noted to be available for the embedded topicalisation example (12); i.e. while (16) cannot be interpreted as a question regarding the manner of fixing, it is still two-way ambiguous, and can be interpreted either as a question regarding the manner of thinking (17) or the manner of saying (18).

Again, if agreement-drop were truly the result of the instantiation of a zero allomorph in the PF, we would not expect the alternation between (16) and (15), which are distinguished only by the absence/presence of overt agreement respectively, to have an impact on the (un)availability of any particular interpretation(s), since the LF representations of the two constructions should, under such a view, be identical.

The existence of a semantic contrast between the two constructions therefore constitutes further evidence in favour of treating agreement-drop within the syntax, rather than post-syntactically as a PF alternation.

That there is again a parallel to be drawn between embedded agreement-drop and embedded topicalisation is also suggestive that the two phenomena should be unified, with the parallelisms deriving from a syntactic structure common to both types of constructions.

The parallelisms continue when we consider the interaction of agreement-drop (as well as topicalisation) with ‘regret’-class verbs. In particular, I demonstrate that just as ‘regret’-class verbs have long been observed to be unable to take clausal complements that have a topicalised nominal phrase in their left periphery (Hooper & Thompson 1973), they are similarly unable to take as their complements agreement-drop constructions.

It must first be noted that this incompatibility of ‘regret’-class verbs with topicalisation in their complement clauses cannot be extended to all predicates that take clausal complements, given that there are other predicates which do allow for embedded topicalisation in their clausal complements, such as say, which we previously saw in (7a), repeated below.

Instead, the incompatibility of ‘regret’-class verbs with topicalisation structures, as in (19b), has been suggested to follow from competing semantic requirements imposed by the verbs themselves, and by the topicalisation structure being embedded. More specifically, such verbs (also known as factives) are traditionally treated as triggering a strong presupposition that their clausal complements are true (Stalnaker 1974; 1978; Heim 1982; 1983); however, topicalisation also triggers an existence inference that holds of the topicalised nominal (which manifests in the form of the specificity requirement noted previously). It is the concurrent imposition of these two independent requirements on the clausal complement as well as the topicalised nominal contained within that results in the incompatibility noted above (see, e.g. Hooper & Thompson 1973).

Abstracting away from the exact implementation of the mechanisms underlying the satisfaction of the interpretational requirements of factives and topics respectively, such that they ultimately end up in complementary distribution, we are left with the fact that there is indeed an incompatibility, and, crucial to the purposes of this paper, that this incompatibility surfaces in agreement-drop constructions as well: while agreement-drop can in principle apply to embedded clauses (20), it patterns with topicalisation in that it is not allowed when targeting the predicate of the clausal complement of a verb such as regret (21).

This is unexpected under the PF approach to agreement-drop, since the agreement morpheme is, under such a view, syntactically represented even in the cases where the zero-allomorph was ultimately selected for realisation. The semantic conflict that holds between topicalisation and ‘regret’-class verbs should therefore be orthogonal to the exact form of the agreement morpheme, since there is no a priori reason why agreement-drop should trigger an equivalent conflict with the requirements imposed by ‘regret’-class verbs. We should therefore expect there to be no contrast between the grammatical fully-agreeing (19a) and the ungrammatical agreement-drop (21) is predicted, contrary to fact.

Taking the earlier observations into consideration, this paradigm is, conversely, unsurprising if the “subjects” of agreement-drop constructions are, in actuality, topics, since the embedded “subject” he in (21) would be, from the viewpoint of the syntax, a topic; the incompatibility between ‘regret’-class verbs and agreement-drop would consequently resolve into the aforementioned interpretational incompatibility between ‘regret’-class verbs and topicalisation structures.

Yet another piece of evidence in favour of a syntactic treatment of agreement-drop relates to the effects of agreement-drop on quantificational scope readings; agreement-drop voids inverse scope readings which are otherwise possible in their fully-agreeing counterparts.

Example (22) is ambiguous between a surface-scope reading, where the existential introduced by the embedded subject somebody scopes over the universal introduced by the embedded object everyone, and an inverse scope reading, where the universal scopes over the existential. That is to say, the sentence is adjudged true either if there is a specific unique individual such that everybody is loved by that individual or if, for each individual being considered, at least one individual loves them.

Interestingly, agreement-drop constructions do not display this same ambiguity. Instead, the inverse scope reading is completely ruled out in (23).

Again, the two examples (22) and (23) form a minimal pair, with the only point of difference being the realisation/non-realisation of the embedded agreement morpheme.

This therefore constitutes a case of semantic non-equivalence between the fully-agreeing construction and its agreement-drop counterpart; setting the exact analysis of the semantic alternation aside for a moment, the very fact that there is such an alternation at all indicates that a PF-based account of the phenomenon cannot be tenable, since a silent allomorph of the same underlying morpheme should not be able to affect the semantic component.

How is the blocking effect of agreement-drop on inverse scope obtained then?

One possibility is to appeal to differences in the structural height of the existential quantifier in fully-agreeing versus agreement-drop constructions.

Consider the standard English sentences in (24) and (25). As before, inverse scope is availabe in (24) such that the universal quantifier in the object outscopes the existential in the subject. However, if we question the subject of (24), as in (25), the universal quantifier cannot scope over the wh-element (see, e.g. May 1977). That is to say, (25) is only compatible with an answer where there is a single unique individual who loves each and every individual within the domain. (25) is crucially not amenable to an answer which comprises the concatenation of every individual’s distinct lover.

This indicates that while inverse scope can obtain over subjects in the canonical subject position (i.e. Spec,TP), it cannot obtain over elements which have undergone extraction to a higher structural position. The inability of inverse scope to obtain in the agreement-drop construction in (23) may therefore be yet another instantiation of the same effect: the existential quantifier in the “subject” is similarly too high for inverse scope to obtain, i.e. the “subject”s in (22) and (23) respectively cannot have been realised in the same syntactic position.

An alternative analysis appeals to the more general constraints imposed upon sentential topics, rather than their structural height. First, consider that indefinites cannot usually serve as topics, as demonstrated by the contrast in grammaticality between (26) and (27), where the topicalised nominals are marked by the definite article the and the indefinite article a respectively (see also Hankamer 1971; Kuno 1972).

Crucially, an exception is made when the indefinite is interpreted as being specific5 (cf. Erteschik-Shir 1997; 2007; Cohen & Erteschik-Shir 2002; Giurgea 2015: a.o.). Accordingly, (27) improves when the topicalised element is interpreted as a specific indefinite (cf. Davison 1984), i.e. the speaker intends for the reference of the indefinite to be a particular girl whom they have in mind. The same holds of the agreement-drop construction in (23): the construction is acceptable only if someone is interpreted specifically.

This specificity requirement is generally taken as an example of the so-called “exceptional wide scope” phenomena, and can be formalised in a number of ways. Indeed, the exact formalism has been hotly debated in the literature, with some of the options including the (forced) interpretation of the indefinite as referential (Fodor & Sag 1982), the imposition of some sort of existential closure, such as through choice functions or Skolem functions (see, e.g. Reinhart 1997; Winter 1997; Kratzer 1998: a.o.), and the stipulation that the specificity comes by way of presupposition (Yeom 1998). One analysis that is particularly aligned to the current paper is Endriss’s (2009) analysis of exceptional wide scope phenomena as being directly correlated to topicalisation.

The exact implementation of the specificity requirement is, however, in no way central to the present discussion. The critical observation is only that the specificity requirement itself exists as a so-called exceptional wide scope phenomenon. Ergo, the absence of the inverse scope reading in (23) is a result of the topical (quantificational) indefinite someone being exceptionally required to take wide scope over the universally quantified object everyone.

In fact, the exact implementation of the absence of inverse scope, while interesting, is itself orthogonal to the more general point that I am trying to make in the current section, which is simply that agreement-drop “subjects”, much like topics, do not allow for the possibility of inverse scope, contrary to the predictions made by the PF account of agreement-drop, though the observation is certainly in line with this quote from Endriss (2009): “Quantifiers receive wide scope due to their interpretation as topics” (p. 10).

Furthermore, the specificity requirement that may be the source of the forced wide scope interpretation of quantificational topics, in fact applies more generally to indefinites in the agreement-drop “subject”-position, as I will briefly demonstrate in the next subsection.

Existential quantifiers aside, universally quantifiers are themselves also indefinite, and as such, have been argued to generally be unable to serve as topics (Reinhart 1981). We would therefore, in light of the findings thus far, predict universally quantified external arguments to be ruled out exactly in the context of agreement-drop. This prediction is indeed borne out (28a vs. 28b; 29a vs. 29b).

The distribution of quantifiers in Singlish therefore provides another domain in which the “subjects” of agreement-drop constructions parallels aboutness-topics.

Lasnik & Uriagereka (1988) observe that wh-phrases in English cannot undergo topicalisation, on the basis of examples such as the following:

As such, we might expect that the wh-subject questions to be impossible in agreement-drop constructions, since the wh-subject would be expected to be forced to function as a topic, given the trend that we have seen throughout the present section.

The impossibility of topicalising wh-phrases appears to be variable cross-linguistically, however. Grohmann (2006), on the basis of earlier work (Grohmann 2000: a.o.), argues for the existence of wh-topics on the basis of German multiple wh-questions, arguing further that these wh-topics are subject to a constraint he labels Discourse Restricted Quantification (34).

What this means is that “all wh-phrases in a German multiple wh-question must be D(iscourse)-linked (possibly, but not necessarily in the sense of Pesetsky (1987))” (Grohmann 2006: p. 269). This characterisation extends to Singlish: unlike English, wh-topics are indeed allowed, so long as they are in compliance with some version of the DRQ.

That is to say, in exactly such cases, basic subject-question can cooccur with agreement-drop (note the impossibility of interpreting 36 as a regular, non-D-linked question; cf. 35).

Given their divergent behaviour with respect to the availability of (i) object topicalisation, (ii) adjunct-extraction, (iii) embedding under regret-class verbs, (iv) inverse scope readings, and (v) quantificational “subjects”, as well as (vi) the interpretational constraints on wh-“subjects”, the distinction between Singlish agreement-drop constructions and their fully-agreeing counterparts cannot be attributed to a PF-alternation, but must instead be treated syntactically, particularly in such a way that agreement-drop “subjects” are syntactically topics.6 In the next section, I sketch out in greater detail the syntactic configuration implicated by the empirical facts that have been presented.

The suggestion that Singlish has Topic projections akin to those found in the Sinitic languages is not in itself entirely novel, with Sinitic-style topic constructions having long been observed to exist in Singlish. In particular, Bao (2015) and Sato (2016) have both made proposals within the generative framework to this effect. However, it is important to note that the analysis being proffered in this paper is distinct in that it establishes a clear bidirectional entailment between agreement-drop and the obligatory projection of Top. This, notably, entails in turn that fully-agreeing Singlish clauses are distinct from agreement-drop constructions in that they do not obligatorily include a Top projection, pace Sato (2016).

Bao (2015) also does not make an equivalence between agreement-drop subjects and topics, simply demonstrating that Singlish has what he terms “Chinese-style topics”, which are characterised as being distinct from English-style topics in that they do not correspond to any pronominal variable in the main clause, as reflected by the accompanying glosses in (37).

On the basis of the parallels between the Singlish (38) and the Mandarin (37), Bao (2015) concludes that “(t)he topic structure of Singapore English is clearly derived from Chinese” (p. 100).7

Sato (2016) makes a similar observation, albeit beginning from the observation that there is a subject-object asymmetry that takes the form of a definiteness “preference” that applies to subjects but not to objects. This “preference” is effectively a strong bias towards the definite interpretation of the reference of preverbal subjects, as demonstrated in (39), which is not found when considering the reference of objects. This, he suggests, can be accounted for if subjects in Singlish move to the specifier of a higher topic projection, thereby creating an operator-variable chain. It is this operator-variable chain that results in the definiteness “preference”, and the subject-object asymmetry follows from the the fact that objects do not undergo any comparable movements within the derivation.

As was previously noted, Sato (2016) also treats the realisation of agreement as being orthogonal to the underlying syntactic structure, i.e. he considers agreement-drop constructions to be syntactically identical to their fully-agreeing counterparts. In his view, the basic minimal pair (3 and 4, repeated below) are syntactically equivalent; in other words, Sato (2016) adopts the view that agreement-drop is a post-syntactic alternation, i.e. PF in nature.

Note also that Sato (2016) crucially requires the subject to have moved to the higher topic position, since the operator-variable chain is critical for establishing the relevant subject-object asymmetry the analysis was designed to capture. Sato’s (2016) analysis therefore necessarily involves the operation of short subject topicalisation, which has been demonstrated by Lasnik & Saito (1992) to be unavailable in English, as illustrated by the contrast between (40) and (41). When an embedded object undergoes topicalisation, it can take an antecedent from a higher clause. It stands to reason then, that if the subject could undergo local topicalisation, it would also be able to take an antecedent from a higher clause. This is, however, not possible, leading Lasnik & Saito (1992) to conclude that subject topicalisation is disallowed in English.

This conclusion is further supported by the contrast in (42) vs. (43), where wh-extraction from the embedded topicalised subject results in a higher degree of degradation than wh-extraction from an otherwise-equivalent object topicalisation structure.

This kind of movement has been argued to be banned in many languages, such as Kinande (see, e.g. Schneider-Zioga 2000; 2007; Bošković 2016: a.o.), Kaqchikel (Erlewine 2014; 2016), as well as Florentino and Trentino (Brandi & Cordin 1989; Rizzi 1990), and is in fact suggested to be universally banned (see, e.g. Bošković 2016; Erlewine 2020).

The implementation of Sato’s (2016) analysis would, however, necessarily involve the stipulation that there is no such ban on short subject topicalisation in Singlish. Conversely, I suggest that an alternative derivation of agreement-drop “subjects” in Singlish that does not violate Lasnik & Saito’s (1992) ban on short subject topicalisation is available: consider the possibility that “subject-topics” in Singlish are not derived through movement, but are instead base-generated in the left-peripheral position in which they surface. Since they do not undergo movement to their derived position, the generation of the “subject-topic” would also not be in violation of the kind of (anti)locality conditions which have been suggested to be the source of the ban on short subject topicalisation, as argued for by Bošković (2016) and Erlewine (2020).

The canonical subject position (if it is there in the first place; see the discussion in the previous subsection), I suggest, is filled not by a trace of the moved argument (since there has been no movement), but a null pronoun instead, i.e. pro, which is semantically bound by the base-generated topic.

Even in English, topics can correspond to pronominals situated within the main clause, as seen in hanging topic constructions (see, e.g. Bianchi & Frascarelli 2010); much like the proposed structure for agreement-drop constructions, these hanging topic constructions come with a resumptive pronoun and are taken not to be derived by movement, and exhibit a parallel condition on definiteness/specificity on topics corresponding to both the clausal object (44a vs. 44b), as well as the clausal subject (45a vs. 45b).

Singlish has also been independently noted to have pro by Sato & Kim (2012), who use a generalisation put forth in Saito (2007), namely that languages which have radical pro drop also do not have overt morphological agreement, as a springboard to argue that Singlish has radical pro drop in exactly those contexts where agreement morphology is missing, ergo agreement-drop constructions (cf. 46B1 vs. 46B2, 46B3 vs. 46B4, 47B1 vs. 47B2).8

That Singlish has null pronominals was also noted in Bao (2015), who observes that the existence of empty categories in Singlish has been posited as far back as Tay (1979), and has largely been attributed to Sinitic influence on the variety.

In particular, Bao (2015) demonstrates that there is indeed parallelism between the distribution of empty categories in Singlish and that in Mandarin; null subjects are available in Singlish, with a distribution parallel to what is found in Mandarin.

I have presented a structural analysis of agreement-drop constructions that is motivated by the empirical facts established in 2, where agreement-drop constructions come with a topicalisation projection that is obligatorily merged higher in the structure, with the “subject” located in this projection rather than in the “usual” subject projection.

Such an analysis is necessitated by the empirical facts, which point to radical differences in syntactic behaviour and interpretational possibilities between agreement-drop clauses and fully-agreeing clauses in Singlish. This, by extension, points to radical differences between agreement-drop clauses and English more generally.

Accordingly, it stands to reason that English cannot have been the source of the clausal configuration that Singlish deploys in exactly these contexts. Instead, the closest analogue to, and thus the best candidate as the source for, agreement-drop constructions can be found in the Sinitic languages, seeing as the basic Sinitic clause shares with the Singlish agreement-drop clause all of the relevant properties, i.e. the absence of overt agreement morphology, the topicalisation structure, and the availability of empty categories.9

One other characteristic that sets Singlish apart from English is its large inventory of sentence-final particles, and the distribution of these particles may provide further evidence in support of a structural difference between agreement-drop and fully-agreeing constructions, in the form of an apparent complementarity between overt agreement morphology and a number of these sentence-final particles.

For example, although the particle sia is compatible with overt agreement morphology (cf. 46), the question forming particle meh is perfectly fine when appended to an agreement-drop construction (50), but results in degradation when attached to the same construction’s fully-agreeing counterpart (51).

Though the analysis of the syntactic structure of sentence-final particles in Singlish (for a nanosyntactic analysis of parallel particles in Cantonese, see Sybesma & Li 2007) is well beyond the scope of the present paper, I suggest the complementary distribution of agreement and the sentence-final question particle meh is due to the functional material where the particle is merged being available only within the clausal configuration which precludes agreement morphology. Independently, this also serves as further evidence against the PF analysis of agreement-drop.

Of note too is the fact the insertion of the particle also alleviates the incompleteness effects that are baseline in simplex agreement-drop constructions, though it must be noted that this is not an innocent rescue operation, since the particle will invariably introduce its discourse/semantic properties into the interpretation of the construction.

This alleviation can also be observed in the presence of other sentence-final particles, such as one.

The alleviating effect of particles reported here appear to suggest that the structural differences that set agreement-drop constructions apart from their fully-agreeing counterparts may be even more extensive than I have suggested, with additional structure potentially being required to achieve full acceptability for basic agreement-drop clauses. However, I should note that the incompleteness effect is likely to be pragmatic in nature. The intuition that belies the incompleteness is that upon hearing an agreement-drop construction out of the blue, there is an accompanying sense of uncertainty as to why the sentence is being put on the table in the first place.

If this is indeed the source of incompleteness, that the addition of the particles meh and one in particular should alleviate it is unsurprising, since they are known to have discourse effects (Lim 2007; Wee 2004), which can address the uncertainty felt.

Note that this characterisation of the incompleteness also suggests that it should dissipate in the presence of additional contextual cues, and this is indeed the case: note the full acceptability of the agreement-drop construction in the enriched context provided in (54).

The incompleteness effect is therefore not a structural issue per se, though the addition of more structure in the C domain in the form of overt particle use is one way of providing the necessary pragmatic cues to fully alleviate the incompleteness effects seen when an agreement-drop construction is presented in a vacuum.