English fronting constructions as a window to the semantics of tense: the case of belief reports

Cross-linguistic contrasts in the temporal interpretation of attitude reports have often been used as evidence that languages manifesting such contrasts possess distinct tense systems. This approach is primarily grounded in the expectation that tenses sharing identical semantics should demonstrate consistent behavior in similar contexts. When such consistency is lacking, it is inferred that the semantic content of present or past tenses differs across the respective languages.

A notable example of this rationale pertains to English (a representative of the so-called Sequence-of-Tense [SOT] languages) in contrast with languages such as Hebrew, Japanese, and Russian, which are classified as non-SOT languages. The interpretation assigned to English present and past tenses within complement clauses markedly differs from that of their counterparts in the aforementioned languages. Consider a case of a complement present tense in (1):

1. (1)

   John thought that Mary is hungry.^{Footnote 1}

This present-under-past report licenses only one interpretation known as double-access (Smith 1978). Roughly, according to the double-access interpretation, at a time in the past John ascribed Mary the state of being hungry, which he perceived as ongoing (John could say to himself “Mary is hungry”). In addition to this, the speaker views the temporal duration of the state of being hungry (ascribed to Mary by John) as extending up until and including the utterance time (cf. Abusch 1994; Ogihara 1995, 1996). Thus, under the double-access interpretation, the state of being hungry is understood to overlap John’s subjective “now” at the time of attitude, and also the speaker’s “now” (i.e., the utterance time).

As illustrated by (2), a similar present-under-past configuration in Russian (as well as in Hebrew and Japanese—cf. Ogihara and Sharvit 2012) allows for not only the double-access interpretation, but also the so-called simultaneous interpretation.

1. (2)

   

According to the simultaneous interpretation, John perceived Mary’s state of being hungry as ongoing at the time of his attitude, akin to the double-access interpretation. However, unlike the double-access interpretation, the simultaneous interpretation does not require an overlap with the utterance time.^{Footnote 2}

Ogihara and Sharvit (2012, p. 638), among many others, point out that the nature of the present tense in English-type languages is revealed by its inability to license the simultaneous interpretation in a present-under-past report.^{Footnote 3} Consequently, it is often believed that when English present tense is interpreted, it can only denote the utterance time (or a time that overlaps the utterance time). By contrast, the interpretation of the complement present tense in Russian, Hebrew, and Japanese is viewed as different, requiring only an overlap with the attitude holder’s “now”.

As is well known, simultaneity under a matrix past tense in English is expressed in terms of a past-under-past report like (3):

1. (3)

   John believed that Mary was hungry.

Sentence (3) can also have the back-shifted interpretation, according to which John had a retrospective attitude perceiving the state of Mary’s being hungry as something in the past, i.e., preceding his own subjective “now”. However, the simultaneous interpretation is the most natural and this fact constitutes the special nature of the past tense in English-type languages.

The most revealing manifestation of this special nature of the English past tense is provided by examples like (4),^{Footnote 4} where the most embedded past tense allows for the simultaneous interpretation, suggesting that John’s intended words are “We are having our last meal together”.

1. (4)

   A week ago, John decided that in ten days at breakfast he would tell his mother that they were having their last meal together.

Equivalents of (4) in Russian, Hebrew, or Japanese do not allow for the simultaneous interpretation. The only available interpretation is back-shifted, suggesting John’s intended words to be “We were having our last meal together”.^{Footnote 5} In these three languages, the simultaneous interpretation requires the use of the present tense in the most embedded clause.

Such observations have led many theorists to view the tense system of English-type languages as fundamentally different from the tense systems of Russian, Hebrew, and Japanese.

This paper argues to the contrary. It presents novel data from fronting constructions in English demonstrating that the simultaneous present-under-past is possible (in addition to the double-access interpretation) in some fronting configurations, but not in others. Consequently, in those English fronting constructions that allow both simultaneous and double-access interpretations, the complement present tense behaves just like its Russian counterpart in (2). This challenges the traditional view in three ways. Firstly, one must explain how the simultaneous present-under-past can in principle be possible in English fronting constructions. Secondly, one needs to explain the contrast between fronting constructions that allow for the simultaneous interpretation and non-fronting constructions like (1), which never do. Finally, contrasts within fronting constructions with respect to the availability of the simultaneous present-under-past must also be accounted for. I argue that a significant revision of our understanding of the tense system of English is necessary to accommodate the new data while still accounting for the full range of known facts. The paper proposes such a revision.

At the core of this revision lies is the null simultaneous tense. This tense is a de re alternative to the more familiar de se zero tense (Kratzer 1998). Similar to Kratzer’s de se zero tense, the null simultaneous tense licenses the simultaneous interpretation of an attitude report and exhibits SOT-effects: its morphology can be licensed by a tense feature transmitted from the matrix tense via the process of Feature Transmission at PF. However, unlike the de se tense, the null simultaneous de re tense can license “strict” readings under ellipsis and surface with the default present tense morphology in cases where Feature Transmission at PF has not occurred. Moreover, it is argued that this tense has the potential to entirely replace the more familiar version of the present tense in English. In other words, it is proposed that the tense system of English contains two tenses: the null simultaneous tense and the past tense.

This discussion of the temporal system of English lays the groundwork for a unified perspective on the tense systems of SOT and non-SOT languages. It is shown how the temporal readings of belief reports arising in non-SOT languages can be derived using the tense system developed for English. According to the proposed account, the known cross-linguistic contrasts are reduced to the differences in Feature Transmission effects and not in tense systems. Further cross-linguistic evidence suggesting variations in SOT-effects across SOT-languages is explored.

The proposed argument unfolds in the following order. Section 2 presents new evidence from VP- and CP-fronting versions of English present-under-past attitude reports, revealing a theoretically unexpected availability of the simultaneous interpretation. Additionally, it highlights unexpected contrasts in the availability of the simultaneous interpretation within fronting constructions. Section 3 discusses the challenges that the new data pose for the traditional approaches in terms of temporal de se and temporal de re. It argues against the temporal de se approach and calls for further development of the temporal de re treatment of tense. Section 4 presents the main proposal of the paper: a theory of tense consisting of the null simultaneous tense and the past tense. Section 5 extends the approach proposed for English to non-SOT languages. Section 6 discusses some further predictions of the proposed account. Section 7 explores data (known and new) suggesting that SOT-effects vary across SOT-languages. Section 8 concludes.

The literature on tense semantics has not extensively focused on fronting constructions. As anticipated in Sect. 1, these constructions reveal theoretically unexpected contrasts, posing new challenges to a semantic theory of English tense.^{Footnote 6}

2.1 Simultaneous present-under-past in VP-fronting constructions

2.1.1 Basic cases

Consider the example in (5) involving VP-fronting:

1. (5)

   Think that Mary is tired, John did. (…but Bill didn’t.) (^✓simultaneous, ^✓double-access)

Native speakers report that (5) can have the simultaneous interpretation, describing a present-oriented attitude held by John at a specific past interval. This present-oriented attitude could be expressed as “Mary is tired”. The report can be true in a situation where Mary is no longer believed to be tired or is not even alive (thus excluding the double-access interpretation). In other words, the report in (5) can be synonymous with the past-under-past simultaneous report in (6):

1. (6)

   John thought that Mary was tired.

At the same time, the non-fronting version of (5), given in (7), can only have the double-access interpretation and cannot be synonymous with (6).

1. (7)

   John thought that Mary is tired. (^✗simultaneous, ^✓double-access)

Another illustration is (8), which has the simultaneous (non-double-access) interpretation, just like the non-fronting past-under-past configuration in (9).

1. (8)

   Think that her husband is scared, Mary did at the time of the crime.

1. (9)

   Mary thought at the time of the crime that her husband was scared.

At the same time, the non-fronting version of (8) is perceived as infelicitous.

1. (10)

   #Mary thought at the time of the crime that her husband is scared.

The infelicity of the report in (10) is dependent on how long ago the crime took place. If the crime is understood to have taken place far enough in the past, making it unnatural to expect the state of being scared to persist until the utterance time, then the present-under-past report (with double-access as its only interpretation) is infelicitous. Since states of being scared are normally short-lived, the interpretation under which (10) is infelicitous is the most salient.^{Footnote 7}

More examples of VP-fronting versions of attitude reports revealing the availability of the simultaneous interpretation of present-under-past are given in (11):

1. (11)

   

The contrast between the fronting and non-fronting configurations is puzzling, given that fronted predicates are known to undergo complete syntactic reconstruction at LF (Heycock 1995, among many others). This suggests that the LF of the VP-fronting configuration in (5) should be identical to the LF of its non-fronting variant in (7). Yet, (5) somehow possesses the simultaneous reading expressed by (6), whereas (7) does not.

2.1.2 More complex cases: the availability of a “strict” interpretation under ellipsis

VP-fronting present-under-past constructions reveal another remarkable affect: they license a “strict” interpretation under ellipsis. Consider (12):

1. (12)

   

Native speakers report that the discourse in (12) can truly describe the situation in (13):

1. (13)

   A crime happened at a past time t. At t, Mary thought, “My husband is scared”. Since t, Mary has changed her mind and now she doesn’t think that her husband was scared at t.

This means that if the elided VP in (12b) is stated overtly, it will look as shown in (14):

1. (14)

   … but now she doesn’t <think that her husband was scared>

The configuration in (14) is a past-under-present configuration in which the attitude holder is understood to be past-oriented and the interpretation is, therefore, back-shifted. At the same time, (12a) is a present-under-past configuration with the simultaneous interpretation. The example in (12) is thus a case of “strict” ellipsis with respect to a given past time. That past time is referred to by the embedded present tense in (12a). The elided VP in (12b) is understood to refer to the same time.

2.2 Simultaneous present-under-past in CP-fronting constructions

In contrast to the VP-fronting constructions like (5), Think that Mary is tired, John did, native speakers report that the CP-fronting versions of such reports do not license the simultaneous interpretation. The only interpretation available for (15) is double-access.

1. (15)

   That Mary is tired, John thought. (^✗simultaneous, ^✓double-access)

CP-fronting versions of present-under-past attitude reports thus pattern with their non-fronting variants.

At the same time, native speakers report that CP-fronting constructions can license the simultaneous present-under-past when a matrix quantifier binds into the complement CP.

1. (16)

   That her₁ kids are home alone, every mother₁ thought (when the boss made everyone stay late at work). (^✓simultaneous)

1. (17)

   That her₁ students miss her₁, every professor₁ believed during the winter break. (^✓simultaneous)

The contrast between (15), on the one hand, and (16) and (17), on the other, presents a puzzle that must be addressed by a semantic theory of tense.

The data presented above pose a challenge to the two popular approaches that derive the simultaneous interpretation of past-under-past constructions in English: the temporal de se approach and the temporal de re approach.^{Footnote 8} Under the temporal de se approach (e.g., Abusch 1994, 1997; Ogihara 1996), the embedded past in a past-under-past configuration indicates the temporal center of the doxastic alternatives (also known as “the attitude holder’s ‘now’’’). According to the temporal de re approach (Abusch 1994, 1997; Sharvit 2018, 2021), the embedded past denotes the temporal res, i.e., an actual time preceding the utterance time and overlapping the time denoted by the matrix past. A de re mechanism associates the temporal res with a temporal concept perceived by the attitude holder and overlapping the attitude holder’s “now”. In what follows, I discuss the predictions of each of these two approaches.

3.1 The predictions of the temporal de se approach

3.1.1 Capturing the basic examples from VP-fronting after a minor ‘fix’

For past-under-past reports like (18), the literature offers more than one realization of the temporal de se approach.

1. (18)

   John believed that Mary was tired.

One idea is expressed by the so-called “deletion” accounts (e.g., Ogihara 1996; von Stechow 2003), according to which the English present tense always denotes the utterance time, whereas the English past tense back-shifts into the past introducing a time preceding the evaluation time of the past tense morpheme. When a complement tense is merged, its present or past feature is always interpreted. However, that feature can optionally be “deleted” when the matrix clause contains the same type of tense. Tense “deletion” occurs at LF and when it takes place, the “deleted” complement tense is understood to indicate the attitude holder’s “now”. This interpretation is akin to the interpretation predicted for the (inborn) relative present tense in non-SOT languages like Russian, Hebrew, Japanese.

Simultaneous present-under-past configurations like (5), Think that Mary is tired, John did, pose a challenge for “deletion” accounts: at LF, the fronted VP reconstructs to its base position, tense “deletion” is not licensed since the matrix and embedded tenses are not the same. Consequently, the simultaneous interpretation is not predicted.

Another way to model the temporal de se is the so-called zero tense approach (Kratzer 1998). According to this approach, English contains a so-called zero tense morpheme Ø, which has the pronominal semantics suggested in (19).

1. (19)

   \([\!\![\emptyset _{\mathrm{i}}]\!\!]^{\mathrm{g}}\) = g(i)

Moreover, Ømust be locally bound at LF and lacks the features that determine its morphology (being a minimal pronoun over times). The zero tense borrows tense features from the matrix tense at PF as part of a special process called Feature Transmission (outlined below).

To illustrate the predictions of the zero tense approach in more detail, I adopt the double-indexed pronominal semantics for the past tense (Ogihara and Sharvit 2012, among others) provided in (20) and a temporal de se lexical entry for believe in (21). In (20), the lower index i indicates the topic time denoted by the past tense morpheme, whereas the higher index j is the anchor index standing for the time from which the past tense back-shifts, “<” stands for “temporally precedes”.

1. (20)

   \([\!\![\mathrm{Past}^{\mathrm{j}}_{\mathrm{i}}]\!\!]^{\mathrm{g}}\) = g(i), defined only if g(i) < g(j)

1. (21)

   

The LF predicted for the SOT past-under-past configuration in (18), John believed that Mary was tired, is given in (22) and the truth conditions in (23):

1. (22)

   

1. (23)

   〚(22)〛^g,c(c(time))(w) = 1 iff ∀<w’,t’> ∈ Dox(John)(w)(g(2)): Mary be tired in w’ at t’, defined only if g(2) < c(time)^{Footnote 9}

In (22), λ3 is “responsible” for the attitude holder’s “now” and binds the only index on the de se zero tense; the λ-abstract over possible worlds, λ4, binds the index on the world variable of the predicate. In (23), and throughout this paper, c(time) stands for the utterance time, meaning the time parameter of the context.

Having shown how the zero tense approach predicts the simultaneous truth conditions for past-under-past reports like (18), I will now outline a Kratzer-style process of transmission of tense features building on the corresponding discussion in Cable (2015). The Feature Transmission process consists of the following rules:

1. (24)

   

It is understood that the structure in (22) is seen at PF too.^{Footnote 10} The process of transmission of tense features from the matrix past tense onto the embedded null tense proceeds as illustrated in (25).

1. (25)

Consequently, the embedded SOT past tense in (18) is not a real past tense. It is rather a “fake” past because at LF the past tense morphology is underlain by a zero de se simultaneous tense.

Certain complications can arise when applying this approach to sentences like (26) and (27), where the main verb believe appears in a nonfinite form. This suggests that Feature Transmission has not occurred, yet an SOT interpretation of the embedded tense is still observed (cf. Bjorkman 2015).

1. (26)

   John didn’t believe that Mary was tired.

1. (27)

   John will believe that Mary is tired.

To extend the Feature Transmission mechanism to such cases, von Stechow (2009, p. 17) proposes a theory of tense in which the time argument of the verb is syntactically represented as PRO, undergoes QR creating a λ-abstract and leaving a trace, and “disappears” in accordance with Chomsky’s principle of Full Interpretation (cf. Grønn and von Stechow 2010, p. 130). Instead of (25), report (18) receives the structure in (28). Features transmitted to the syntactically represented time argument of a verb and are understood to percolate to the whole VP consisting of the verb and its time argument (i.e., from \(t_{3}\) to \(VP^{2}\) in (28)). To this structure, the rules of Feature Transmission suggested in (24) can apply as shown:

1. (28)

In such a framework, tense features are transmitted to nonfinite verbs as well, where they remain invisible, but can be transmitted further.^{Footnote 11}

In what follows, I follow Grønn and von Stechow (2010) and assume that nonfinite verb forms also bear tense features. I will not overcomplicate the syntax and use Tense Lowering to account for this. Feature Transmission in sentences like (26) will be understood to proceed as follows:

1. (29)

The case of (27) is somewhat different because, unlike the negation in (26), the modal will in (27) not only intervenes between the tense and the main verb, but also itself bears tense morphology. To account for this, I adopt Heim’s (1994, pp. 151-152) view that will is composed of the present tense and the tenseless verb stem woll (an idea supported by the will/would alternation). I further adopt Cable’s (2015, p. 17) treatment of woll as undergoing head-raising to T⁰, (re-)merging with tense and thus acquiring the tense feature. I call the process of putting tense feature on woll “Feature Sharing”. To avoid unnecessary overcomplication, I present structures that do not explicitly reflect this movement of woll. Feature tTansmission in sentences like (27) receive the following analysis:

1. (30)

It seems that the zero tense approach to modeling the temporal de se fares better than “deletion” accounts when faced with simultaneous VP-fronting present-under-past reports like (5), Think that Mary is tired, John did, because the zero tense approach relies on Feature Transmission at PF. A relatively simple “fix” can allow this approach to predict cases like (5) as instances of the temporal de se. The “fix” in question says that whenever a zero de se tense is outside the scope of the matrix tense at PF (as is the case in (5)), it surfaces with a “default” present tense morphology. The simultaneous present-under-past report in (5) would then be an instance of a de se configuration in which the embedded de se zero tense surfaces with such default present tense morphology.

Despite this welcome prediction, the zero tense approach faces problems with further data from Sect. 2, not being able to account for the “strict” ellipsis example, incorrectly predicting the availability of the simultaneous interpretation in all CP-fronting constructions, and, consequently, failing to account for the contrast between fronting constructions that license the simultaneous present-under-past and those that do not.

3.1.2 Inability of the temporal de se approach to capture the “strict” ellipsis interpretation

Recall the example from (12), repeated in (31), which licenses the “strict” interpretation of (31b):

1. (31)

   

The “strict” interpretation requires that (31b) be understood as suggested in (32):

1. (32)

   … but now she doesn’t <think that her husband was scared>

In other words, (31b) must be able to receive the back-shifted interpretation: the complement clause in (31b) (=(32)) describes the same time as the time described by the complement clause in (31a), but, unlike (31a), the attitude holder’s perspective towards that time is retrospective, because the matrix clause in (31b) contains the present tense. In other words, the “strict” ellipsis interpretation in (31) involves the sameness of the time described by the complement tense in the antecedent and in the ellipsis site together with a shift in the attitude holder’s perspective as we move from (31a) to (31b).

The de se zero tense approach cannot predict this “strict” interpretation of (31b). As was shown above, under the de se zero tense approach, the antecedent in (31a), which is the simultaneous present-under-past construction, is analyzed as having the VP with the LF in (33) and interpretation in (34):

1. (33)

   De se LF for the VP in (31a):

   [_VP2 think_{de se}-w₁ [_CP λ3 λ4 that [_TP \(\emptyset _{3}\) [_VP1 her husband-w₄ be scared-w₄]]]]

1. (34)

   

This interpretation is inevitably simultaneous. Under the common assumption that an elided VP must be semantically identical to the antecedent VP (Keenan 1971; Merchant 2019, among many others), the same interpretation must also be assigned to the elided VP in (31b). This would lead to interpreting (31b) as (35), which is a “sloppy” interpretation:

1. (35)

   … but now she doesn’t <think that her husband is scared>

The “sloppy” interpretation would suggest that Mary doesn’t think that her husband is scared now, and not that she doesn’t think that he was scared at the time of the crime. This is not the desired interpretation.

3.1.3 Can the temporal de se approach be “saved” by an appeal to a mismatch in features under ellipsis?

It is well-known that in focus alternatives and in an ellipsis site, φ-features on pronouns and definite DPs can be ignored (Partee 1989; Rooth 1992; Fiengo and May 1994; Rullmann 2004; Heim 2008; Sauerland 2013; Bassi 2021):

1. (36)

   a. I did my homework, and  b. you did <do your homework> too.

In light of examples like (31), a question arises whether the required difference in the morphology of the complement tense visible in (31a) and expected in (31b) is another example of the same effect and, if so, whether the predictions of the de se theory could be somehow made compatible with the observed facts. In other words, could (31a) receive a de se interpretation, and could the interpretation arising in (31b) be an instance of a switch in tense features licensed under ellipsis?

I claim that the answer is negative: the de se theory cannot be “saved” in such a way. Accounts licensing such mismatches in features still require that semantic identity under ellipsis be satisfied. For examples like (36) Bassi (2021, p. 102) writes:

Ellipsis requires identity of meaning (semantic value) with an antecedent […] at the level of the elided constituent […]. Crucially, […] the DP/pronouns with offending features do not constrain the semantic value proper of the structure, but [are] added late. Because of that, the ellipsis clause and the antecedent clause respect identity at the level relevant for ellipsis licensing: the elided VP, for example in [(36)], is semantically identical to its antecedent.

This suggests that if the antecedent sentence in a “sloppy” ellipsis configuration like (36) is understood to contain a structure interpreted as [λx. x did x’s homework], then so is the sentence containing the elided constituent. In other words, given that my in (36a) is understood to be bound by the subject I, the pronoun your in (36b) cannot be viewed as deictic and coreferential with the subject you; your can only be viewed as bound by you.^{Footnote 12}

It seems that φ-features on pronouns can be ignored only under “sloppy” ellipsis. When the ellipsis is “strict”, the features on pronouns in the elided constituent must be the same as on the corresponding pronoun in the antecedent. What makes cases involving tense special is the fact that a mismatch in features can arise not only under a “sloppy”, but also under a “strict” ellipsis.

An example of the former case is given by Alxatib and Sharvit (2017) in their discussion of tense in relative clauses:

1. (37)

   

According to Alxatib and Sharvit (2017), the embedded present tense in (37a) has a “sloppy” (“bound”) construal. The same bound construal is reproduced in (37b). Had the elided part in (37b) been pronounced, it would have to involve different morphology (namely, past), as suggested in the part inside the angle brackets.

Example (31) is a case, where the ellipsis is “strict”: one and the same time is being referred to in both complement clauses. The difference in the morphology expected in the ellipsis site arises from the difference in the perspective taken towards that concrete time in each sentence. In (31a), that time is perceived as ongoing and the perspective is simultaneous, whereas in (31b), that time is perceived as preceding and the perspective is back-shifted. This transition from the simultaneous to the back-shifted perspective with respect to the same time would be associated with different morphology on the complement tense in the antecedent and in the elided constituent (compare (31a) and (32)). Thus, (31) is an example of a mismatch in tense features under “strict” ellipsis.

Being in principle compatible with mismatches in tense features under ellipsis, the theory of temporal de se must still preserve the simultaneous interpretation. By contrast, in order to explain (31), it is required not only to account for the mismatch in tense features, but also for the change in the temporal orientation (i.e., a switch from the simultaneous to the back-shifted perspective). The latter requirement cannot be satisfied by the temporal de se approach. Consequently, for cases like (31), the theory of temporal de se under-generates.

3.1.4 Inability of the temporal de se approach to predict the contrasts in the temporal interpretation of fronting constructions

The de se zero tense approach does not predict any contrast between VP- and CP-fronting. In both cases, the embedded de se zero tense identifies the attitude holder’s “now” and is outside the scope of the matrix tense. Consequently, allowing a de se zero tense to surface with the “default” present tense morphology (when Feature Transmission at PF has not applied) incorrectly predicts the availability of the simultaneous present in CP-fronting constructions like (15), That Mary is tired, John thought, which only allows for the double-access interpretation. Therefore, in addition to under-generating (as discussed in the previous subsection), the temporal de se theory also over-generates.^{Footnote 13} Finally, this theory cannot explain why the simultaneous present-under-past unavailable in (15) becomes available when the fronted CP contains an element bound by a matrix quantifier (as illustrated in (16) and (17)).

3.2 The predictions of the temporal de re approach

As was already mentioned in the beginning of this section, for past-under-past reports like (3), John believed that Mary was tired, temporal de re approaches provide an analysis along the following lines: the embedded past tense receives an indexical interpretation denoting a time preceding the utterance time. That time is then associated with a time concept perceived by the attitude holder. Two additional constraints, known as the Upper Limit Constraint (ULC) and Intensional Isomorphism (II), further apply to the temporal de re interpretation of an attitude report in order to rule out undesired predictions.

The ULC roughly stipulates that and the time denoted by an embedded tense cannot be later than the time of the attitude. As demonstrated in Abusch (1994), the ULC prohibits past-under-past reports like (39) to have the so-called forward-shifted interpretation and truthfully describe scenarios like (38). For such scenarios, only future-under-past configurations like (40) are acceptable.

1. (38)

   Scenario. Thunderstorm. Today is Friday. There was a thunderstorm on Thursday (i.e., yesterday). On Wednesday (i.e., two days ago), Mary already knew that the thunderstorm was expected and she entertained the following thought, “John will be afraid during the thunderstorm”.

1. (39)

   Mary believed that John was afraid during the thunderstorm.

1. (40)

   Mary believed that John would be afraid during the thunderstorm.

II roughly states that the temporal alignment between the temporal res and the attitude time is mirrored by the temporal alignment between the time concept perceived by the attitude holder and their subjective “now” (cf. Abusch 1994, pp. 123-124; Heim 1994, p. 158, fn. 28).^{Footnote 14} II prohibits a sentence like (41) from being true in a scenario where John has a retrospective attitude towards Mary’s state of being tired (thinking “Mary was tired”), but the speaker believes that the state continues up until and including the utterance time and uses the present-under-past construction to report this belief.

1. (41)

   John thought that Mary is tired.

According to temporal de re approaches, the English complement present tense is always indexical under a matrix past tense. Given that, there are two options in principle for interpreting the present tense in a present-under-past configuration like (41). One option is to interpret the present tense as indicating directly the utterance time. This option is ruled out by the ULC, because the temporal res indicates a time that follows the attitude time. The second option is to interpret the present tense as denoting a time that overlaps the utterance time and extends far enough into the past to overlap the attitude time. In this case, the ULC is not violated. However, now that there is an overlap between the temporal res and the attitude time, II requires the attitude holder to perceive the time concept (which is the counterpart of the temporal res in each doxastic alternative) as overlapping the “now” of each alternative. Consequently, a sentence like (41) cannot be true in a scenario where the attitude holder has a retrospective orientation towards the time concept.

Example (31) poses a challenge for temporal de re theories, because they predict that present-under-past configurations can only receive a double-access interpretation. So, they cannot predict the simultaneous interpretation for (31a) to begin with. Consequently, they cannot predict the contrast between VP- and CP-fronting constructions as well as the impact of binding on the availability of the simultaneous present-under-past in CP-fronting constructions.

However, in contrast to temporal de se accounts, the temporal de re approach has the potential of predicting “strict” ellipsis in (31). If only the embedded present tense in (31a) could receive a temporal de re interpretation according to which it denotes a contextually salient past time, the complement tense in the ellipsis site in (31b) would also be able to denote the same past time as is required by the “strict” interpretation.

The next section proposes a development the temporal de re approach into a theory that allows the present-under-past in (31a) to have the simultaneous interpretation, captures the “strict” interpretation of (31b) and accounts for all the other contrasts.

In what follows, I adopt a neo-Reichenbachian approach to tense and view tenses as expressions indicating topic (or reference) times (Klein 1994). Under this approach, topic times are related to eventuality times by viewpoint aspect. Consequently, I view tenses as expressions composing with aspectual phrases (AspPs). However, for the sake of simplicity, I omit AspPs in the LFs and use VP instead of AspP throughout. I also adopt the pronominal theory of tense (in the spirit of Partee 1973 and Heim 1994). Tenses denote contextually given times that are presupposed to precede or overlap (depending on the specific tense) a time anchor (also known as the evaluation time). Following von Stechow (2009) and Ogihara and Sharvit (2012), I assume that tenses have two indices: the upper index responsible for the time anchor and the lower index responsible for the denoted topic time. Further assumptions and proposals are introduced in the following sections as the argumentation develops.

4.1 Preliminaries: Sharvit’s (2018) theory of time-concept generators

I adopt Sharvit’s (2018) theory of time-concept generators as an in situ approach to the de re interpretation of complement tenses. A time-concept generator is defined as follows (see Sharvit 2018, p. 221, ex. (11)):

1. (42)

   

According to Sharvit, a suitable time concept is a concept that the attitude holder can use to describe a time to themselves.

The definition of a time-concept generator in (42) encodes the ULC and II. (42biii) requires that when a time-concept generator applies to the temporal res at world and time of attitude, the output be the temporal res itself. Assuming that a time-concept always encodes temporal ordering with respect to the evaluation time, a time-concept generator applying to the temporal res at the world and time of attitude will return the temporal res only if the temporal res is ordered with respect to the time of attitude in the same way as the time perceived by the attitude holder is ordered with respect to the attitude holder’s “now”.^{Footnote 15} In light of this, it can be said that II is hard-wired into the notion of a time-concept generator. (42biv) pertains to the ULC. It prohibits the time concept from picking out a time in the future of the attitude holder’s “now”. By II, this also implies that the time denoted by a complement de re tense cannot follow the time of the attitude.

As an illustration, consider the past-under-past report in (43) and the predicted LF in (44):

1. (43)

   John believed that Mary was tired.

1. (44)

   LF involving time-concept generators:

   [λ0 λ1 \(\mathrm{Past}^{0}_{2}\) John believe_CG-w₁ [_CP1 λ4 λ5 λ6 [_TP [[[G₄ \(\mathrm{Past}^{0}_{3}\)] t₅] w₆] [_VP Mary be tired-w₆]]]]

In this LF, the embedded past tense receives an in situ de re interpretation. It is not anchored to the attitude holder’s “now” (associated with λ5). Instead, the embedded past tense is anchored to the utterance time (as indicated by the anchor time index 0 that it bears; this index is bound by the abstract λ0, which stands for an argument position to be saturated post-syntactically). Relative to a contextually given assignment function g, this embedded past tense denotes a past time g(3), which is presupposed to precede the utterance time.

The complement CP is an expression of type <<i,<i,si>>,<i,st>> denoting a function from time-concept generators to relations between times and worlds. The attitude verb is understood to take the CP as argument. The lexical entry predicted for \(\mathit{believe}_{CG}\) under this approach is as shown in (45). The truth conditions for (43) under the LF in (44) are given in (46).

1. (45)

   〚believe_CG〛^g,c = [λw. λP_{<<i,<i,si>>,<i,st>>}. λx. λt. ∃G(G is a time-concept generator suitable for x in w at t and ∀<w’,t’> ∈ Dox(x)(w)(t): P(G)(t’)(w’) = 1)]

1. (46)

   〚(44)〛^g,c (c(time))(w) = 1 iff ∃G(G is a time-concept generator suitable for John in w at g(2) and ∀<w’,t’> ∈ Dox(John)(w)(g(2)): Mary be tired in w’ at G(g(3))(t’)(w’), defined iff g(2) < c(time) and g(3) < c(time)

According to (46), the interpretation function (relativized to assignment function g and context c) is defined for (43) under the LF in (44) in world w (the actual world) at the utterance time (c(time)), only if g(2) precedes the utterance time and g(3) precedes the utterance time. If defined, (43) is true in w at c(time), if and only if there is a time-concept generator G suitable for John in w at g(2), i.e., at the time of believing, and in all worlds w’ with the temporal centers t’ that are compatible with what John believes in w at g(2), Mary is tired in w’ at G(g(3))(t’)(w’), where G(g(3))(t’)(w’) is the time-concept generated by the application of G to g(3) and evaluated w.r.t. t’ and w’ (i.e., the intensional counterpart of g(3) in John’s doxastic alternatives).

In the next subsection, I propose a theory of locally anchored de re tenses, which is different from the theory of tense used for the embedded past tense in (44). However, the new theory is fully compatible with Sharvit’s theory of time-concept generators outlined above.

4.2 The null simultaneous tense and the past tense: examples from matrix clauses

I propose that English tense system consists of a past tense and a null simultaneous tense with the lexical entries in (47) and (48), where ‘⊗’ stands for ‘overlaps’. I further introduce the requirement in (49):

1. (47)

   \([\!\![\mathrm{Past}^{\mathrm{j}}_{\mathrm{i}}]\!\!]^{\mathrm{g}}\) = g(i), defined only if g(i) < g(j)

1. (48)

   \([\!\![\emptyset ^{\mathrm{j}}_{\mathrm{i}}]\!\!]^{\mathrm{g}}\) = g(i), defined only if g(i) ⊗ g(j)

1. (49)

   Obligatory anchoring requirement:^{Footnote 16}

   The anchor time index of a tense must be bound by a λ-binder located in the left periphery of the clause that tense occurs in.

The null simultaneous tense in (48) also lacks its own morphology borrowing it from a c-commanding tense via Feature Transmission at PF (like Kratzer’s de se zero tense) or surfacing with a default present tense morphology whenever Feature Transmission at PF has not occurred. Consequently, under the current proposal, what is known as “present tense” is the default realization of null simultaneous tense (bearing either default or transmitted morphology).

For a matrix sentence like (50), the LF in (51) and truth conditions in (52) are predicted.

1. (50)

   Mary was tired.

1. (51)

   LF: [λ0 \(\mathrm{Past}^{0}_{1}\) Mary be tired]

1. (52)

   〚(51)〛^g,c(c(time)) = 1 iff Mary was tired at g(1), defined iff g(1) < c(time)

In (51), λ0 binds the anchor-index 0 on \(\mathit{Past}^{0}_{1}\) suggesting that the denotation of this tense, namely g(1), is presupposed to precede the time that saturates anchor-argument associated with λ0. For matrix clauses, this time is always the utterance time.

For a matrix sentence like (53), the LF in (54) and truth conditions in (55) can be predicted.

1. (53)

   Mary is tired.

1. (54)

   LF: [λ0 \(\emptyset^{0}_{1}\) Mary be tired]

1. (55)

   〚(51)〛^g,c(c(time)) = 1 iff Mary is tired at g(1), defined iff g(1) ⊗ c(time)

In (54), λ0 binds the anchor-index 0 on \(\emptyset^{0}_{1}\) suggesting that the denotation of this tense, namely g(1), is presupposed to overlap the time that saturates anchor-argument associated with λ0. For matrix clauses, this time is always the utterance time. The morphology on the null simultaneous tense is “default” since no feature transmission is possible.

4.3 The temporal interpretation of complement clauses

Recall the past-under-past configuration in (43), repeated here as (56).

1. (56)

   John believed that Mary was tired.

The complement CP for a sentence like (56) can be associated with either one of the following two structures:

1. (57)

   LF: [_CP2 λ7 [_CP1 λ4 λ5 λ6 [_TP [[[G₄ \(\emptyset^{7}_{3}\)] t₅] w₆] [_VP Mary be tired-w₆]]]]

1. (58)

   LF: [_CP2 λ7 [_CP1 λ4 λ5 λ6 [_TP [[[G₄ \(\mathrm{Past}^{7}_{3}\)] t₅] w₆] [_VP Mary be tired-w₆]]]]

Here, the structure of CP¹ is just like the temporal de re structure of CP¹ in the Sharvit-style analysis given in (44). The difference is that here the binder of the time-anchor of the embedded tense, i.e., λ7, is added in the left periphery, making the embedded tense anchored in accordance with the Obligatory Anchoring Requirement in (49). The whole complement CP is now represented as CP² having a more complex semantic type <i,<<i,<i,si>>,<i,st>>>.

Importantly, the time anchor argument (represented by λ7) is not associated with the attitude holder’s “now”. In these LFs, it is λ5, which is “responsible” for the attitude holder’s “now”, just like in Sharvit-style theory discussed above. Furthermore, I assume a prohibition of vacuous binding, which means that λ7 in (57) and (58) must bind an index.^{Footnote 17}

These CPs receive the following interpretation:

1. (59)

   〚(57)〛^g,c = [λt: t ⊗ g(3). λG. λt’. λw’. Mary be tired in w’ at G(g(3))(t’)(w’)]

1. (60)

   〚(58)〛^g,c = [λt: t > g(3). λG. λt’. λw’. Mary be tired in w’ at G(g(3))(t’)(w’)]

As can be seen from (59) and (60), the anchor argument of each complement tense remains unsaturated at this stage of the derivation. Observe also that the morphology with which the null simultaneous tense in (57) will be pronounced is not yet determined. If Feature Transmission at PF applies, then the null simultaneous tense will be realized with the tense feature that will be transmitted (present or past). If no Feature Transmission at PF applies, then the null simultaneous tense will be realized with the default present tense morphology. No such question arises in (58): the complement tense bears the past tense morphology.

What time can be the anchor of an embedded tense? Ultimately, I show that it can be the attitude time as well as the utterance time. At this stage, however, I concentrate on the former option.

4.4 Composition of a complement CP with the attitude verb

I adopt the lexical entry for \(\mathit{believe}_{CG}\) provided in (45) and repeated here as (61):

1. (61)

   〚believe_CG〛^g,c = [λw. λP_{<<i,<i,si>>,<i,st>>}. λx. λt. ∃G(G is a time-concept generator suitable for x in w at t and ∀<w’,t’> ∈ Dox(x)(w)(t): P(G)(t’)(w’) = 1)]

Observe that \(\mathit{believe}_{CG}\) (after it has taken its world argument) cannot compose with a CP, like (57) or (58), by Function Application, because such a CP now contains an extra time argument and is an expression of type <i,<<i,<i,si>>,<i,st>>>. A type mismatch arises.

To compose such CPs with \(\mathit{believe}_{CG}\), I introduce a new rule of semantic composition that is based on Jacobson’s (1999) type shifter z:^{Footnote 18}

1. (62)

   Rule Z:

   If α is a branching node and β, γ are its daughters such that 〚β〛 ∈ D_<δ,τ> and 〚γ〛 ∈ D_{<τ,<υ, <δ, σ>>>}, where τ, υ, δ, σ are semantic types, then 〚α〛 ∈ D_{<υ, <δ,σ>>} and 〚α〛 = [λp_υ. λq_δ. 〚γ〛(〚β〛(q))(p)(q)].

This rule allows \(\mathit{believe}_{CG} -w_{i}\) and a complement CP, like (57), to compose in such a way that the anchor time argument of the function denoted by the CP is saturated by time argument of the attitude verb. Only then 〚believe_CG-w_i〛^g,c takes that function as its argument:

1. (63)

   

The whole attitude report (56), John believed Mary was tired, under the simultaneous interpretation gets the LF in (64) and the truth conditions in (65).

1. (64)

   LF: [λ0 λ1 \(\mathrm{Past}^{0}_{2}\) John believe_CG-w₁ [_CP2 λ7 [_CP1 λ4 λ5 λ6 [_TP [[[G₄ \(\emptyset^{7}_{3}\)] t₅] w₆] [_VP Mary be tired-w₆]]]]]

1. (65)

   

In prose: true at the utterance time in w (relative to assignment function g and context c), iff there is a time-concept generator G suitable for John in w at the time of believing g(2) such that in each of John’s doxastic alternatives w’ with temporal center t’, Mary is tired at the time to which the concept G(g(3)) maps t’ and w’ (i.e., at the intensional counterpart of g(3) in John’s doxastic alternatives).

Since (56) does not involve reconstruction, I follow Kratzer (1998) and Heim (2008) and assume that the structure seen at PF is not significantly different from the LF in (64) in the sense that all indices and their binders remain visible. Consequently, the LF in (64) creates a suitable configuration for feature transmission at PF:

1. (66)

The embedded null simultaneous tense thus surfaces with the temporal morphology it “borrows” from the matrix past tense.

Correspondingly, the back-shifted interpretation of sentence (56), John believed Mary was tired, arises when the embedded tense is interpreted as past. The corresponding LF and truth conditions are given below:

1. (67)

   LF: [λ0 λ1 \(\mathrm{Past}^{0}_{2}\) John believe_CG-w₁ [_CP2 λ7 [_CP1 λ4 λ5 λ6 [_TP [[[G₄ \(\mathrm{Past}^{7}_{3}\)] t₅] w₆] [_VP Mary be tired-w₆]]]]]

1. (68)

   

To conclude this subsection, it should be said that an appeal to a rule like Rule Z might appear undesirable due to concerns that such a rule could render the overall system too powerful, leading to undesired consequences elsewhere. In this paper, such concerns are not investigated.^{Footnote 19}

4.5 Capturing the simultaneous interpretation in VP-fronting present-under-past reports

4.5.1 Deriving the simultaneous truth conditions

The proposed treatment can now be applied to explain the new data from VP-fronting constructions.

Under the simultaneous interpretation, the present-under-past report from (12a), repeated here as (69), receives the same LF as the past-under-past non-fronting configuration in (70):

1. (69)

   Think that her husband is scared, Mary did.

1. (70)

   Mary thought that her husband was scared.

Under the assumption that fronted VPs reconstruct at LF, both reports are predicted to have the same LF given in (71). The simultaneous truth conditions are also the same and are given in (72):

1. (71)

   LF: [λ0 λ1 \(\mathrm{Past}^{0}_{2}\) Mary [_VP2 think_CG-w₁ [_CP2 λ7 [_CP1 λ4 λ5 λ6 [_TP [[[G₄ \(\emptyset^{7}_{3}\)] t₅] w₆] [_VP1 her husband be scared-w₆]]]]]]

1. (72)

   

In prose: the interpretation function (relativized to assignment function g and context c) is defined for the LF in (71) at the utterance time in world w (the actual world), only if g(2) precedes the utterance time and g(3) precedes the utterance time. If defined, (43) is true in w at the utterance time, if and only if there is a time-concept generator G suitable for Mary in w at g(2), i.e., at the time of thinking, and in all worlds w’ with the temporal centers t’ that are compatible with what Mary thinks in w at g(2), her husband is tired in w’ at G(g(3))(t’)(w’), where G(g(3))(t’)(w’) is the time-concept generated by the application of G to g(3) and evaluated w.r.t. t’ and w’ (i.e., the intensional counterpart of g(3) in Mary’s doxastic alternatives).

The only difference between the two reports is the morphology on the null simultaneous de re tense. In (69), Feature Transmission at PF has not applied. This is so, because at PF the VP think that her husband is scared is in the fronted position and outside the scope of the matrix tense. Consequently, the null simultaneous tense can only surface with the default present tense morphology.^{Footnote 20} In (70), the morphology on the null tense is past, because the embedded tense is in the scope of the matrix past at PF and feature transmission applies along the lines suggested in (66).

This establishes a method by which (69) and (70) can share an LF despite the difference in the morphology of the embedded tense.

4.5.2 Capturing the “strict” ellipsis interpretation

The account proposed above for (69), repeated here as (73), can now be used to capture the “strict” ellipsis interpretation of the continuation in (74):

1. (73)

   Think that her husband is scared, Mary did.

1. (74)

   Now she doesn’t <think that her husband was scared>.

Assuming that the whole report in (73) is analyzed in the way proposed in the previous subsection, the interpretation of the fronted VP is as shown in (75):

1. (75)

   

The “strict” identity interpretation requires that the denotation of the bracketed VP in (74) be as shown in (76).

1. (76)

   

Both VPs have the same assertive content (in accordance with the requirement of semantic identity). Their difference lies in the presupposition associated with g(3). This difference in presuppositions arises from the difference in the times of each attitude and, consequently, from the difference in the temporal perspective on the same temporal res. Here, it is assumed that this difference is not visible under ellipsis because the corresponding presuppositional component is not part of the meaning shared by the antecedent and the ellipsis site (in line with Fiengo and May’s 1994 mechanism of Vehicle Change, Heim’s 2008 theory minimal pronouns, and Bassi’s 2021 valuation from context approach).

The truth conditions predicted for the whole report in (74) are provided in (77):

1. (77)

   

In prose: whenever defined, this report is true, iff there is no time-concept generator G suitable for Mary in w (the actual world) at g(2), i.e., at the time of thinking, and in every world w’ with the temporal center t’ that is compatible with what Mary thinks in w at g(2), her husband is tired in w’ at G(g(3))(t’)(w’), where G(g(3))(t’)(w’) is the time concept generated by the application of G to g(3) and evaluated with respect to t’ and w’ (i.e., the intensional counterpart of g(3) in Mary’s doxastic alternatives).

These truth-conditions are back-shifted. According to them, the temporal res, g(3), precedes the time of thinking, g(2), and, correspondingly, the intensional counterpart of g(3) precedes the attitude holder’s “now”.

4.6 Interim summary 1

The preceding subsections introduced the first part of the proposed theory of tense. Tenses contain two indices (an index for the anchor time and an index for the topic time). The anchor time index of the main tense of a clause is bound by a λ-abstract in the left periphery of the clause. Consequently, each clause is a predicate of time anchors. The tense system of English consists of the null simultaneous tense and the past tense. The null simultaneous tense can be pronounced with what is known as the present or the past tense morphology depending on whether Feature Transmission at PF has applied. The null simultaneous tense presupposes that the topic time overlaps its anchor, whereas the past tense presupposes that the topic time precedes the anchor.

For matrix clauses, the time anchor is the utterance time provided post-syntactically when the clause is uttered. For complement clauses, it was shown that the anchor can be the attitude time (i.e., the time argument of the attitude verb). The anchoring of a complement tense to the time argument of the matrix verb becomes possible due to Rule Z, a mode of semantic composition between the attitude verb and its CP complement. In such cases, a complement null simultaneous tense denotes a time overlapping the attitude time and a complement past tense denotes a time preceding the attitude time.

The interpretation of a complement tense anchored to the attitude time remains fully extensional: its anchor is an “actual” time and its topic time is provided by the assignment function. Consequently, such a tense requires a temporal de re analysis. A theory of time-concept generators is adopted.

It was proposed that present-under-past VP-fronting constructions allowing for the simultaneous interpretation are instances of a complement null simultaneous tense anchored to the attitude time and surfacing with the default present tense morphology because feature transmission at PF cannot apply in fronting constructions. In a non-fronting version of the same construction, Feature Transmission at PF applies and the null simultaneous tense surfaces with the past tense morphology resulting in what appears as the simultaneous past-under-past report. Given that fronted VPs obligatorily reconstruct at LF, it is predicted that what appears to be a fronting simultaneous present-under-past attitude report and what appears to be a non-fronting simultaneous past-under-past report can be two different realizations of one and the same LF.

4.7 Further predictions of the analysis proposed so far

4.7.1 Capturing the “sloppy” ellipsis interpretation in terms of a bound null simultaneous de re tense

Up until this point, the null simultaneous de re tense has been discussed as referential denoting a concrete time provided by the contextually sensitive assignment function. However, besides the “strict” ellipsis interpretation, examples like (78) also allow for a “sloppy” interpretation. Under the “sloppy” interpretation, (78b) is understood as suggested in (79). Under the “sloppy” interpretation, Mary is most naturally understood to have in mind two different states of her husband being tired. This subsection demonstrates how the “sloppy” reading can be generated in terms of the proposed approach.

1. (78)

   

1. (79)

   … but now she doesn’t <think that her husband is tired>.

In (80), a special case of the null simultaneous tense is identified in which the anchor and the topic time indices are the same:

1. (80)

   \([\!\![\emptyset ^{\mathrm{i}}_{\mathrm{i}}]\!\!]\)^g,c = g(i)

The “sloppy” reading of (78b) arises when (78a) is understood to have the LF in (81).

1. (81)

   LF: [λ0 λ1 \(\mathrm{Past}^{0}_{2}\) Mary [_VP2 think_CG-w₁ [_CP2 λ3 [_CP1 λ4 λ5 λ6 [_TP [[[G₄ \(\emptyset^{3}_{3}\)] t₅] w₆] [_VP1 her husband be tired-w₆]]]]]]

VP² receives the interpretation in (82), where the time of thinking and the temporal res are the same:

1. (82)

   

The elided VP in (78b) is understood to have the same interpretation. The truth conditions predicted for (78b) are simultaneous:

1. (83)

   

According to (83), Mary does not now think that her husband is scared.

4.7.2 Some further applications of a bound null simultaneous de re tense

A bound null simultaneous de re tense can be used to predict the simultaneous interpretation of present-under-future cases like (84).

1. (84)

   John will think that Mary is pregnant.

Here, I adopt Heim’s (1994) lexical entry for woll given in (85):

1. (85)

   〚woll〛 = [λP_<i,t>. λt. ∃t’(t’ > t and P(t’) = 1)]

Assuming that future times are not sufficiently determined and are, therefore, not given to us directly (e.g., Abusch 1994, pp. 109-110), we can only talk about them in terms of quantification.^{Footnote 21} For this reason, the future attitude time, as well as the res counterpart of the time at which Mary is believed to be pregnant, cannot be provided by an assignment function. The embedded tense in (84) cannot be referential.^{Footnote 22}

Nevertheless, it is still possible to provide (84) with a de re LF and truth conditions in terms of a special case of the null simultaneous tense (proposed in (80)) as illustrated below:

1. (86)

   LF: [λ0 λ1 [_TP2 \(\emptyset^{0}_{2}\) [_VP4 woll [_VP3 John [_VP2 think_CG-w₁ [_CP2 λ3 [_CP1 λ4 λ5 λ6 [_TP1 [[[G₄ \(\emptyset^{3}_{3}\)] t₅] w₆] [_VP1 Mary be pregnant-w₆]]]]]]]]]

1. (87)

   

So far, we have only seen cases where the present tense morphology is default. However, it can also be transmitted. In (84), the matrix null simultaneous tense surfaces with the default present tense morphology. This “default” morphology is then transmitted onto the embedded null simultaneous de re tense.^{Footnote 23}

1. (88)

A VP-fronting version of (84) given in (89) also receives the LF in (86) (because the fronted VP reconstructs) and truth conditions in (87). The only difference is that the embedded present tense morphology in (89) is “default”.

1. (89)

   Think that Mary is pregnant, John will (when he sees her next week at the costume party).

The same approach to the null simultaneous tense can be used to explain the following version of Breakfast example in (90), which has been simplified for expository purposes:

1. (90)

   (A week ago,) John thought (that) (in ten days at breakfast) he would say (that) they were having their last meal (together).

This example is often believed to show that a temporal de se approach to past-under-past attitude reports is indispensable. What the current discussion suggests is that it is possible to capture such examples in terms of a bound null simultaneous de re tense. The following LF and truth conditions are predicted for (90):

1. (91)

   LF: [λ0 λ1 [_TP2 \(\mathrm{Past}^{0}_{2}\) [_VP5 [_VP John [_VP think_CG-w₁ [_CP2 λ3 [_CP1 λ4 λ5 λ6 [_TP1 [[[G₄ \(\emptyset^{3}_{3}\)] t₅] w₆] [_VP woll [_VP he [_VP say-w₆ [_CP2 λ7 [_CP1 λ8 λ9 λ10 [_TP1 [[[G₈ \(\emptyset^{7}_{7}\)] t₉] w₁₀] [_VP1 they be having their last meal-w₁₀ ]]]]]]]]]]]]]]]

1. (92)

   〚(91)〛^g,c(w)(c(time)) = 1 iff ∃G’(G’ is a time-concept generator suitable for John in w at g(2) and ∀<w’,t’> ∈ Dox(John)(w)(g(2)): ∃t”(t” > G’(g(2))(t’)(w’) and ∃G”(G” is a time-concept generator suitable for John^{Footnote 24} in w’ at t” and ∀<w’’’,t’’’> ∈ Say(John)(w’)(t”): they are having their last meal in w’’’ at G(t’)(t’’’)(w’’’), defined iff g(2) < c(time).

Feature Transmission at PF is licensed and all occurrences of the null simultaneous tense surface with the transmitted past tense morphology as shown below:

1. (93)

4.8 CP-fronting present-under-past reports: capturing the double-access interpretation

Recall that CP-fronting present-under-past attitude reports like (94) only allow for the double-access interpretation.

1. (94)

   That Mary is tired, John thought. (^✗simultaneous, ^✓double-access)

It has long been known that CP-fronting constructions obviate condition C effects, unlike their non-fronting versions:^{Footnote 25}

1. (95)

   

Such data led theorists to believe that fronted CPs do not reconstruct at LF (cf. Takahashi 2010).

Building on Takahashi (2010), I treat fronted CPs as located inside a DP-shell headed by a covert definite determiner the. I further propose that the anchors its complement CP to the utterance time giving it an independent (indexical) interpretation. The lexical entry for the is given in (96):

1. (96)

   〚the〛^g,c = [λP_{<i,<<i,<i,si>>,<i,st>>>}. P(c(time))]

Thus, in a CP-fronting construction like (94) a complement tense is always understood to be anchored to the utterance time. The LF in (97) is predicted for (94).

1. (97)

   LF: [λ0 λ1 [_TP2 [_DP the [_CP2 λ7 [_CP1 λ4 λ5 λ6 [_TP1 [[[G₄ \(\emptyset^{7}_{3}\)] t₅] w₆] [_VP Mary be tired-w₆]]]]] [λ8 \(\mathrm{Past}^{0}_{2}\) John think_CG-w₁ t₈]]]

In (97), the fronted CP/DP is an expression of type <<i,<i,si>>,<i,st>> (the time anchor argument of the CP is saturated). This expression leaves a trace of the same type, \(t_{8}\) in (97), which composes with the attitude verb by Function Application.

Because the time anchor of the embedded null simultaneous tense is saturated by c(time), the embedded null simultaneous de re tense denotes a time overlapping the utterance time, which leads to the double-access interpretation.

1. (98)

   〚(97)〛^g,c(c(time))(w) = 1 iff ∃G(G is a time-concept generator suitable for John in w at g(2) and ∀<w’,t’> ∈ Dox(John)(w)(g(2)): Mary be tired in w’ at G(g(3))(t’)(w’)), defined iff g(2) < c(time) and g(3) ⊗ c(time)

Feature transmission at PF is not predicted, because in this fronting construction the matrix tense does even not c-command the embedded tense at PF. The embedded null simultaneous tense surfaces with the default present tense morphology.

4.9 Other cases of indexical complement tense under a matrix past tense

The approach proposed in the previous subsection for double-access present-under-past CP-fronting constructions can now be used to predict the double-access intereprtation of present-under-past in non-fronting reports.

For non-fronting constuctions, a DP-shell for the complement CP remains optionally available. In other words, an in situ CP complement can remain without a DP-shell (composing with the attitude verb by Rule Z), or it can be inside a DP-shell (composing with the attitude verb by Function Application). In the former case, the embedded tense is anchored to the attitude time. In the latter case, the embedded tense is anchored to the utterance time. Consider (99), which is a non-fronting present-under-past construction:

1. (99)

   John thought that Mary is tired.

The fact that the embedded tense surfaces with the present tense morphology shows that feature transmission at PF has not applied. Consequently, it is not possible that the embedded clause contains a null simultaneous tense anchored to the attitude time, because in that case the complement tense would have surfaced with the transmitted past tense morphology. The LF predicted for (99) is given in (100). This structure is also seen at PF. Feature transmission is not predicted, because the blocks the use of Predication:

1. (100)

It is thus predicted that present-under-past non-fronting reports (just like their CP-fronting versions) can only have the double-access interpretation (with the truth conditions in (98)).

Up to now, VP-fronting constructions like (5), repeated in (101), have been mostly discussed as licensing the simultaneous interpretation.

1. (101)

   Think that Mary is tired, John did.

However, their complement CP can also occur inside a DP-shell. In that case, (101) is predicted to have the double-access interepretation (with the truth conditions in (98)). Because in (101) the fronted VP reconstructs, the predicted LF is as shown in (100). At PF, however, the VP remains in the fronted position. Feature transmission is not predicted. The embedded null simultaneous tense surfaces with the default present tense morphology.

Finally, the more familiar cases of de re past-under-past in which the embedded past is anchored to the utterance time are also predicted whenever the complement CP containing the past tense is inside a DP-shell.

4.10 Reconstruction for binding licenses the simultaneous present-under-past in CP-fronting constructions

As was illustrated earlier with examples like (102), the simultaneous present-under-past is licensed in a CP-fronting construction containing a pronoun bound by a matrix quantifier:

1. (102)

   That her₁ kids are home alone, every mother₁ thought (when the boss made everyone stay late at work). (^✓simultaneous, ^✓double-access)

CP-fronting constructions in which the fronted CP contains a pronoun bound by a matrix quantifier have been argued to undergo syntactic reconstruction at LF (e.g., Takahashi 2010, pp. 351-352). Consequently, in the LF of such reports the CP appears in its base position. In base position, the complement CP containing a null simultaneous tense need not be encapsulated inside a DP shell and can compose with the attitude verb by Rule Z, leading to the simultaneous interpretation. At the same time, since Feature Transmission at PF is not licensed in (102), the embedded null simultaneous tense surfaces with the default present tense morphology.

The option to interpret the embedded tense relative to the utterance time is also available and can be carried out as described above.

4.11 Interim summary 2

It was suggested that when a CP occurs inside a DP-shell, it receives an independent temporal interpretation: the complement tense becomes anchored to the utterance time.

For CP-fronting constructions like (94), it is required that the complement CP occur inside a DP-shell. Consequently, reports like (94) only have the double-access interpretation. The morphology on the embedded null simultaneous tense is present by default, because Feature Transmission at PF is not predicted.

Building futher on existing proposals that fronted CPs undergo reconstruction for binding and are in their base position at LF, it was claimed that such reconstructing CPs need not be incapsulated inside a DP shell and that they can license the simultaneous interpretation of the embedded null simultaneous tense in addition to the double-access interpretation. In either case, the null simultaneous tense is still realized with the default present tense morphology.

Finally, this approach was used to account for the fact that double-access is the only interpretation available for non-fronting present-under-past constructions like (99). In non-fronting constructions, the embedded null simultaneous tense is preceded by the matrix past tense at PF. The fact that the embedded tense is pronounced with the present tense morphology means that the underlying configuration is incompatible with Feature Transmission at PF (otherwise the null simultaneous tense would have surfaced with the transmitted past tense morphology). This is only possible if the embedded null simultaneous tense is inside a DP-shell and is anchored to the utterance time, which necessarily leads to the double-access interpretation.

In this section, I demonstrate how the tense system developed for English can be applied to capture the interpretation of the main configurations of attitude reports in such non-SOT languages as Russian, Japanese, and Hebrew. It will be shown that we can view the tense systems of those languages as also consisting of the past and the null simultaneous tense. All the main contrasts between SOT and non-SOT languages will be reduced to the presence of Feature Transmission in the former and its absence in the latter. The key differences will emerge in the realization of the embedded null simultaneous tense. In non-SOT languages, where Feature Transmission is lacking, this tense is predicted to consistently appear with the default present tense morphology.

5.1 Predictions for complement tenses anchored to the utterance time

It was proposed above that in English a complement tense can always be anchored to the utterance time. This happens when the complement CP is inside a DP-shell. In such cases, feature transmission is not predicted, regardless of whether the construction is fronting or non-fronting. In such configurations, the null simultaneous tense always appears with the default present tense morphology and receives the “indexical present” de re interpretation. The embedded past tense receives the “indexical past” de re interpretation. This framework can be extended to Russian, Japanese, and Hebrew.

In these languages, a complement null simultaneous tense anchored to the utterance time is also predicted to give rise to the simultaneous interpretation under a matrix present, and the double-access interpretation under a matrix past. Both interpretations are attested in these languages.

When a complement past tense is anchored to the utterance time and occurs under a matrix null simultaneous tense, the result is the familiar past-under-present configuration with the back-shifted interpretation. This is the only interpretation that has been reported for such configurations across all of these languages. When such an indexical complement past tense occurs under a matrix past, the prediction is that it should give rise to the simultaneous as well to the back-shifted interpretation. This prediction is also borne out. Russian past-under-past constructions have been reported to allow not only for the back-shifted, but also the simultaneous, interpretation (cf. Khomitsevich 2007, pp. 2-3; Altshuler 2008, p. 21). Ogihara and Sharvit (2012, p. 640) report similar facts for Hebrew past-under-past reports.

Japanese is the only language that is usually reported to license only the back-shifted interpretation for past-under-past construction.^{Footnote 26} However, as was observed already in Kusumoto (1999), the presence of sonotoki ‘then’ in the embedded clause makes the simultaneous interpretation of a past-under-past report possible:

1. (103)

   

Kusumoto (1999, p. 218) explains:

The sentences [in (103)] are ambiguous, however: when there is a contextually salient past interval that the speaker has in mind, sonotoki ‘then’ is understood deictically. The sentence has the shifted reading. Sonotoki can, however, also be anaphoric to the time of the matrix events. In this case, the sentence has the simultaneous reading.

Tsilia (2021, p. 32, fn. 26) also reports the availability of the simultaneous interpretation of past-under-past reports in Japanese even in the absence of sonotoki.^{Footnote 27,Footnote 28}

I conclude that the interpreations predicted for the English complement tense anchored to the utterance time are also available in Russian, Japanese, and Hebrew.

5.2 Predictions for complement tenses anchored to the attitude time

In English, a past tense anchored to the attitude time always gives rise to the back-shifted interpretation. A null simultaneous tense with such anchoring yields the simultaneous interpretation and always appears with transmitted features.

In Russian, Japanese, and Hebrew complement clauses, a past tense anchored to the attitude time is predicted to produce the back-shifted interpretation, regardless of the type of tense in the matrix clause. This prediction holds for all three languages. A locally anchored null simultaneous tense is predicted to give rise to the simultaneous interpretation and to appear with default present tense morphology, as Feature Transmission is absent in non-SOT languages. This prediction is also confirmed: present-under-past constructions in these languages give rise to the simultaneous interpretation.

5.3 All relevant contrasts are attributed to the presense or absence of feature transmission

The contrasts between the SOT and non-SOT languages that are traditionally discussed in terms of the existence of “deleted” tense in the former and “inborn relative present” in the latter are now explained under the assumption that the tense systems in both types of languages are fundamentally the same. The observed contasts are reduced to the presence or absence of feature transmission in a given language. In all these languages, using a complement clause with a null simultaneous tense anchored to the attitude time is the most natural way to convey the simultaneous interpretation. However, in English-type languages Feature Transmission determines the morphology of this tense. In non-SOT languages like Japanese, Russian, and Hebrew, where Feature Transmission does not apply, a locally anchored simultaneous tense consistently appears with default present tense morphology.^{Footnote 29}

6.1 No “fake” past in fronting constructions

The contrast between English VP-fronting attitude reports, which reveal the existence of the simultaneous present-under-past, and their non-fronting versions, which do not, was discussed above as evidence in favor of a Kratzer-style approach, according to which Feature Transmission in English occurs at PF. Consequently, this approach predicts that a SOT past (i.e., transmitted past morphology on the simultaneous tense) is not licensed in a fronting version of an attitude report. Moreover, the approach predicts that an occurrence of past tense morphology inside a fronted constituent can only be interpreted as “real” (relative or absolute) past (unless the fronted constituent also contains a “real” c-commanding past). This prediction appears to be supported by examples like (104).

1. (104)

   That they were having their last meal together, John decided that he would tell his mother at breakfast.

Five native speakers who I consulted report that in (104), the back-shifted interpretation is either the only one possible or, at least, very strongly preferred.^{Footnote 30}

6.2 New evidence in defense of the ULC

The observations made in the previous subsection can be used to shed more light on one puzzle related to the nature of the ULC.

It was observed by Klecha (2014, 2016) that verbs like hope and pray allow the embedded clause to describe an event in the future of the attitude holder’s “now” and, therefore, seemingly violate the ULC (see also Abusch 1994, p. 137, fn. 13; Schlenker 2004).

1. (105)

   Martina hoped Carissa got pregnant. (Klecha 2014)

Sentences like (105) present a seeming violation of the ULC, since hope here is an attitude verb and the embedded clause describes a situation in the future of the attitude holder’s “now”. Examples like (105) have been discussed as counterexamples to the ULC.

The data explored in this paper could be used to defend the ULC as a constraint on de re tense even for verbs like hope. Consider the CP- and VP-fronting versions of (105) in (106):

1. (106)

   

Native speakers report that, unlike (105), the attitude reports in (106) do not allow for the forward-shifted interpretation. They only allow for the back-shifted interpretation.

As was suggested in the previous subsection, the past tense morphology in a fronted constituent is expected to be “real”. In light of this, a forward-shifted interpretation in (106a, b) could in principle be expected only if the embedded past were anchored to the utterance time. However, this interpretation is not observed, suggesting that the ULC is satisfied by de re past-under-past constructions even with verbs like hope. Consequently, example (105) cannot be viewed as a counterexample to ULC-based theories of temporal de re.^{Footnote 31}

6.3 Present-under-past and CP-fronting in a non-SOT language

CP-fronting constructions in a non-SOT language like Russian easily license simultaneous present-under-past:

1. (107)

   

This might seem puzzling if English and Russian have similar tense systems: English CP-fronting analogues do not license the simultaneous interpretation.

According to the current proposal, this should mean that fronted CPs in Russian necessarily reconstruct at LF (in contrast to fronted CPs in English, which reconstruct only when this is required by such factors as binding). This prediction seems to be borne out (supported by data regarding condition C-effects in Russian CP-fronting constructions). Recall from Safir’s (1999) example, repeated in (108), that English CP-fronting configurations obviate Condition C effects:

1. (108)

   

The Russian equivalents are both equally bad:

1. (109)

   

If Russian fronted CPs obligatorily reconstruct, then the current proposal predicts that they would license the simultaneous interpretation of present-under-past.

More generally, the Russian CP-reconstruction data (and their contrast with English) illustrate cross-linguistic variation in the behavior of fronted CPs. The observation that Russian always requires CP-reconstruction and English blocks it (unless there are independent reasons like binding) suggests the possibility of there being languages where CP reconstruction is optional. The possibility of an optional CP-reconstruction in a non-SOT language would make such a language compatible with the availability of the simultaneous present-under-past and obviation of condition C effects in CP-fronting constructions.

In this final section, I would like to discuss some known, as well as new, cross-linguistic evidence suggesting that the transmission of tense features is a more complex phenomenon than is typically assumed. The goal of this section is to provide additional support for the idea that cross-linguistic contrasts in the temporal interpretation of belief reports can arise from differences in Feature Transmission, rather than from variations in tense systems.

Up until now, English has been viewed as a language with an obligatory Feature Transmission mechanism. However, further evidence from the literature suggests that the picture is more complicated. Wurmbrand (2014) reports that the present-under-future-in-the-past configuration in (110) can also have the simultaneous interpretation. Sharvit (2014) reports similar facts about Abusch’s (1994) Breakfast example, illustrated in (111):^{Footnote 32}

1. (110)

   Last week, the weatherman hoped that he would announce on Christmas Eve that it is snowing.

1. (111)

   Last week John decided that in ten days at breakfast he would tell his mother that they are having their last meal together.

In terms of the perspective proposed in this paper, these data suggest that if would intervenes between a matrix past tense and the most embedded null simultaneous tense, Feature Transmission becomes optional even in English, and the most embedded null simultaneous tense can surface with the default present tense morphology. Consequently, English appears not to be a language with obligatory Feature Transmission throughout: while English attitude verbs seem to be “holes” for feature transmission, the future auxiliary functions as a “filter” optionally blocking feature transmission.

In languages like Modern Greek, attitude and modal verbs behave as optional transmitters of tense features. Tsilia (2021) reports that Modern Greek demonstrates exactly this sort of behavior: it licenses the simultaneous interpretation in present-under-past attitude reports (just like Japanese, Herbew, and Russian, and unlike English) and it also permits the simultaneous interpretation for the most embedded past in an analogue of the Breakfast example (just like English, and unlike Japanese, Hebrew, and Russian):

1. (112)

   

1. (113)

   

Next is German, which was reported to licence the SOT past-under-past in attitude reports in addition to the simultaneous present-under-past (von Stechow 1999, pp. 97-98):

1. (114)

   

At the same time, my three German consultants report that the simultaneous (SOT) past is impossible in the German version of the Breakfast example:^{Footnote 33}

1. (115)

   

A version of the same sentence with the present tense in the most embedded clause (dass er [gerade] sein letztes Treffen mit ihr hat) easily licenses the simultaneous interpretation.

At first, it might seem that German is, after all, not an SOT language because it reveals the same behavior as Russian in allowing the simultaneous past-under-past in reports like (114) and dissallowing it in cases like (115). However, the following data suggest that the situation is more complicated and that German and Russian are not identical.

German patterns with English in allowing the “sloppy” ellipsis reading for the b-variant suggesting that John at 10 thought that it was +15 at his “now”, not at the time that Mary had in mind.

1. (116)

   At five, Mary thought it was +15 outside, and (b) so did John at 10.

1. (117)

   

Assuming that the sloppy simultaneous past-under-past arises in the presence of a bound null simultaneous tense, the above data suggest that German past-under-past reports license feature transmission onto the embedded null tense.

Russian on the other hand reveals a different behavior.^{Footnote 34} In (118), the past-under-past report (118a) can be understood to have the simultaneous interpretation. However, in (118b), unlike German and English, a “sloppy” simultaneous interpretation is not licensed. This suggests that the simultaneous past-under-past in Russian is expressible only in terms of a temporal de re configuration.^{Footnote 35}

1. (118)

   

The sloppy simultaneous interpretation can arise in (118b), only when simultaneity under past in (118a) is expressed in terms of a present-under-past construction as illustrated in (119):

1. (119)

   

These data provide evidence to view Russian as a genuine non-SOT language, whereas German appears to be a language where attitude verbs optionally allow for Feature Transmission, and modals like würde completely block it. These observations suggest the following typology:

The preliminary typology in Table 1 contributes to the evidence in favor of a unified view of the semantics of (complement) tense across languages and in favor of explaining the observed contrasts (especially with respect to the configurations that license the simultaneous reading) in terms of restrictions on Feature Transmission.^{Footnote 36}

With respect to the temporal interpretation of attitude reports, Ogihara (1996, p. 148) already noted that “a pure de se analysis can be recast in terms of a de re analysis”. In this paper, I argued that such a reformulation is necessary. The new data from English fronting constructions were used to support a more unified semantic perspective on tense. However, the conclusions reached in this paper, even if on the right track, can be considered preliminary. More data must be examined. The proposed analysis has not been extended to the temporal interpretation of adjunct clauses, which are known for various cross-linguistic puzzles. Future research will hopefully provide more clarity on these issues.