complement tense in contrast: the sot parameter in russian and english

In an sot-language like English, “past under past” may have a simultaneous interpretation, i.e., we have temporal agreement. In a non-sot language like Russian, we get the shifted interpretation. In English, the temporal morphology of the embedded verb can be determined by the matrix tense via a binding chain through verbal quantifiers such as ‘say’ or ‘think’. In Russian, these attitude verbs break the binding chain, hence the morphology of the embedded verb is determined locally by an embedded relative present or past. The main tense of an attitude complement will be a semantically empty PRO. We propose that the difference between English and Russian is derived from the sot-parameter, saying that a language L is an sot-language if and only if the verbal quantifiers of L transmit temporal features. The paper gives a precise formulation of the syntax and semantics of the constructions involved. The essential idea is that temporal features are transmitted via semantic binding; following the sot-parameter, verbal quantifiers may or may not act as barriers for feature transmission. The paper takes up a recent challenge by Daniel Altshuler and Olga Khomitsevich against existing accounts: verbs of perception and, occasionally, attitude verbs and factive verbs in Russian may express simultaneity by “past under past”. For attitude verbs, we show that the problem is in fact nonexistent when the complement is imperfective. Concerning factives, we argue that the complement tense is an independent de re past. Finally, perception verbs are normally not verbal quantifiers and hence not subject to the sot-parameter. [1] introduct ion to complement tense The role of tense is ubiquitous in natural language, yet many phenomena pertaining to subordinate tense remain to be properly investigated and understood. In this paper we will provide a semantically motivated explanation – the sot parameter – which is intended to capture the difference between tense agreement and non-agreement languages. In other words, we explain the distribution of complement tense in sequence-of-tense languages (notably Germanic and Romance languages) and non-sequence-of-tense languages like Russian. The proposal restores [110] grønn & von stechow the traditional viewof Russian as a non-sot language despite the challenging data, notably from factives and perception verbs, recently discussed by Altshuler and Khomitsevich. In future work, we intend to extend the theory to capture adjunct tense, as well. In this section, we will introduce the sot parameter, our general tense architecture, and some key notions central to the theory – dependent (vs. independent) tense, verbal quantifiers, and zero (vs. fake) tense. In section [2], we will introduce the basic patterns found in Russian and English complement tense. Our methodology is based on retrieval of authentic data from parallel corpora. Next, in section [3], we give a formal implementation of our theory. Finally, in section [4], we show that Russian “past under past” with imperfective complements and/or matrix factives and perception verbs are compatible with our theory and the proposed sot parameter. [1.1] The SOT parameter The contrast in (1) illustrates the difference between sequence of tense found in Germanic languages, and non-sequence of tense characteristic of Russian1. (1) R On skazalPAST-PF, čto živetPRES pod Moskvoj. (Viktor Pelevin, “Pokolenie P”) E He said he was living just outside Moscow. N Han fortaltePAST at han boddePAST utenfor Moskva. Tenses express relations between times: backward shift (precedence), simultaneity (overlap), or forward shift (succession). Russian apparently represents the most natural form-meaning mapping. The embedded tense in (1R) is a (relative) present tense – expressing simultaneity with the higher past tense of the verbum dicendi. In our feature system, the present tense morphology is licensed by a relative present operator in complements of attitude verbs in Russian (Ogihara 1989), (Kusumoto 1999), (Schlenker 1999) and (von Stechow 2003). Since the feature transmission is done locally in the complement in non-sot languages, Russian morpho-syntax neatly matches the semantics in this domain. (We speak of the feature “present” originating with the semantic present and transmitted to the finite verb; the mechanism will be explained in detail below, notably in section [3]). In sequence-of-tense languages like English and Norwegian, morphology is not in a one-to-one relation with semantics due to the phenomenon of temporal agreement: the past tense morphology (“was” (Eng.), “bodde” (Nor.)) in the complement in (1E)/(1N) is semantically void and simply agrees with the past tense operator in the matrix. This fact requires syncategorematic rules (here: long[1] E = English, F = French, G = German, N = Norwegian, R = Russian. OSLa volume 2(1), 2010 complement tense in contrast [111] distance transmission of temporal features following the sot parameter, i.e., nonlocal binding). The difference between English and Russian is tentatively illustrated in Figure 1. past He said ∅-tense he was living outside Moscow (English)

complement tense in contrast: the sot parameter in russian and english Oslo and Tübingen/Konstanz a b s t r a c t In an sot-language like English, "past under past" may have a simultaneous interpretation, i.e., we have temporal agreement.In a non-sot language like Russian, we get the shifted interpretation.In English, the temporal morphology of the embedded verb can be determined by the matrix tense via a binding chain through verbal quantifiers such as 'say' or 'think'.In Russian, these attitude verbs break the binding chain, hence the morphology of the embedded verb is determined locally by an embedded relative present or past.The main tense of an attitude complement will be a semantically empty PRO.We propose that the difference between English and Russian is derived from the sot-parameter, saying that a language L is an sot-language if and only if the verbal quantifiers of L transmit temporal features.
The paper gives a precise formulation of the syntax and semantics of the constructions involved.The essential idea is that temporal features are transmitted via semantic binding; following the sot-parameter, verbal quantifiers may or may not act as barriers for feature transmission.
The paper takes up a recent challenge by Daniel Altshuler and Olga Khomitsevich against existing accounts: verbs of perception and, occasionally, attitude verbs and factive verbs in Russian may express simultaneity by "past under past".For attitude verbs, we show that the problem is in fact nonexistent when the complement is imperfective.Concerning factives, we argue that the complement tense is an independent de re past.Finally, perception verbs are normally not verbal quantifiers and hence not subject to the sot-parameter.
[1] i n t r o d u c t i o n t o c o m p l e m e n t t e n s e The role of tense is ubiquitous in natural language, yet many phenomena pertaining to subordinate tense remain to be properly investigated and understood.In this paper we will provide a semantically motivated explanation -the sot parameter -which is intended to capture the difference between tense agreement and non-agreement languages.In other words, we explain the distribution of complement tense in sequence-of-tense languages (notably Germanic and Romance languages) and non-sequence-of-tense languages like Russian.The proposal restores [2] grønn & von stechow the traditional view of Russian as a non-sot language despite the challenging data, notably from factives and perception verbs, recently discussed by Altshuler and Khomitsevich.In future work, we intend to extend the theory to capture adjunct tense, as well.
In this section, we will introduce the sot parameter, our general tense architecture, and some key notions central to the theory -dependent (vs.independent) tense, verbal quantifiers, and zero (vs.fake) tense.In section [2], we will introduce the basic patterns found in Russian and English complement tense.Our methodology is based on retrieval of authentic data from parallel corpora.Next, in section [3], we give a formal implementation of our theory.Finally, in section [4], we show that Russian "past under past" with imperfective complements and/or matrix factives and perception verbs are compatible with our theory and the proposed sot parameter.
[1.1]The SOT parameter The contrast in (1) illustrates the difference between sequence of tense found in Germanic languages, and non-sequence of tense characteristic of Russian 1 .
(Viktor Pelevin, "Pokolenie P") E He said he was living just outside Moscow.N Han fortalte PAST at han bodde PAST utenfor Moskva.
Tenses express relations between times: backward shift (precedence), simultaneity (overlap), or forward shift (succession).Russian apparently represents the most natural form-meaning mapping.The embedded tense in (1R) is a (relative) present tense -expressing simultaneity with the higher past tense of the verbum dicendi.In our feature system, the present tense morphology is licensed by a relative present operator in complements of attitude verbs in Russian (Ogihara 1989), (Kusumoto 1999), (Schlenker 1999) and(von Stechow 2003).Since the feature transmission is done locally in the complement in non-sot languages, Russian morpho-syntax neatly matches the semantics in this domain.(We speak of the feature "present" originating with the semantic present and transmitted to the finite verb; the mechanism will be explained in detail below, notably in section [3]).
In sequence-of-tense languages like English and Norwegian, morphology is not in a one-to-one relation with semantics due to the phenomenon of temporal agreement: the past tense morphology ("was" (Eng.), "bodde" (Nor.)) in the complement in (1E)/(1N) is semantically void and simply agrees with the past tense operator in the matrix.This fact requires syncategorematic rules (here: long- complement tense in contrast [3] distance transmission of temporal features following the sot parameter, i.e., nonlocal binding).
The difference between English and Russian is tentatively illustrated in Figure 1.
past He said ∅-tense he was living outside Moscow (English) We adopt a feature theory whose conceptual attractiveness comes from the explicit encoding of every detail in the morpho-syntax and the corresponding dependence on semantic operators.The system gives us a precise formulation of the syntax and semantics of the constructions involved.The basic machinery is the following (more details are given below in section [3]): Every verb has a temporal argument.The temporal argument is satisfied by a variable coming from the tense morphology with an uninterpretable feature.The interpretation of this variable is determined by a binding relation with a semantic operator having a corresponding interpretable feature.With complements in sot languages we get agreement throughout the c-command domain, with a semantic operator checking several instances of the same uninterpretable feature (a kind of Multiple Agree).In non-sot languages, the morphology is licensed locally by a corresponding semantic operator.Khomitsevich (2007) adopts a framework with feature checking as well but seeks primarily a syntactic explanation of subordinate tense data.Following the work of Abusch (1994b), Schlenker (1999), von Stechow (2003) among others, we focus on the semantics of the matrix verb in our explanation of the data.
The difference between English and Russian is derived from the sot parameter: a language l is an sot language if and only if the verbal quantifiers of l transmit temporal features.(The sot parameter) In English, the temporal morphology of the embedded verb is determined by the matrix tense via a binding chain through verbal quantifiers such as "say".In Russian, "skazat'" ("say") breaks the binding chain.The morphology of the embedded verb is determined by an embedded relative present or past. [4]

grønn & von stechow
The sot parameter gives the right predictions and explains the different distribution of subordinate tenses in sot and non-sot languages such as the expression of simultaneity in (1) as "past under past" in sot and "present under past" in non-sot languages.
[1.2] Architecture of tenses For an easier understanding of our paper, the following remarks on our view of the architecture of tenses may be helpful.Tenses have the structure schematically presented in Figure 2. Partee (1973) proposed to treat tenses in general as pronouns -deictic or anaphoric.In our theory, a semantic tense consists of an obligatory temporal pronoun (T-center, perspective time, local evaluation time) and an optional shifter.The pronoun n(ow) is deictic and refers to the speech time.Tpro is an anaphoric pronoun bound by a higher tense.TPRO is a semantically empty pronoun and must be moved for type reasons.This movement creates a λ-operator binding the temporal variable in the T-center.TPRO is what has been called zero tense (λ-tense) in the literature, cf.(Kratzer 1998), (von Stechow 1995).
The shifters are existential quantifiers based on the temporal center and temporal relations such as "before", "after" and "identical" and possibly others.p(ast) means "there is a time before the temporal center T", f(uture) means "there is a time after the temporal center T" and pres Rus means "there is a time identical with the temporal center T".We assume that the temporal center is the same in all languages.We further think that the quantifying force of the shifters is always existential, but the relations may vary across languages: for instance, English has no pres Rus , German and Japanese use the relation "not before" for the present; there might be temporal overlap, temporal inclusion, "no after" and others.
If the center is n, we have a deictic tense; this is the tense found in the matrix clause.If the center is Tpro, we have an anaphoric tense bound by some higher tense; this kind of tense is found in relative clauses and other adjuncts.If the center is TPRO, we have a bound tense, i.e., a temporal variable locally bound by a λ-operator.complement tense in contrast [5] [1.3] Dependent vs. independent tense The sot parameter applies to dependencies between matrix tense and subordinate tense.Obviously, tense in subordinate clauses often has a different status from matrix tense, since the former can be dependent on the latter.In deictic tenses, the pronominal T-center refers to the now of the context, but in dependent tenses, the pronominal center may be anaphoric, as in relative clauses, or λ-bound as in complements.
The distinction between independent (deictic) and dependent tense is crucial in our survey.When we consider complex data from subordinate tense, we should keep in mind the possibility of the lower complement tense being independent of the higher matrix.In such cases, the morphology of the subordinate tense is licensed by the speaker's deictic now in the utterance situation or by a de re past, and the sot parameter does not directly apply to the construction.
The difference between dependent and independent tense is illustrated below: (2) N Var PAST det ikke de gamle romerne som sa PAST at alt er PRES i evig forandring?(independent) (Nikolaj Frobenius, "Latours katalog") E Wasn't it the ancient Romans who said that everything was in a process of eternal flux?(dependent) F Les vieux Grecs ne disaient-PAST-IMPF -ils pas que tout est-PRES en perpétuel changement?(independent) R Razve drevnie rimljane ne govorili PAST-IMPF , čto vse nepreryvno izmenjaetsja PRES ?(dependent or independent) The embedded present in the Norwegian original and French translation requires an independent interpretation 2 : the complement reports a universal truth which was true at the time of the ancient Romans and remains true for the author at the moment of writing -a kind of "double access" interpretation.Arguably, we have an independent de re extended now in the complement (Abusch 1997).For sot languages like Norwegian and French, a dependent interpretation in (5N)/(5F) would have required tense agreement, hence past tense morphology both in the matrix and complement.This tense agreement is precisely what we find in the English translation.
Concerning the Russian translation, we cannot distinguish between a dependent and independent ("double access") reading.The surface syntax in (5R) matches the Norwegian original, but since Russian is not a tense agreement language, the present tense in the Russian complement remains indistinguishable between a relative present (a simultaneous interpretation dependent on the matrix) and an independent present.
[2] Germanic and Romance languages do not have a relative present of the Russian kind, see section [3.4].
The paradox is illustrated in the following sentence: (3) At 5 o'clock Mary thought it was 6 o'clock.
The sentence is coherent and makes perfect sense although Mary is obviously wrong about the time in the context of (3).However, if the complement is analysed in the Hintikka-style as a proposition (set of worlds) we end up attributing the absurd belief to Mary that "5 o'clock = 6 o'clock".The solution is to let attitudes quantify over worlds and times (and individuals, neglected here).Hence the complements must be properties of times, whose highest semantic tense is a zero tense (a temporal abstract), cf.(Kratzer 1998).Our semantics for attitude predicates follows the insights of Lewis (1979): ) is compatible with everything y believes of (w, t) in w at time t → P (w 1 )(t 1 )] In order to understand complement tense under attitudes, it must be stressed that the semantics of attitudes gives us an abstraction over the highest temporal variable in the complement.This subtle fact is a permanent source of confusion, and we will return to this point below.
We will not try to give an exhaustive list of verbal quantifiers in natural language, 4 but we will make one demarcation in the next subsection.
[1.5] Zero tense vs. fake tense A zero tense embedded under an attitude verb is semantically void, it does not have any meaning.However, some caution is called for.Consider the following sentence, with an embedded counterfactual: (5) E John believes PRES that Mary would PAST-MODAL come INF if she was PAST invited PART .G John glaubt PRES dass Marie käme PAST-SUBJ , wenn sie eingeladen PART würde PAST-SUBJ R Vanja verit PRES , čto Maša prišla PAST-PF by SUBJ.PART , esli by SUBJ.PART ee priglasili PAST-PF .
[3] i.e., "there is a time in the future/past …".Formally, these verbs embed a lambda abstract over times.[4] The future auxiliaries "will/would" (Eng.) and "budet" (Rus.) as well as their mirror operator -the perfect auxiliary "have" (Eng.)-are quantifiers over times, hence verbal quantifiers in our understanding.
complement tense in contrast [7] Although the embedded verbs have past morphology, it is obvious that this is not a real past tense, since the embedded counterfactual is not evaluated at a time preceding the subjective now but at a time succeeding the subjective now.The embedded tenses cannot be in a relation of temporal agreement with the matrix tense (attitude verb), since the latter is in the present tense.So ( 5) is a case of what Iatridou (2000) calls fake past: past tense morphology does not point to anteriority.The verb form does not have its normal temporal feature, but a feature pointing to a subjunctive interpretation.This is what we see more clearly in German (4G) and Russian (4R), where the fake past is translated by forms of subjunctive past (Konjunktiv 2), a mood used in counterfactuals.In English and many other languages, the subjunctive is lost and we have a sort of tense-mood syncretism.Thus, fake tense needs a different treatment, and is not to be confused with zero tense.
The example in (4E) shows that there is no 1-to-1 correspondence between semantic features and morphology.The same morphology may encode different semantic features. 5For the purposes of this study we will rely on an intuitive and informal distinction between zero tense and fake tense exemplified above.Verbal quantifiers which transmit their mood features as fake tense morphology (e.g. the counterfactual operator would) are not directly relevant for the sot parameter.
[2] c o m p l e m e n t t e n s e i n p a r a l l e l a n d c o n t r a s t The examples 6 in our study are naturally occurring data taken from two searchable parallel corpora developed at the University of Oslo -the Oslo Multilingual Corpora (OMC) and the RuN corpus, as well as the English-Russian parallel corporus provided by the Russian National Corpus.The first item listed in the examples is the original source -typically a Norwegian or Russian fiction text -then follow the translations made by professional translators.Our main focus will be on Russian and English, but we will occasionally also comment on examples from other languages, notably Norwegian, German and French.

[2.1] Simultaneous interpretation under past attitudes
The simultaneous interpretation under past attitudes is expressed by a "past under past" configuration in sot languages and "present under past" in non-sot languages. [5] The opposite is also true: the same semantic feature may be encoded by different morphological means even in one single language.To give an example: the German past form war 'was' and the present perfect form bin gewesen 'have been' mean the same.[6] Glossing is reduced to a minimum in the examples, making use of the following abbreviations: AUX = auxiliary, COND = conditional mood, FUT = future tense, IMPER = imperative mood, IMPF = imperfective aspect, INF = infinitive, PART = participle, PF = perfective aspect (including le passé simple in French), PRES = present tense, REL.PRON = relative pronoun, SUBJ = subjunctive, SUBJ.PART = subjunctive particle.sot languages include Germanic languages (here: English, German, and Norwegian) and Romance languages (here: French).The present tense in the complement of the Russian sentence (6R) is a zero copula, cf.The intensional nature of attitude verbs shows up in German, in which the embedded complement can display different versions of the subjunctive, instead of the expected "past under past".Accordingly, German attitude verbs optionally have a feature like subjunctive, which is transmitted to the complement verb.For instance, in (7G), the subjunctive present is used in German: ( If the matrix tense is expressed by a present perfect in an sot language, we expect to find a present tense in the complement -in agreement with the present auxiliary of the matrix.This pattern is found in the Norwegian translation (8N) -"present under present_perfect" -which contrasts with the English (8E) "past under past". [7] The French translation in (7F) poses interesting questions.The auxiliary a has present morphology, but it licenses the past morphology of the embedded sentait.In our theory we will have to express this by stipulating the feature combination uN, iP for the auxiliary a.In this respect, French (and other Romance languages like Italian) differs from English.In this respect, it should be noted that feature transmission carries on through non-finite verb forms like past participles as in (8N) or the infinitive "say" in (8E).
The translator's choice of lexical aspect (Aktionsarten) or grammatical aspect can influence the tense configuration, blurring the expected "past vs. present" dichotomy between sot and non-sot.The following example illustrates this point. (9) N Folk sa PAST at ville dyr bebodde PAST gemakkene.
The Norwegian original in (9N) displays "past under past" with a stative complement verb "bebo".The German translation follows this pattern, but the English translation makes use of the inchoative VP "take up residence", which excludes a simultaneous interpretation with the matrix.Hence, the English construction displays an additional second layer of past tense 8 : the past tense suffix of "had" is empty and agrees with the matrix past, while the perfect operator -the lexical content of "have" -expresses the required tense transposition by converting the VP into a stative.This gives the wanted result: The resultant state of the VP "take up residence" is simultaneous with the matrix event.Note that also the Russian translator has chosen an inchoative perfective verb, a choice which requires tense transposition and a "past under past" in Russian.
The various strategies employed in the translations of (9) can be easily accounted for by our theory once we incorporate aspect -and we will do so in section [4] (see also section [2.4] for an analysis of eventive/perfective embeddings).The literature on sot has mostly been concerned with embeddings of stative and imperfective complements, where the opposition sot vs. non-sot is (apparently) more transparent in the canonical case of a simultaneous interpretation: "past under past" vs. "present under past".However, in section [4] we will address data in Russian which complicate this picture, notably constructions involving factive matrix verbs and perception verbs.
Below we give some examples with factives that correspond to the "expected" pattern for non-sot languages, i.e., "present under past" with a simultaneous in- [8] But see some examples below.If the complement is eventive, a simple "past under past" construction in Germanic can often have a backward shifted interpretation.E Medea understood that he really was very keen to take her to this restaurant of his.N Medea forsto PAST at han hadde PAST veldig lyst til å gå INF på dette "Kaukasus"-stedet sammen med henne.
Note that the distinction between factives and other verbs of attitudes such as verbs of speech can be quite subtle.In the context of (15) below, the factive verb "learn" means "be told" (+ a factive presupposition): ( [2.2] Forward shifted interpretation under past attitudes A forward shifted interpretation may require the insertion of a covert future operator at LF in sot languages. (16) E I asked what time the attack was to be and they said as soon as it was dark.(Ernest Hemingway, "A Farewell to Arms") R Ja sprosil PAST-PF , v kotorom času načnetsja FUT-PF ataka, i mne skazali PAST-PF , čto kak tol'ko sovsem stemneet FUT-PF .
The covert future operator does not break the checking relation, as we see in Figure 4.
past they said ∅-tense as soon as it was fut dark (the attack was to be) (English) non-local agreement past skazali čto ∅-tense kak tol'ko fut stemneet (Russian) local agreement figure 4: Forward shifted interpretation under past attitude Viewpoint aspect is not represented in Figure 4, but a full analysis requires two perfective like operators in the Russian sentence, one in the matrix and one in the complement.Perfective verbs with present tense morphology like "stem-OSLa volume 2(1), 2010 [12] grønn & von stechow neet -becomes dark" receives a future (and perfective) interpretation, since the combination pf(n) -perfective and present -is semantically inconsistent.Accordingly, the usual pf operator is replaced by a semantic tense fut Rus , which has its own feature iF. 10  In example ( 17), the forward shifted meaning is explicitly conveyed by a modal or periphrastic construction in the complement of the sot languages.The Russian (17R) displays the familiar "perfective_future under past": With a non-finite complement, as the Russian (20R), the relative future is covert: (20) N Hun overbeviste PAST Goupils om at de burde PAST-MODAL sette INF ned rentene.(Nikolaj Frobenius, "Latours katalog") [10] Note that our treatment of the synthetic Russian "perfective future" is different from our analysis of the periphrastic future in English and Russian inasmuch as auxiliaries like "will" and "budet" are verbal quantifiers with a feature uN, hence subject to the sot parameter.The English auxiliary "have" is a verbal quantifier as well.[

2.3] Backward shifted interpretation under past attitudes
In Russian, "past under past" is expected to have the shifted reading, meaning that the time of the complement precedes the matrix.In sot languages, where "past under past" by default produces a simultaneous interpretation, the backward shifted interpretation is normally conveyed through tense transposition using a past perfect in the complement.A typical example is given in ( 22) with a feature analysis as sketched in Figure 5.  volume 2(1), 2010 [14]

grønn & von stechow
The English auxiliary "had" is a verbal quantifier. 11It has the same semantics as the past operator.In our example, its morphology is checked by the higher past associated with the matrix.The semantic precedence relation in the sot languages thus comes from the perfect "have" operator in (22E) and (22N).
In non-sot languages like Russian, the past tense morphology in the complement points to a local semantic past operator.Recall from our introductory remarks above that tenses are shifters, i.e., quantifiers based on temporal relations, and that the highest tense under the attitude is a zero tense, bound by lambda abstraction.Thus, when we have a semantic past under a verbal quantifier as in (22R), it is not the "past variable" that is bound, but the "perspective variable" to which this variable is related.
The same patterns as in ( 22) occur frequently in the parallel corpora.Example (23) below is similar, except for an aspectual difference in Russian: In (23R), the matrix is imperfective and the complement is perfective, while in (22R) the matrix was perfective and the complement imperfective.In the examples under discussion, this aspectual distinction does not seem to make any difference for the temporal ordering of the events -in both cases the complement event/state precedes the matrix event/state, hence a shifted reading (but see section [4.1] below on the role of aspect in Russian non-sot configurations).( 23 The present form "has" has a special semantics as it may express the "extended now"; cf.(Dowty 1979, chap. 7).
While "past under past" typically has a simultaneous interpretation in sot languages, the context can also license a shifted interpretation similar to the Russian "past under past" construction.In fact, it is well-known from the sot-literature that "past under past" leads to a shifted interpretation for Germanic and Romance languages when the embedded sentence is perfective (or eventive). 12This is illustrated in the English translation in (26E) with a simple "past under past" -in contrast to the past perfect found in the Norwegian translation (26N).The fact that the translators in (26) have chosen different forms is "accidental" -both English and Norwegian have both constructions.Thus, a simple past in English and Norwegian can have the same meaning as a past perfect, but it need not, cf. Figure 6.
... ∅-tense A.P. sent ... ∅-tense A.P. past sent ... ∅-tense A.P. had sent } synonymous -backward shifted interpretation feature transmission from higher tense figure 6: Embedded simple past vs.past perfect Although (26E) and (26N) are truth-conditionally indistinguishable, only the latter relies on the sot parameter for the licensing of the verb form (the auxiliary "hadde" with past morphology, and not "har" with present morphology) embedded under the highest verbum dicendi.The past tense morphology on the auxiliary "hadde" is thus semantically void, the anteriority being conveyed by the lexical semantics of the verbal quantifier "ha" -"have".[12] This forced backward shifted interpretation has a certain parallel in the well-known fact for all speakers of Russian that the combination of present tense + perfective aspect requires a shifted future interpretation due to the incompatibility of the perfective complete event configuration with a punctual present topic time.
OSLa volume 2(1), 2010 [16] grønn & von stechow For the second embedding under "sent word", both (26E) and (26N) require the insertion at LF of a silent future.The feature transmission for the English sentence in (26E) is shown in Figure 7. Thus, the two embeddings in (26E) demonstrate both local (p -"said" and p -"sent word") and non-local ("were") agreement. 13 The Russian original in (26R) also deserves special mention.We have two verba dicendi -the first embeds a relative past and the second a relative future.As expected for a non-sot language, the relative future is expressed with a perfective future.On the other hand, surprisingly, the relative past is expressed with an imperfective past, "prisylala PAST-IMPF " ("sent"), instead of the perfective aspect.This peculiar use of the imperfective with a "perfective" complete event interpretation -the so-called konstatacija fakta ("the factual imperfective") -is quite common in Russian and was treated extensively in (Grønn 2003), see also (Altshuler 2010).

[2.4] Simultaneous interpretation under future attitudes
We ignore data with a present tense matrix since the dependent/independent distinction is typically blurred in this environment.On the other hand, configurations with a future matrix attitude verb are expected to comply with the sot parameter.We observe "present under future" both in English and Russian: Despite the apparent similarities between the temporal surface structures in (27E) and (27R), the sot parameter forces different LFs for the two languages.This is depicted in Figure 8.In English, both "will" and "think" are verbal quantifiers which transmit the feature N from the deictic utterance situation.In Russian, [13] We will, however, see that our implementation of feature transmission enables us to maintain obligatory feature transmission in sot-languages.The feature uP originating in the highest matrix past is transmitted to the temporal center of an embedded semantic past.Here its voyage ends.The embedded past associated with "sent word" transmitts its own feature uP to the embedded verb "were".In the case of temporal agreement, as between "sent word" and "were", the feature is transmitted to the variable created by a zero tense (TPRO) and further transmitted to the verb.27), i.e., "present under future".However, the context typically suggests an independent (deictic) interpretation of the complement, hence the data are not directly related to the sot parameter.An example of independent complement tense is given in (28). 14(28) R Mne ne veriš' PRES-IMPF , sprosi IMPER-PF starikov; každyj tebe skažet FUT-PF , čto ryba teper' sovsem ne ta, čto byla PAST-IMPF .
(Anton Čechov, "Svirel'") E If you don't believe me ask the old people; every old man will tell you that the fish are not at all what they used to be.
[2.5] Forward Shifted interpretation under future attitudes With a forward shifted interpretation under a future matrix we expect to find an additional future operator in the complement.The pattern is illustrated in two examples below: (29) E 'Well,' said the gipsy, 'I'll tell you what I will do.'(Kenneth Grahame,"The Wind in the Willows") R Ladno, skazal PAST-PF cygan, ja tebe skažu FUT-PF , čto ja sdelaju FUT-PF .
(30) E I am going to reply to Mark Darcy's invitation and say quite clearly and firmly that I will be unable to attend.
[14] The present tense complement in the English translation ("the fish are not ...") of this example corresponds to a present tense zero copula in Russian ("ryba teper' sovsem ne ta").

grønn & von stechow [2.6] Backward shifted interpretation under future attitudes
The backward shifted interpretation is highly frequent in the parallel corpora.
As expected, Russian expresses the relative past with past tense morphology: "past under future".In sot languages of the Germanic type we typically find "present_perfect under future" in this environment (with feature transmission through the infinitive complement of "will").Compare the constructions in (31) and their temporal feature checking in Figure 9.
(Ljudmila Ulitskaya, "Medeja i ee deti") E I'll tell them I just came in to find a room for the night, no more than that.N Jeg skal PRES-AUX si INF til de typene at jeg ganske enkelt har PRES-AUX tinget PART meg nattelosji her.
The analysis of the "present_perfect under future" in (32N) is straightforwardthe present tense auxiliary in the complement gets its morphology from the deictic n in the matrix through the verbal quantifiers "skal -will" and "si -say".The English "past under future" shows that we must allow for the insertion of a relative past in the complement also in sot languages, similar to what we observed with a backward shifted "past under past" in section [2.3] (importantly, a relative pres like in Russian cannot be inserted in the same environment in English).
OSLa volume 2(1), 2010 complement tense in contrast [19] Constructions like the "past under future" or "present_perfect under future" are often ambiguous between a dependent and independent (deictic) interpretation of the complement (Abusch 1994a).A dependent interpretation amounts to the precedence relation "complement tense ≺ matrix tense".However, the context may pragmatically license a stronger interpretation, viz. the precedence relation "complement tense ≺ utterance time".
[3] a n a l y s i s The data in section [2] convincingly show that: • English is a clear sot language -simultaneity under attitudes is expressed by temporal agreement.
• Russian is a non-sot language -simultaneity under attitudes is expressed by the present.
We will now present the details of a formal analysis.We also refer the reader to (von Stechow 2009) for an introduction to tense semantics and feature theory.

grønn & von stechow
The time argument is the first by convention.Like "happy", "called" has a tenseless semantics!The morphology of the latter is checked by a semantic tense.At deep structure (DS), the time argument is filled by the semantically empty pronoun PRO, which is moved for type reasons at LF (cf. the PRO-theory of (Heim & Kratzer 1998, 226-228)).PRO-movement is an essential ingredient for the construction of binding chains for feature transmission.
( The perfect temporal auxiliary have is a verbal quantifier, which has the same meaning as the semantic past.But as a verb it has its own morphology, which has to be licensed by a semantic tense.(The extended now meaning for have pres is ignored here).
(39) has/had: type i(it,t) features: has uN; had uP The future auxiliary will is the mirror image of have: (41) will: type i(it,t) feature uN The temporal auxiliary would has the same meaning but the feature uP.
Russian budet ("will") has the same semantics as English will, and it has the feature uN.In addition it subcategorises for an imperfective verb, hence the embedded infinitive is always in the imperfective. 16This subcategorisation merely serves the purpose to implement aspect selection and is purely syntactic, cf. the traditional notion of "status government".The phenomenon is similar to the requirement (in English) that modals like "must" and "can" subcategorise for an infinitive, while "have" subcategorises for a past participle.This kind of subcategorisation can be captured by special subcategorisation features.
[3.2] Feature Theory: Temporal agreement Following Chomsky (1995), Zeijlstra (2004) (no inherent temporal feature) Meaning of all these: λwλtλx.xcalls in w at time t Our principles of feature transmission under semantic binding follow (Heim 1994b) and (Heim 2005): A semantic tense p or n transmits a feature [uP]/[uN] to the time variable it binds.If the variable is an argument of a tensed verb form, the feature has to agree with the tense feature of the verb, i.e., with the verbal morphology.
We furthermore assume the conventions for semantic binding outlined in (Heim & Kratzer 1998).In particular, a phrase or operator α may bind a variable via a λ-operator.In addition to quantifier raising (QR), λ-abstracts are created by PRO-and WH-movement.As said above, our theory makes crucial use of PROmovement, as can be seen by comparing the deep structure (DS) and logical form (LF) of a toy sentence in Figures 10 and 11.
First, a note to the percolation of tense features.Features percolate along the [16] Imperfective verbs must be licensed by a semantic imperfective operator, but this operator is not subcategorised by "budet" directly. [22] grønn & von stechow figure 11: LF (interpretable) head line.The feature of a temporal variable either agrees with the inherent feature of the head (as in finite verbs) or it is transmitted to the head (and percolates to the phrase).Since the semantic past is the head of the semantic tense [p n], the feature [iP] percolates to the phrase [p n].
The LF in Figure 11 is created by PRO-movement: PRO is semantically void and has to be moved for type reasons.Assuming Heim & Kratzer's QR conventions, PRO leaves the variable t 1 in situ.The movement index of PRO is interpreted as the λ-operator.Given that PRO is semantically empty, it is deleted at LF by Chomsky's principle of Full Interpretation (our trees do not represent the deletion).Thus PRO i can be read as λ i .Note that the transmission mechanism sends the feature iP of the semantic past p to the bound variable t 1 .Here the feature locally agrees with the inherent temporal feature uP of called.Non-finite forms OSLa volume 2(1), 2010 complement tense in contrast [23] have a temporal feature on their variable as well, but since they lack inherent temporal morphology, we do not need the features for licensing the morphology.We may need them instead for feature transmission to further embedded tenses.

[3.3] Intensions
We introduce an intensional λ-language with the following recursive definition of the interpretation function based on a function F that interprets the lexicon and a variable assignment g, ⟦.⟧ F,g : (i) Let α be a lexical entry of type a.Then ⟦α⟧ M,g = F (α).
(ii) Let x be a variable of type a.Then ⟦x⟧ M,g = λw.g(x),g(x) ∈ D a .
(iii) fa: Let α have type b and daughters β of type ab and γ of type a.
Verbs of speech ("say", "tell") and verbs of thought ("think", "believe") follow this scheme.The same also holds for factive verbs ("know", "understand"), which have an additional factive presupposition that the complement is true.We will say more about their semantics in section [4.3].Control verbs like "promise" or "convince" behave similarly with the additional proviso that we find a covert relative future in non-finite complements: [24] grønn & von stechow (45) a. John promised Jane he would vote for Obama.b.John promised Jane to vote for Obama.c. John convinced Jane to vote for Obama.
The HE in (47a) is a de se pronoun.Like PRO it is semantically void and has to be moved for type reasons. 19This gives a parallel analysis for finite and non-finite complements, a desideratum pronounced in (Katz 2002).

English attitudes
Thus, attitude verbs are verbal quantifiers.The sot parameter says that these verbs in English transmit their temporal features under binding to the temporal variable in the clausal complement.
Here are the relevant LFs: (48) At 5 o'clock Mary thought it was 6 o'clock. 20  The temporal variable of the subordinate "was" inherits its feature from the [17] The formula ignores the fact that the subject (y) addresses the indirect object (x).

[18]
The causative component of convince is ignored here, hence the rule does not make explicit that the subject (y) excerces pressure on the object (x).

[19]
The de re case with mistaken identity has to be analysed differently, say, with the method of structured propositions; see e.g.(von Stechow & Cresswell 1982).[20] Identity (t 1 = 6 o'clock) is transcendental, i.e., not dependent on worlds.Therefore the abstraction over worlds (λw 1 ) applies vacuously in Figure 12.The complement contains its own relative past that checks the uP of the past verb form "left". Hence we get the shifted interpretation. 21

Russian Attitudes
The Russian verbal quantifier budet and verbs of attitudes do not transmit temporal features.This is what the sot parameter says for non-sot languages.
The relative present plays an important role in Russian complements: (50) Russian Present ⟦pres Rus ⟧ = λwλtλP it .P (t) feature iN Semantically pres Rus is identity, i.e., void.The trivial operator merely serves the purpose to assign the feature uN to the temporal variable it binds.This is the present analysis for non-sot languages of Ogihara (1996). 22 pres Rus has the same semantics as the copula (e.g.English is or Russian byl).Therefore it is natural to assume that the Russian (covert) present copula has the feature iN and the Russian past copula has the feature iP.Accordingly, the Russian copulas always have two features, i.e., uN/iN and uP/iP.The u-feature marks the morphology and the i-feature marks the fact that the copula is a verbal quantifier.In a way, this is equivalent to saying that the Russian copula does not block feature [21] A remark to the notation in Figures 12 and 13: "Mary thinks in w at t[λw 1 λt 1 .P (w 1 )(t 1 )]" is, of course, equivalent with the notation used elsewhere in this paper: (∀w 1 , t 1 ) ∈ DoxMary(w, t) → P (w 1 )(t 1 ).
OSLa volume 2(1), 2010 [26] grønn & von stechow transmission.In our examples, the feature assignment by copulas plays no role and is ignored.English-type languages seem to lack a relative present of the Russian kind, otherwise we cannot explain the impossibility in English of "present under past" with a simultaneous, dependent interpretation (cf.example (1) above): (51) He said that he lives outside Moscow.*simultaneous, dependent interpretation.(only "double access") In addition to pres Rus , Russian has the deictic present n, which also checks the feature uN.In complement clauses, n is ruled out for type reasons.Therefore, the temporal argument of pres Rus at DS is always PRO.
( The morphological present of the (silent) copula is locally licensed by pres Rus in (52).Consider now the anteriority construction in (53) below.Since Russian is a non-sot-language, the verbal quantifier "skazal" ("said") does not transmit its feature uP.Therefore we need an additional p in the complement, which gives us the backward shifting.( 53 complement tense in contrast [27] (Paslawska & von Stechow 2003) and (Grønn 2007), we need a covert relative past operator in a language like Russian which lacks a composite perfect.We can insert a relative past under past or future to obtain a semantic pluperfect or future perfect.An illustration from subordinate tense is provided below, cf.example (54) and Figure 16.How do we know that we have to insert a covert past in this construction?The particle uže ("already") requires a stative as evidenced by the following contrast in English: (55) Every veteran will already *receive/OK have received warranty.
The perfect have converts the achievement "receive warranty" into a stative.Without the perfect the sentence is ungrammatical.
[4] c h a l l e n g e s t o r u s s i a n a s a n o n -s o t l a n g u a g e According to the sot parameter, "past under past" in Russian should give us a shifted anteriority reading.Claims in the literature to the contrary -i.e., that "past under past" in Russian sometimes allows for a simultaneous interpretation -at first seem to challenge our theory.We will show that three kinds of explanations in defence of the sot parameter are available depending on the data under discussion: • It follows from our semantics of aspect that complements which carry imperfective aspect may still be compatible with a simultaneous event interpretation despite the topic time in the complement being backward shifted.Hence these cases are not in contradiction with the sot parameter.
• The matrix is not a verbal quantifier, hence not subject to the sot parameter.
[4.1] Shifting under attitudes and aspect in Russian Altshuler (2008) objects against the traditional view that "past under past" means backward shifting in non-sot languages like Russian.He says that embedded imperfectives in the past can have a simultaneous or a backward shifted reading.Embedded perfectives are always anterior.
To simplify the exposition, we have not given a semantic analysis of aspect so far.The fact that we can have a kind of "simultaneity" with "past under past" in Russian if the embedded verb is imperfective is straightforwardly explained by considering the meaning of the imperfective.We use E for properties of events.τ (e) is the running time of the event e.The meaning of the imperfective is crudely simplified, but good enough for our purposes. 23 Here are some examples of the kind which motivated Altshuler's objections to the treatment of Russian as a non-sot language: (57) E Harris said that, to himself, it was always a mystery how people managed to get sick at sea. (Jerome K. Jerome, "Three Men in a Boat") R Garris skazal PAST-PF , čto lično dlja nego vsegda bylo PAST-IMPF zagadkoj, kak ėto ljudi uchitrjajutsja PRES-IMPF stradat' INF morskoj bolezn'ju.The imperfective has at least three different uses: • Progressivity: this interpretation needs an intensionalisation in the style of (Dowty 1979).
• For Russian, one also has to consider the general-factual interpretation, see (Grønn 2003).
It is a matter of dispute whether a unified account can be given for all these readings.
OSLa volume 2(1), 2010 complement tense in contrast [29] las' PAST-PF čast' vozov za les; no podumali PAST-PF , čto kozaki gotovilis' PAST-IMPF sdelat' INF zasadu; tože dumal PAST-IMPF i francuzskij inžener.(Nikolaj Gogol', "Taras Bul'ba") E From the tower of the town hall the sentinel only perceived that a part of the waggons had been dragged into the forest; but it was thought that the Cossacks were preparing an ambush -a view taken by the French engineer also.
In the examples above, we have "past under past", but the imperfective state in the complement clearly holds before, at and, perhaps, after the matrix event.This allows for an analysis where the topic time of the complement (a proper subinterval of the (progressive) state) precedes the topic time of the matrix.We show how this works by giving an analysis of the last example above: 24 (60) podumali PAST-PF , čto kozaki gotovilis' PAST-IMPF sdelat' INF zasadu (it was thought that the Cossacks were preparing an ambush) We have backward shifting in the embedded clause.But since the VP expresses a progressive state and the topic time is in the time of the state, the state might continue at the "subjective now" t 1 .This gives us the feeling that the reading is simultaneous.This is a point stressed emphatically in (Klein 1994).
Note that according to our theory, the past progressive in the English translation (58E) can, strictly speaking, also get its simultaneous reading -in terms of temporal overlap between the matrix and complement events -through a backwarded shifted topic time, as in Russian.As formulated above in section [1.1], our sot parameter does not require long-distance feature transmission for English, hence it does not by itself block the insertion of a local semantic past in the complement.Our theory does not say where the semantic tense is located at LF. 25  The two simultaneous readings for English obtained through non-local or local agreement are indeed hard to distinguish.For Russian, on the other hand, the sot parameter forces local binding from a semantic past in the complement.
As pointed out by a reviewer, temporal feature transmission from the matrix forces a simultaneous interpretation for the English sentence (58E), while the proposed truth conditions for (58R) do not.We think that this is as it should be.Indeed, although (58E) is an adequate translation of (58R), the two constructions are not completely equivalent.While the English translation (in principle) [24] Recall from section [3] that PRO i can be read as λ i .The official notation would have an erased PRO and λ i .Often we simply write λ i . [25] In our system, feature transmission is not optional, but there may be different possible licensers. [30] grønn & von stechow allows for two different tense configurations (cf.previous paragraph), the Russian original (in principle) allows for two different aspectual configurations.As argued in (Grønn 2007), the so-called "factual" imperfective past in Russian (i.e., impf with a complete event interpretation) is, again in principle, i.e., semantically, compatible with a backshifted relative past configuration.However, due to competition from the perfective past, which explicitly encodes a proper backward shifted reading under attitudes, the imperfective is pragmatically strengthened to encode a simultaneous (progressive) reading.Thus, if the embedded past VP in Russian is in the perfective aspect, we always have the anterior reading.This was confirmed by our survey of the data in section [2].We give the analysis of one example here: (61) E I've told them I've found someone.
(Helen Fielding, "Bridget Jones's Diary") R Ja uže skazala PAST-PF im, čto našla PAST-PF koj-kogo.( 62) It is clear that the reading has to be anterior because the backward shifted interval contains a complete or "quantised" event.A property of events is "quantised" if it does not apply to subintervals of its running time.
[4.2] Tense under perception verbs in Russian Not every verb that selects a CP is a verbal quantifier.Perception verbs are not verbal quantifiers in their prototypical use; hence we should not be surprised to find "past under past" with a simultaneous interpretation in Russian.Nevertheless, constructions with perception verbs often follow the familiar patterns predicted by the sot parameter.We will review the different data below.
Simultaneous interpretation under past perception verbs: "past under past" Khomitsevich (2007) observes that "past under past" frequently expresses simultaneity in perception constructions even for embedded achievements or accomplishments.She correctly claims that this is an effect of the semantics: on the default, direct perception reading, the fact/event perceived must be simultaneous with the perception.In this construction, the complementiser in Russian is typically "kak" ("how") instead of "čto" ("that").Here are some examples from our parallel corpora: (63) N Han så PAST tuppen av en pisk som falt PAST mot Valéries kropp, og han hørte PAST at mannen roste PAST det stygge trynet hennes.

Simultaneous interpretation under past perception verbs: "present under past"
With a "present under past" in Russian, the perception verb construction is reminiscent of verbs of attitudes.The perspective is not that of the speaker, but of the perception holder, hence we get a kind of evidential reading.
(66) N Men han så PAST at hun ikke var PAST helt seg selv.
(Herbjørg Wassmo, "Dinas bok") E But he saw that she was not completely herself.G Aber er sah PAST , daß sie nicht ganz sie selbst war PAST .
Note also that the distinction between perception verbs and factives is sometimes very subtle.Evidence for knowing can be visual or acoustic.The Russian translator in (66R) below can therefore use a perception verb to convey the meaning of the factive verb in the original: (67) N Bou-Bou visste PAST at han løy PAST .
The evidence for him being a liar in (66R) can be indirect.Constructions with perception verbs are therefore often ambiguous between direct perception and indirect perception.
In analogy with complements of attitude verbs, we find the expected complementiser "čto" ("that") in the two previous Russian examples.The facts are more complicated though, as witnessed by the next example: In (68), only the Norwegian original ("past under past") and the Russian translation ("present under past") have a finite complement clause.With the non-finite complements in the English, German and French translations, the perception construction can only have a direct perception reading.The Russian complementiser "kak -how/as" -instead of "čto -that" -is also an indication of direct perception.Given a Higginbotham style analysis of perception verbs (see below), one would expect a "past under past" construction also in Russian.The translator's choice of "present under past" makes the construction more "vivid" and similar to attitude verbs: the complement is reported from the perspective (acoustic alternatives) of the perception holder.

Analysis of tense under perception verbs
Thus, perception verbs express simultaneity either by embedding a past or a present.If these verbs were analysed like attitudes, present would be expected, with OSLa volume 2(1), 2010 complement tense in contrast [33] a sort of evidential reading.If past and present were freely interchangeable in these constructions without a change in meaning, then that would be a problem.
We suggest that the alternation may to some extent reflect an ambiguity between direct perception from the speaker's perspective (⇒ "past under past") and visual/acoustic alternatives from the perception holder's perspective (⇒ "present under past").Our analysis of direct perception closely follows (Higginbotham 1983). 26Higginbotham analyses perceptions as a relation between an individual and an event, the eventive reading.The analysis will entail that these verbs are (normally) not verbal quantifiers and hence not subject to the sot parameter.Let us relate the discussion to a toy example from Altshuler (2004): (69) a. Dina videla PAST-IMPF , kak (čto) voda lilas' PAST-IMPF iz vedra.'Dina saw how (that) water was pouring [poured] from the basket' b.Dina videla PAST-IMPF , čto (kak) voda l'ëtsja PRES-IMPF iz vedra.
'Dina saw that (how) water was pouring [pours] from the basket' The complement in (69a) describes an event and may, of course, localise it in time.We expect a de re past in the complement, given that the truth-condition of (69a) is something like this: (70) There was a water pouring e & Dina saw e.
On this account, it is obvious that "see" is not a (normal) verbal quantifier, at least when the verb is used to denote direct perception.This analysis of the construction has interesting consequences for the semantics of aspect.Here is the result we want for (69a) (in Altshuler's Russian example the perception verb itself is also in the imperfective aspect): We have two independent deictic pasts, so no temporal agreement is needed.Furthermore, the existential quantifier on top binds the variable e that occurs as an argument of the aspect in the the subordinate clause and as the direct object of the matrix verb.This means that the event variable in the aspect is not always locally bound by an existential quantifier.Accordingly, we have to modify the logical type of the aspectual operator: ( We can regain the familiar existential analysis for aspect by existential closure of the variable e. The complement clause is headed by "kak" ("how"), here analysed as an existential quantifier over events, as in (73).The compositional build up of the LF for the complement clause is depicted in Figure 17 The "kak"-clause is the object of the perception verb videla 1 , which has the following semantics: (74) ⟦videla 1 ⟧ = λwλe 1 λe 2 λx.e 2 is a seeing of e 1 by x in w type (v(v(et))) features: uP, uIMPF In order to be able to have the "kak/how"-clause as an object, we must QR the generalised quantifier from the object position.We end up with the LF in Figure 18, which eventually brings us back to the truth-conditions which we started out with in (71).
A note concerning feature transmission is in order: Temporal features are transmitted across the aspect operator to the variable that the latter binds.Thus complement tense in contrast [35] [ CP kak ∃ λ 3 p(n) λ 2 impf(e 3 )(t 2 ) λ 1 voda lilas'(e 1 )] iP iIMPF uP uP, uIMPF λ 4 [ S p(n) λ 5 (∃ 6 ) impf(e 6 )(t 5 ) λ 7 Dina videla(e 4 )(e 7 )] iP iIMPF uP uP, uIMPF figure 18: LF of (69a) with temporal and aspectual features the aspect operator transmits two kinds of features: its own aspect feature and the temporal feature inherited from the semantic tense of the sentence.This analysis explains the fact that "past under past" may express simultaneity, i.e., direct perception, in perception constructions.Let us now turn to "present under past" and example (69b).
Indirect perception verbs are verbal quantifiers.Their analysis is similar to believing or knowing.The possible difference is that the attitude is based on visual, acoustic or other sensomotoric evidence.For instance, for the truth of (69b) it is sufficient to see that the desk under the basket gets more and more wet.In the limiting case the evidence is direct, in which case the "seeing that" is something like realizing by direct visual evidence.There might be a de re variant as well, similar to our de re analysis of factive verbs (See section [4.3] below). 27 Thus, since perception verbs with subordinate present are verbal quantifiers, they have the standard semantic assumed for attitudes.For concreteness, we indicate the lexical entry for videla in (69b): where Acc x (w, t) are the world-times compatible with the visual evidence that x has in w at time t.
We assume that "kak/čto" is a complementiser without meaning in constructions involving these verbs.Under these assumptions, (69b) is analysed exactly as any belief-sentence and means: 28   [27]  The relevant paraphrase for the double access reading of "present under past" is this: "There is a state s of water pouring and she saw s".A present state of water pouring may have had its beginning some time ago, and she saw that very state in the past.This is precisely the idea behind the double access reading of "John believed that Mary is pregnant.",cf.(Abusch 1997).[28] We ignore the event arguments and aspects in (76).Concerning tense, we repeat from above what should be clear by now: Since the complement clause is tenseless, it has to be controlled by the matrix tense, hence we have dependent tense and the sot parameter applies.As we did for present tense complements under normal attitude verbs in Russian, we must insert the identity function pres Rus in order to license the present tense morphology in the complement. [36] grønn & von stechow If the cognitive evidence is reliable, the complement of a perception verb is true and there is virtually no difference in truth conditions between the direct and indirect perception construction.This is witnessed by the fact that both constructions can be conjoined, i.e., we find alternation between "past over past" and "past over present", as in ( 77): (77) R Ona gljadela PAST-IMPF , kak sizye kol'ca ot sigary Azarello uplyvali PAST-IMPF v kamin i kak kot lovit PRES-IMPF ich na konec špagi.
(Michail Bulgakov, "Master i Margerita") E She watched as Azarello blew (Rus: past!) smoke-rings at the fireplace and the cat spiked (Rus: present!) them on the end of his sword.N Hun så PAST på Azazello, som sendte PAST ringer av sigarrøk inn i peisen, og på katten, som fanget PAST dem opp med kordespissen.
[4.3] Tense under factive verbs Our informal corpus study in section [2] shows that many authentic examples with factives comply with the patterns characteristic of attitude verbs in Russian, i.e., "present under past" is used for a simultaneous interpretation.However, as with perception verbs, we find data with "past under past" and a "simultaneous" interpretation also in factive constructions.We will argue that these data are not counterexamples to our sot theory, since factives have a special semantics which is responsible for this vacillation.Here are some relevant data with "past under past" in Russian: (78) N Alle visste PAST at lensmannsfrua var PAST flere måneder på vei da hun ble PAST-AUX skoldet PART i hjel.(Herbjørg Wassmo, "Dinas bok") E Everyone knew the sheriff's wife was several months pregnant when she was scalded to death.G Alle wußten PAST , daß die Lehnsfrau schon mehrere Monate schwanger war PAST , als sie zu Tode verbrüht PART wurde PAST-AUX .R Vse znali PAST-IMPF , čto žena lensmana byla PAST-IMPF beremenna, kogda obvarilas' PAST-PF ščeloč'ju.
(Lev Tolstoj, "Anna Karenina") E Today was Saturday, and he knew that the letter contained reproaches for not being back at the time fixed.N I dag var PAST det lørdag, og han visste PAST at brevet inneholdt PAST bebreidelser over at han ikke var PAST-AUX kommet PART hjem i rett tid.Khomitsevich (2007) writes that embedded past and present are freely interchangeable in Russian in these constructions without change in meaning.Indeed, the difference in truth conditions may be negligible, but the derivation at LF must still be different.Note that as a first try we could naïvely adopt the same explanation for these data as for the examples discussed in section [4.1] above where a simultaneity reading occurred with imperfective past complements under attitude verbs.For those cases we pointed out that the sot parameter only requires the topic time of the complement to precede the topic time of the matrix, hence we can still have a kind of simultaniety of the events described in the matrix and complement.However, this approach would not do justice to the semantics of the factive construction.Our semantics for factives will make it clear that we have an independent de re past in the complement of examples like ( 78) -( 80), similar to what we observed with "past under past" and perception verbs in the previous section.Thus, we agree with Khomitsevich that "past under past" with factives is never a (non-local) tense agreement phenomenon in Russian.Our analysis of these readings will follow Kratzer (1990)'s analysis of factual belief, according to which one of the roles of the complement clause is to describe the known fact.This will bring the complement clause in a transparent position and explain why the subordinate tense is interpreted independently from the matrix tense.In other words, the temporal locations of the two facts are described independently, which does not exclude the two facts from overlapping in time (simultaneity).
In order to understand the semantic rationale behind these findings, we remind the reader of the analysis of facts given in (Kratzer 1990).Consider the following sentence and its analysis: (81) John knew that it was raining.
(82) a.There was a state/event e, which was a raining, and b.John was acquainted with e, and c.John believed of e the property of raining.
The complement clause is used in condition (a) and in condition (c).(a) is a purely extensional environment and the interpretation of the past tense is independent of the tense of the main clause.(c) is an intensional environment, where the com-OSLa volume 2(1), 2010 [38] grønn & von stechow plement clause serves for the description of a tenseless property (a zero tense; a relative bound present in Russian).
Given that the complement clause has to fulfil two roles for getting the content right, we face a dilemma of expressibility: if the complement is in the present, the complement can describe only condition (c) correctly.If the complement is in the past, it can describe only condition (a).
In order to implement the two strategies of projecting the complement clause, we will adopt a trick: if the visible complement clause is in the past, it will describe the res, i.e., the event/situation the subject holds the attitude of, and the property believed of the res will be represented by a free property variable, which is supposed to be the tenseless counterpart of the complement.If the complement is in the present, it will be the property attributed to the res and the tensed counterpart will be represented by a free property variable.Now, let us see how this works with a minimal pair from Russian (cf.example (10) above): (83) a.On znal PAST-IMPF , čto ona stojala PAST-IMPF u okna.'He knew she was standing by the window.' The analysis is sketched in Figure 19 on the next page.The crucial points to notice are the following: We are flexible in choosing between the two versions of the aspects from (72).In the complement, we have the new version where the event argument is not locally bound, while in the matrix, aspect has its traditional logical type.In the end, the structure gives us existential closure of the event argument in both cases.The complementiser "čto" is a λ-operator (an abstraction over the event variable).We assume that Q in (83a) gets the following interpretation: This is the version where the past in the complement denotes a tense that is independent from the matrix tense.Hence we can have simultaneity, cf. the result we obtain from computing the derivation in Figure 19: The relation believes_of(he, e, e 1 , Q, w) can be analysed further in the style of (Lewis 1979).This is rather complicated because it involves the particular re-  Here it can be paraphrased roughly as "he believed being in a state of acquaintance with a situation that was a standing at the window of her".
Next, consider the case where we have present tense in the complement clause: (87) a.On znal PAST-IMPF , čto ona stoit PRES-IMPF u okna.
The upper part of the LF tree is exactly as before, but the complement clause is in the present.Thus, the complement depicted in Figure 20 is the overt version of our former property Q.We now need a new rule where Q plays the role of the past tensed counterpart of this property: Therefore, the proposition is the same as before, and we have an explanation of why we can express simultaneity either by "past under past" or by "present under past" .There are of course other factive verbs than knowing, e.g., the emotive factive.We cannot go into these here.We hope that similar explanations are available.For the time being we conclude that factive verbs do not constitute counterexamples to the sot parameter.
The sot parameter thus provides a nice diagnostics, indicating how the constructions in question should be analysed.Given that Russian is a non-sot language, we have been guided in section [4] by the sot parameter in the following sense: If we have a "present under past", the matrix is likely to act as a verbal quantifier (i.e., a kind of attitude verb).On the contrary, a "past under past" with a simultaneous reading in the strict sense of temporal overlap between the topic times of the matrix and complement is not possible in Russian on the assumption of tense dependency between the complement and the matrix.This is what the sot parameter tells us, and our findings and analysis of perception verbs and factives with a "past under past" indeed confirm that different factors conspire to produce an independent tense interpretation in these cases.
[5] f u t u r e e x t e n s i o n s In the study of tense dependencies and subordinate tense, three syntactic environments are particularly interesting: • Tense in complements • Tense in relative clauses

• Tense in adverbial clauses
We have only dealt with complement tense in this article, but we intend to further explore the sot parameter with respect to tense dependencies in adjuncts, both relative and adverbial clauses.This issue is less explored in the literature since adjuncts are not syntactically dependent on the matrix in terms of c-command.Indeed, adjunct tense is often independent of the matrix, hence deictic.Still, certain configurations -notably with the verbal quantifier will in Englishclearly display tense dependency and tense agreement in sot languages.
[6] s u m m a r y Below, we sum up some of our main claims and findings.
• Non-local tense agreement in English complements is licensed by verbal quantifiers such as will/would, have and verbs of attitude.The feature transmission is achieved via binding chains that are created by PRO-movement.
• In Russian, no feature transmission across temporal quantifiers is possible.This fact motivates the formulation of the sot parameter, according to which Russian is a non-sot-language.
• There are obvious differences between the English and the Russian tense systems.Russian has no perfect auxiliary have and may therefore require the insertion of a covert past.Another difference is the assumption of a relative pres Rus for Russian.English does not have such a tense.The need for this local licenser in Russian follows directly from the sot parameter and the requirement that tensed forms have to be licensed by a semantic tense.In English, embedded tenses can be licensed from semantic tenses in the matrix across temporal quantifiers, but this is not possible in Russian.
• Recent objections against the view that Russian is a non-sot-language are refuted by a more careful investigation of the semantics of the verbs involved.Not all verbs with complements are verbal quantifiers.
-Verbs of perception are normally not verbal quantifiers and therefore not subject to the sot parameter.The syntactic complement serves the purpose to describe the event perceived.There is no temporal subordination.
-Factive verbs like "knowing" are more complicated.There the complement serves two purposes: (a) it describes the res situation, and this description is not subordinated to the main verb; (b) it describes the tenseless property ascribed to the res.The visible complement may relate to the (a) purpose or the (b) purpose.In the first case, the factive behaves like a perception verb and we find "past under past" for simultaneity.In the second case, the factive behaves like a verb of attitude, i.e., a verbal quantifier, and we find "present under past".
a c k n o w l e d g m e n t s We wish to thank Daniel Altshuler whose comments at various stages of our work forced us to work really hard.
PAST at han hadde PAST-AUX vaert PART den betydeligste i red-[11] Simultaneous interpretation under future attitude the checking relation is broken by the verbal quantifier "dumat'/think", hence a relative present must be inserted in the complement to license the present tense morphology of "zatevaete -be up to something".Many examples of English and Russian in the parallel corpora display a similar pattern as in ( below. is the speech time.Thus the semantic present simply denotes the speech time.p is an indefinite relative past.15Theargument of p is always n in matrix clauses.In subordinate clauses, p can have a time variable t as argument that is bound by a higher tense or locally bound by a λ-operator. * b.Past, type i(it,t): ⟦p⟧ = λwλtλQ it .(∃t 1 )[t 1 ≺ t & Q(t 1 )] (Heim 1997) s * e ⟧ = λw.John b. ⟦called i(et) ⟧ = λwλtλx.xcallsin w at t. feature: uP c. ⟦happy i(et) ⟧ = λwλtλx.xis happy in w at t. no tense feature [15] For the Partee Problem, see (von Stechow 2009).OSLa volume 2(1), 2010 [ TP [ T p t 0 ] PRO λ 1 [ VP John [called t 1 ]]] = λw.(∃t≺s * ) John calls in w at t.Copulas are trivial verbal quantifiers.In English, they transmit temporal features, in Russian they do not.In most cases feature transmission by copulas does not matter for the analysis of the data.