201 94 4MB
English Pages 174 [176] Year 1998
Linguistische Arbeiten
379
Herausgegeben von Hans Altmann, Peter Blumenthal, Herbert E. Brekle, Gerhard Heibig, Hans Jürgen Heringer, Heinz Vater und Richard Wiese
Regine Eckardt
Adverbs, Events, and Other Things Issues in the Semantics of Manner Adverbs
Max Niemeyer Verlag Tübingen 1998
Für Luzie, die dann doch schneller
war.
Die Deutsche Bibliothek - CIP-Einheitsaufnahme Eckardt, Regine: Adverbs, events, and other things : issues in the semantics of manner adverbs / Regine Eckardt. - Tübingen : Niemeyer, 1998 (Linguistische Arbeiten ; 379) ISBN 3-484-30379-4
ISSN 0344-6727
D 93 © Max Niemeyer Verlag GmbH & Co. KG, Tübingen 1998 Das Werk einschließlich aller seiner Teile ist urheberrechtlich geschützt. Jede Verwertung außerhalb der engen Grenzen des Urheberrechtsgesetzes ist ohne Zustimmung des Verlages unzulässig und strafbar. Das gilt insbesondere für Vervielfältigungen, Übersetzungen, Mikroverfilmungen und die Einspeicherung und Verarbeitung in elektronischen Systemen. Printed in Germany. Gedruckt auf alterungsbeständigem Papier. Druck: Weihert-Druck GmbH, Darmstadt Buchbinder: Industriebuchbindeiei Hugo Nädele, Nehren
Table of Contents
Preface
vii
1. Introduction 1.1. Manner Adverbs 1.2. The Operator Approach of Thomason and Stalnaker 1.3. The Narrow Operator Approach 1.4. The Event Based Approach 1.5. Appendix: Operators Are Not Essential
1 1 3 8 10 12
2. Ontology 2.1. Events and Sets of Times 2.2. Events as Spatio-Temporal Regions 2.3. Semantic Participants 2.3.1. The Polyadicity of Verbs 2.3.2. Thematic Roles or Semantic Participants 2.3.3. Manner Adverbs and the VOLI relation 2.3.4. Beneficients and Instruments 2.3.5. Telicity and Atelicity 2.3.6. Post States 2.3.7. Moved Objects - Grammatical or Semantical Information? 2.3.8. Subtle Parameters 2.4. Summary
14 14 16 20 20 22 29 34 35 37 39 40 42
3. Events and Their Names 3.1. Attribution 3.1.1. Spreading and Non-spreading Predicates 3.1.2. Partiality 3.1.3. Representation of Predicates 3.1.4. A Greediness Parameter? 3.2. Evaluative Adverbs 3.3. Causation 3.3.1. Lewis 3.3.2. Dowty 3.3.3. Real Causal Statements and Pseudo Causal Statements
43 43 46 49 49 50 52 57 58 61 64
4. Mereological Structure 4.1. Evidence in Favour of Complex Events 4.1.1. Complex Events and Complex Objects 4.1.2. Temporal Modification of Complex Events
71 71 73 75
vi 4.1.3. Spatial Modification of Complex Events 4.1.4. Causation and Complex Events 4.1.5. Complex Events in Lexical Semantics 4.1.6. Manners of Complex Events 4.1.7. Anaphoric Reference to Complex Events 4.2. Persistent Event Structures 4.2.1. Lasersohn's Structures 4.2.2. A Constructive Definition of Simple Event Structures with Persistency (SEP) 4.2.3. A Characterization Theorem for Lasersohn-Structures 4.2.4. Evaluating Lasersohn-Structures 4.3. Non-boolean Conjunction 4.3.1. Using or not Using Persisteny 4.3.2. "And" as Summation 4.3.3. Linguistic Applications 4.3.4. Summary
76 77 78 78 79 80 81 87 91 98 103 103 109 113 115
5. Scope 5.1. Modification of Big Events 5.2. Existential Closure, and Diesing's Mapping Hypothesis 5.3. On Little and Big Events
116 116 120 123
6. Manner Adverbs and Word Order 6.1. Manner Adverbs and Indefinite NPs 6.2. Intonation and Normality 6.3. Focus, and the Interpretation of Indefinites 6.3.1. Two Kinds of Foci 6.3.2. Examples 6.3.3. More Examples 6.3.4. Indefinites and Judgement Structure 6.3.5. Lexical Conditions on Judgement Structure 6.3.6. Relation to the Theory of Diesing 6.3.7. A Short Note on Normal Accents 6.4. Adverbs of Degree of Perfection 6.5. Conclusion
127 128 130 136 136 140 143 144 147 153 156 159 164
7. Bibliography
165
Preface
The development of this book has a long history, beginning with my starting to work on my Ph.D. thesis at the Institut für maschinelle Sprachverarbeitung at the University of Stuttgart. I want to thank all my colleagues both at Stuttgart, and at the Seminar für Sprachwissenschaft, University of Tubingen, for the helpful comments and arduous discussions which have shaped the contents of this book. Special thanks are due to Arnim von Stechow, Caroline Féry, Cécile Meier, Fritz Hamm, Hubert Haider, Inga Kohlhof and Mats Rooth who all gave me the opportunity to improve my understanding of my own ideas by trying to communicate them to others. Heart attacks have been caused by Daniel Biiring. Much of the contents of this monograph were further developed at the Institut für Sprachwissenschaft, University of Düsseldorf. I profited greatly from intense discussions with members of the Sonderforschungsbereich "Theorie des Lexikons'', and would like to thank Sebastian Löbner for the lively and inspiring atmosphere of our project and many insightful remarks which have found their way into the present work, in one way or other. I am grateful for the financial support of the Graduiertenkolleg "Grundlagen der maschinellen Sprachverarbeitung", University of Stuttgart, and of the ESPRIT project No. 6852 "DYANA". I had the privilege to have Ede T. Zimmermann and Hans Kamp as my advisors, who are both able to encourage intense and lively scientific discourse, and whose aim in formulating critizisms always has been to come to a mutual understanding of the questions under debate. I want to thank my parents for teaching me all the basic qualities and giving me all the moral support necessary for an enterprise like this.
Konstanz, September 1997
Chapter 1. Introduction
1.1. Manner Adverbs The term "adverbs" covers a broad class of words which is semantically inhomogeneous. This book will be concerned with the class of manner adverbs, the subclass of those adverbial modifiers of which examples like the following are prototypical: carefully, shyly, quickly, loudly, hesitatingly, skillfully,... These modifiers tell us something about the way in which the action denoted by the verb takes place. If we know that Alma sang joyfully then we will have a more specific idea about what took place in the world than if we simply knew that Alma sang. Some of those modifiers, like "quickly", seem to convey only temporal information, others in contrast refer to a participant's attitude in the action in question, like "shyly" or "joyfully". The limits of the class of modifiers to be investigated here might best be marked out by listing phenomena which I will not be concerned with. I will distinguish manner adverbs from modal and other intensional operators, like "probably", "necessarily", or "intentionally". These operators behave systematically different from manner adverbs in that they give rise to opacity effects (see 1.2.). I will also not be concerned with adverbial quantification ("always", "often") or the fact that certain adverbials refer to social standards ("illegally", "professionally speaking"). More importantly, we will distinguish manner adverbs from evaluative adverbs. This distinction is rather subtle for English, but more easily visible in German. While the use of a manner adverb α can be paraphrased by "it was a how ...", sentences with evaluative adverbs β must be paraphrased by "it was β that ...". In English, manner adverbs and evaluative adverbs often are homonymous. Thus, a sentence like (1) is ambiguous: (1) (la) (lb)
Alma carefully opened the fridge. It was careful how Alma opened the fridge, (manner adverb) It was careful that Alma opened the fridge, (evaluative adverb)
In German, almost all evaluative adverbs are derived from manner adverbs by the suffix "weise". Sentence (1) will translate as (2a) in reading (la), and as (2b) in reading (lb): (2a)
Alma öffnete vorsichtig den Kühlschrank. Alma opened carefully
(2b)
the fridge
Alma öffnete vorsichtigerweise den Kühlschrank. Alma opened carefully
the fridge
2 The distinction of manner adverbs and evaluative adverbs is further supported by the observation that they differ in semantic behaviour (see Wyner[94]). We will come back to this distinction in chapter 2. A broad survey of the whole class of adverbs can be found in R.Bartsch's book "The Grammar of Adverbials" (Bartsch [76]). How should manner adverbs be represented semantically? Basically, we find two suggestions with respect to this question in the literature. On one hand, it has been suggested that adverbs 1 should operate as functors on verb and verb phrase meanings which are provided by an independent semantic framework. I will call this approach the "operator approach". It goes back to Thomason and Stalnaker (Thomason/Stalnaker[73]) and will be discussed in the next section. Due to certain difficulties with the operator approach, we will also take into account a variant of it, namely a "narrow operator approach" where adverbs are taken to be operators on verb meanings alone. The alternative view holds that manner adverbs should be represented on the basis of an additional parameter of the verb, the event parameter. The idea that verbs relate individuals to an event goes back to Davidson (Davidson[67]). The event based approach makes use of this idea by assuming that manner adverbs denote predicates of events. This approach already was part of Davidson[67] and was adopted by many later authors, like Bartsch (Bartsch[76]), Parsons (Parsons[90]), Wyner (Wyner[94]), etc. The treatment given by McConnell-Ginet (McConnell-Ginet[82]) does not readily fit into the opposition between operator and event based treatment; I will come back to it in chapter 2. It is important to realize that the crucial difference between the operator approach and the event based approach does not lie in the logical type we ascribe to the meanings of adverbs. I will shortly sketch at the end of this chapter that once we have made up our minds as to what parameters are modified by a manner adverb, we can model this modification with "operators" as well as with simple "predicates" on events. The essential difference lies in the conceptual basis I want to associate with either approach. While the operator approach attempts modelling the semantics of manner adverbs on the basis of previously known ontological categories like "time", "individuals", "space", "worlds", the event based approach will be committed to the use of a new ontological category, namely events. It will therefore be necessary to give an explicit answer to the question whether events can be reconstructed using "old" ontological categories, or whether they constitute an independent category on their own. It has been argued that events are complex entities to be derived from times and worlds, like by Cresswell (Cresswell[85]) or Montague (Montague[69]). In some chapters of Cresswell[85] (which is in fact a collection of papers), the author states that no claim is made as to whether events are basic or derived entities. We will, however, take a different position.
I will use "adverbs" interchangeably for "manner adverbs" if the context makes it clear that no other kinds of adverbs are under debate.
3 This book will be devoted to investigate the interplay of the semantics of manner adverbs and the ontology of events. I will argue that the "old" ontological categories do not provide the correct level of distinction which is captured by "events", if events are individuated according to the requirements set by adverbial semantics. It will be important, though, to make sure that events are not shaped to fit the needs of manner adverbs alone. Event objects were created to be multi-purpose objects, as we will see in chapter 2. We know of a range of phenomena which depend on these objects, and we will make sure that the semantic treatment of manner adverbs does not violate ontological properties of events as required by other classical applications of "event semantics". The next section will recapitulate the operator approach in its classical variant and show that this is not a satisfying way to model the semantics of manner adverbs.
1.2. The Operator Approach of Thomason and Stalnaker
R. Thomason and R. Stalnaker were the first ones to give a treatment of manner adverbs within the semantic framework developed by Montague (Montague[73], Thomason/Stalnaker[73]). Coming from the idea that the meaning of verbs, extensionally speaking, should be represented as sets of individuals, they suggested that adverbs make modified verbal predicates out of plain ones. We take "singing loudly" to be a specific variant of "singing", and "carefully carrying out the garbage" as a specific variant of "carrying out the garbage". Knowing what "carefully" means is the same as being able to tell which instances of a given verbal relation are moreover "done carefully". Thus, adverbs are represented as functors from verb meanings to verb meanings. I will adopt the convention to use [[ α ]] to refer to the denotation of an expression α of natural language within a truth value based semantic representation theory. (1)
[[adverb]]: [[carefully]]:
[[verb]] [[sing]] [[ climb ]] [[ eat a bagel ]] —>
[[ adverb ] ] ( [ [ verb ]] ) [[ carefully ]]( [[sing]] ) [[ carefully ]] ( [[climb]] ) [[ carefully ]] ( [[eat a bagel]] )
etc... If we spell out this idea in an extensional way, we will get the following theory: Assume that we work in a type logical framework like Montague's PTQ (Montague[73]) and its followers. Let D < e t> be the domain of extensions of VP-meanings. Adverbs will be represented as objects of type D < < e t> < e { » . They denote functions from VP meanings to VP meanings. This still looks a bit rough. Not everything we can say about manner adverbs is captured by this simple suggestion. Further restrictions of the possible values of adverbs might be desirable, for instance the following: As it stands, this kind of semantic representation will
4 allow for the modified verbal predicate to be mapped on a set of individuals outside the original unmodified verbal predicate. We could have a model where Judy, Tim and Sue are singing, but where only Paul is singing joyfully. The following kind of implication would no longer be available: (2)
Judy is singing joyfully - * Judy is singing.
Working in a truth value semantic framework, we have further means in order to shape the intended semantic values of words. For instance, we can integrate implications like (2) with the help of meaning postulates: (3)
For all manner adverbs α the semantic representation [[ α ]] has the subset property: If [ [ « ] ] = F in D«e,t>,> *hen for all A in D < e > t>: F(A) is a subset of A.
This will ensure that the above implications are valid. The operator approach in this simple spellout models the motivating intuitions which we gave above. However, it is not fine-grained enough. If adverbs literally denote functors of VP extensions into VP extensions, we will make the prediction that whenever two VP extensions accidentially coincide in some world, then the modified verbal predicates also have to coincide for logical reasons. Let us go through an example. Assume that in some model the set of persons who sing is exactly the set of persons who drum: (4)
[[ sing ]] = { Molly, Bo, A r t } [[ drum ]] = { Molly, Bo, A r t }
Assume further that the adverb "loudly" is represented as a functor on sets of individuals, and that Bo is singing loudly, but nobody else. This is expressed by (5). (5)
[[ loudly ] ] ( [ [ sing ]] ) = {Bo}
The functor denoted by "loudly" operates on the set denoted by "sing" without knowing the verb which denoted this set. Thus, we will also get that (6)
[[ loudly ]] ( [[drum]] ) = {Bo}
and predict that Bo is not only the only one who sings loudly but also the only one who drums loudly. This is not correct. It was too optimistic to assume that the basic idea could be spelled out using the extensions of verbal predicates alone. Thomason and Stalnaker indeed suggest an intensional variant of the initial idea, based on the following kind of reasoning: Although the set of persons who drum might accidentially coincide with the set of persons who sing in this world, we know very
5 well that drumming and singing do not necessarily go hand in hand. Thus if we work in an intensional framework, there will be worlds where the extensions of "drum" and "sing" do not coincide. Let intransitive verbs denote functions from possible worlds into sets of individuals. We then get the following inequality, where the variable i is assumed to range over possible worlds. (7)
XAx.sing(-(x) * λίλχ.άηιπι,ίχ)
Assume that manner adverbs do not denote functors on verb extensions, but on verb intensions. Now it is logically possible to keep the loud singings and the loud drummings apart: (8)
The following is logically possible: F( X¡ta.sing¿(;c)) * F( λ/λχ.ΰηιπι/(;ε)) with F = [[loudly]] (wo), and Ajc.singw0(;c) =
ta.drumy^x).
Again, we can freely add further restrictions like the one that in each world, the extension of the resulting predicate must be a subset of the extension of the argument predicate (translating meaning postulate (3)). We now have the following, more careful theory: Manner adverbs are represented as functors from predicates-in-intension to predicates-in-intension. Unfortunately, the theory is too strong, because it predicts that so-called opacity effects should arise for manner adverbs. Let me elaborate what that means. Intensionality has been used to account for the fact that a certain individual might sometimes be described in various ways. Thus "the queen of Sweden" and "Sylvia" both can denote the same individual. Nevertheless, the two expressions are not always, only sometimes, freely exchangeable. In our world, where Sylvia is the queen of Sweden, the following sentences will either both be true or both false: (9) (10)
Tom kissed the queen of Sweden. Tom kissed Sylvia.
However, if we look at the two VP intensions "kiss the queen of Sweden" and "kiss Sylvia" we find that they are not equal. In worlds where Sylvia is not the queen of Sweden, we may have quite different individuals kissing one or the other. Thus we note that (11)
XiXt(KISS¿( χ , SYLVIA )) it Xikx(3y( KISS,( x,y)&
QUEEN-OF-SWEDEN¿( y ) ))
If adverbs are functors on VP intensions, we will allow for the possibility that these two predicates-in-intension are mapped onto different modified predicates. We have not excluded the possibility that the following sentences are true:
6 ( 12)
Tom tenderly kissed Sylvia, but he did not tenderly kiss the queen of Sweden. Tom loudly welcomed Sylvia, but he did not loudly welcome the queen of Sweden. Tom carefully led Sylvia to her chair, but he did not carefully lead the queen of Sweden to her chair.
This is an incorrect prediction. The above sentences can never be true. Necessarily, if Tom tenderly kisses the queen of Sweden, and Sylvia is the queen of Sweden, then Tom does tenderly kiss Sylvia. (Note that we are not talking about Tom's intentions while kissing. Of course, he might think: "I devote all tenderness in this kiss to Sylvia - 1 do not care for the queen in Sylvia." But this clearly is a separate phenomenon which has nothing to do with the presence of the adverb. Tom might also think: "I hereby kiss the queen of Sweden - not the person Sylvia" when kissing.) There are adverbs where examples like the ones above are characteristic for the adverb's semantic behaviour. The core example of this class of opaque adverbs is "intentionally". Thomason and Stalnaker mention this adverb (Thomason/Stalnaker[73]), discussing the following example. (13)
Oedipus intentionally married locaste, but did not intentionally marry his mother.
We can make the same point for the above sentences. Example (14) is a sentence which may perfectly be true. (14)
Tom intentionally kissed Sylvia, but he did not intentionally kiss the queen of Sweden.
In this case, an analysis where the adverb can distinguish between the respective predicates-inintension is adequate. Matters are different for adverbs like "carefully", "tenderly" or "recklessly". Looking at the data, we just find that the liberty to allow for constellations as in (12) is never made use of. Such sentences are never true. We could claim that this was so just by accident. However, this would be misunderstanding the meaning of manner adverbs. 2 What does a manner adverb express? If we report that (15)
Tom tenderly kissed Sylvia
this means that Tom did a certain action (namely, kiss) to a certain person (namley, Sylvia), and moreover that he did this in a certain distinguished manner ( = tenderly). Intuitively, it is not necessary to know the names of the participants to tell whether the kissing was done tenderly or not. Even if we watched the scene without being able to tell who the persons were at all, we could still be able to tell that the kissing was done tenderly, and this judgement
Moreover, we could not explain why sentence (12) implies "Sylvia is not the queen of Sweden". I am grateful to T.E. Zimmermann for pointing this out to me.
7 would not be altered by anything we are told about the two people: that the person kissed is Sylvia, that she is the queen of Sweden, that she owns a racing car which she drives every Sunday, and so on. Thus, it seems not to be an accident, but part of the meaning of manner adverbs that they are not opaque operators. Co-extensional but non-equivalent NPs may always be substituted in their scope. Defenders of the operator approach could point out that the mere logical type corresponding to manner adverbs will naturally not be restrictive enough to capture all we know about manner adverbs. Maybe we can again use meaning postulates in order to give an appropriate restriction to possible meanings of manner adverbs, like we did above. We need a requirement which expresses the following: (16)
If two predicates-in-intension Ρ and Q differ only in that extensionally equivalent, but intensionally differing nominal arguments N] and N 2 have been combined with the same predicate G, then the adverbial functor α maps both onto the same predicate, at least in our world w 0 . VPVQ(3G3NßN2(
N](W0)=N2(WO)
& P=G(NJ)
&
Q=G(N2)
a(P)(O=a(ß)(w0))) Unfortunately, meaning postulates which express this kind of generalization are impossible to formulate, for very general reasons. Zimmermann (Zimmermann[93]) shows that this kind of meaning postulate is doomed to fail. The relevant constellation in abstract terms is the one in (17), where (18) shows how this constellation is instantiated in the example under debate: (17) ( 18)
There are N P i , N P 2 such that λί(VERB(NP 1 )(i)) *&i(VERB(NP 2 )(i)) and NPi at w 0 = NP 2 at w 0 Xi(KISS(S YLVIA)(i)) * Xi(KISS(QUEEN-OF-SWEDEN)(i)) SYLVIA at w 0 = QUEEN-OF-SWEDEN at w 0
Zimmermann shows that this constellation can always arise by accident. Montagovian semantics is typically based on very rich domains of complex functions. This makes it possible to find some "abnormal" predicate for any pair of predicates-in-intension which, applied to two appropriate NPs will give one or the other of the two predicates in question: (19)
For arbitrary P, Q of type it is always possible to find a predicate V of type and x,y of type such that x{w0) = V(x) = Ρ and V(y ) = Q.
y(w0),
This fact will have the consequence that the tentative meaning postulate in (16) ensures the following: Any adverb will map all VPs to the same modified predicate-in-intension. Thus "shyly kiss Sylvia" will be the same predicate as "shyly eat a sandwich" or "shyly open the door" or "shyly grunt". This is of course more than we wanted to say.
8 Thomason and Stalnaker are aware of this problem and refrain from predicting that "shyly kiss Sylvia" is the same as "shyly kiss the queen of Sweden", once we have ensured that Sylvia is the queen of Sweden. Thus, they miss an important semantic property of manner adverbs.
1.3. The Narrow Operator Approach
The difficulties in connection with the operator approach were due to the fact that we could not pin down the correct level of distinction on the basis of extensions and intensions of verb meanings and VP meanings. Essentially, we need to distinguish different but coextensional verbs but not different but coextensional NPs. This effect could be achieved by using a syntactic trick: Manner adverbs denote functors on verb intensions. Manner adverbs combine with the verb before any nominal arguments do. Sentences (12) of the previous section will then not be a problem. The functor [[tenderly]] could in principle give different values for the arguments [[kiss Sylvia]] and [[kiss the queen of Sweden]], but this combination is assumed not to arise in the representation of (12). The adverb "tenderly" combines with the verb "kiss" in both cases. The resulting predicate combines with the co-extensional, but intensionally different NPs "Sylvia" and "the queen of Sweden", and the overall predicates will correctly be expected to be extensionally equal (in our world) but intensionally different. The feasability of a solution along these lines hinges on at least two predictions which have to be tested: First, we predict that manner adverbs never show scope effects with respect to NPs. Second, we expect that the base position of adverbs is close by the verb. Criterion 1 can be tested quite quickly, but the results are not satisfyingly clear. Criterion 2 clearly is a syntactic criterion and depends on the language under investigation. The last part of this book is devoted to word order regularities around adverbs in German. It will turn out that word order, intonation and focus structure interact in a subtle way which eventually suggestes that manner adverbs should have their position between the subject NP and object NP(s). It will, however, also turn out that the results in German can't straightforwardly be carried over other languages. Let me come back to criterion 1 and ask: Are there meaning differences between sentences where the manner adverb is in different positions? Relevant minimal pairs are given in (1) to (3) (remember that we are not interested in evaluative readings). (la) ( lb)
Alma picked each worm carefully out of the salad. Alma carefully picked each worm out of the salad.
9 (2a) (2b) (3a) (3b)
Bertha threw away all photos reluctantly. Bertha reluctantly threw all photos away. Clara ate most cookies greedily. Clara greedily ate most cookies.
Indeed there seem to be subtle meaning differences between the (a) sentences and the (b) sentences. For (la) to be true, it is enough that Alma's care was devoted to each single worm picking. We get the feeling that the care was directed, for example, towards not hurting the worm. In sentence (lb) Alma's care focusses on some overall task. While she might be indifferent with respect to the health of the single worm, she is concerned about the state of the salad. A similar contrast holds for (2a) and (2b). Examples (3a) and (3b) have been given in order to make sure that the reader does not mix up these subtle differences with a quite stronger effect, namely that the (a) word orders strongly support a focus driven interpretation of the sentence. The phenomenon in question can be demonstrated with (3). The reader may understand (3a) as not talking about the overall cookies, but only about the cookies eaten by Clara: "Most of those she ate, Clara ate greedily." In this reading, Clara may have eaten ten out of 100 cookies, and eight of these ten greedily. The reading is supported by a certain intonational hat contour on determiner and adverb, like in (3a'). The symbol / stands for a raise accent on the respective word, \ notes a fall accent. (3a')
Clara ate most/ cookies greedilyV
No such reading is available for (3b). For (3b) to be true in a situation where there were 100 cookies before, Clara must have eaten at least 51 cookies (under standard assumptions about the meaning of "most"). Thus (3b) and (3a') clearly differ in meaning. I have argued elsewhere that the pattern in (3a') involves association of the rise accented quantifier with a focus on the adverb (Eckardt[96a], [t.a.]). (3a') is essentially a different sentence than (3a). We are of course not interested in the question which word orders support additional focus effects, but only in the meanings of the sentences pure. ( Part II will take up this question for German, though, and we will ask why it is that (3a) tends to "slip into" (3a').) Therefore we must suppress the stronger meaning differences one might get for the (a) and (b) sentences and are left with the subtle distinctions described above. If we accept these as the motivation for a semantic distinction between (la) and (lb), (2a) and (2b), or (3a) and (3b), then we have a further reason do dismiss the narrow operator approach for the representation of manner adverbs. Having found the operator approach not fully satisfying, we will come to a first description of the event based approach in the next section.
10
1.4. The Event Based Approach
D.Davidson (Davidson[67]) suggested that verbs should not only denote relations between the objects that correspond to the nominal arguments of verbs in sentences. He proposes that they should have an additional "event" argument. This event stands for the action going on, the thing that happens. Sentence (10) in section 1.2., for example, will thus be represented as in (2) and not in the simpler form (1). (1) (2)
KISS( TOM, SYLVIA) KISS( TOM, SYLVIA, e)
In (2), e stands for the kissing event. The idea is that sentences with an action verb (like "kiss", in contrast to, say, "know") have to include reference to the respective "event" described in ther semantic representation. On one hand, this event can serve as a further argument for all optional modifications and relations that might further be added in the sentence. We can specify its place ("in Paris"), its time ("at noon"), the instruments used or persons benefitting ("with chrutches", "for Clara"), and also manners in which the action was performed ("greedily", "quickly",... ) Moreover the event in question is the referent for anaphors like in (3), or nominalizations like in
(4). (3) (4)
It happened at noon. Alma's singing was tender and soft.
In order to make the point precise for manner adverbs, let us for example decide to predicate manners on individuals and events. The sentence "Tom kissed Sylvia tenderly" will get a representation like in (5). (5)
KISS( TOM , SYLVIA, e ) & TENDER( TOM, e)
It is clear, though, that we are not finished just because we have invented the word "event" and written down formula (5). We don't know yet inhowfar our ontological potential has been increased by executing these two steps. Let me illustrate this claim. Looking at formula (5), an opponent might give the following argument: "You have written down e in (5), but clearly, what you mean is a temporal parameter of the verb. Even simplest semantic representations will have to take care of the fact that a sentence like "Tom kisses Sylvia' does not hold true forever, but only for a certain time. The formula v TENDER(TOM,e)' means that Tom was tender during the time of the kissing, i.e. kissed Sylvia tenderly. Therefore you claim to have a new theory, but in fact what you do essentially can be copied into an event free approach."
11 The opponent will not be right in this case. In defense of an event approach, we can easily show that adverbs do not denote properties of individuals during a certain time, but relate the individual's behaviour to something more complex. Imagine the following scene: Bertha is listening to the news and at (exactly) the same time frying doughnuts. The news are very absorbing. Bertha, listening attentively, does not devote enough care to her doughnuts, which therefore end up a bit darker than they should. This is reported by (6): (6)
Bertha carefully listened to the news. Bertha carelessly fried doughnuts (at exactly the same time).
If manners simply denoted properties of individuals at a time, the truth of (6) could not be accounted for. The opponent's principal scepticism, however, is not yet answered. As long as the notion "event" has not been filled with content it is not clear why the event based approach essentially differs from a version of the operator approach. (I will give the formulation of the event based approach in terms of "operators" in the appendix to the present chapter, in order to demonstrate that the use of "operators" is no essential point.) Chapter 2 will be devoted to give a more thorough answer to our fictive opponent. I will reconstruct a hierarchy of applications of events which give rise to ever more finegrained notions of events. The special role of manner adverbs and what I will call "semantic participance" will be discussed, showing that they require an independent notion of event. Moreover, individuating events according to the needs of adverbial semantics is not without costs. We will see examples where manner adverb representation gives rise to event individuations and -creations which would probably not exist due to our naive intuitions with respect to the term "event" (proving that it has become a technical term of semantics). Insofar manner adverbs have influence on our understanding of events. Chapter 3 will present three kinds of phenomena which seem to require event individuation by finer criteria than those discussed in chapter 2. It will be argued, though, that all three phenomena should receive alternative analyses, where the notion of event which was developed in chapter 2 can remain unchanged. Next, we will come back to the scope data of section 1.3., examples (1) to (3). We preliminarily turned down the narrow operator approach, observing that adverbs have scope with respect to NPs. It will turn out that the modellation of such (and more) scope effects involves looking not only at simple events but also at more complex objects in an event mereology. Chapter 4 will give an introduction to the mereological structure of events, as motivated by independent applications. Moreover, I will investigate the flow of information in these mereological structures, coming to the conclusion that the so-called persistency principle is not a fruitful guideline in the design of event ontology. The consequences of chapter 4 with respect to the scope data will be discussed in chapter 5. The final chapter 6 can be read (almost) independently from the rest of this book. It is devoted to word order phenomena in German,
12
having consequences for the base position of manner adverbs, but written with a broader perspective. It will lead to the conclusion that not any semantic effects that seem to be connected to word order variations can be treated as "scope phenomena".
1.5. Appendix: Operators Are Not Essential
What would happen to the operator approach of Thomason/Stalnaker[73] if we combined it with the Davidsonian idea of an event parameter of verbs? If verbs carry an event argument in addition to the nominal arguments, then predicates are no longer indistinguishable just because they happen to hold for the same set of individuals. Let us go back to our first example. Assume that we are in a world where (1) { Molly, Bo, Art} are the persons who sing { Molly, Bo, Art } are the persons who drum If we assume an extra argument which refers to "the drumming" and "the singing", these event arguments will certainly be different for the two predicates. Whatever we know or do not know yet about events, it seems plausible that no "drumming" can ever be a "singing" or vice versa. This is something we know about verbs, "drumming" never ever is conceptually the same thing as "singing", no matter who is actually drumming or singing in our world. This conceptual difference is brought out by the use of an event parameter. We will now be able to express in our above world that the "drummings" and the "singings" have nothing in common, neither extensionally nor intensionally, apart from coincidentially being done by the same persons. This is made visible by the events: (2)
SING( MOLLY, ej)& SING( BO, e2) & SING( ART , ej) DRUM( MOLLY, e4) & DRUM( BO, e5) & DRUM( ART, e6)
However, it will not change the events in question whether the persons who do the drumming or the singing are referred to by different names. The singing by Bo will remain the same singing if Bo is referred to with "the tallest boy in the syntax class", "the boy who owns a red bike", and so on. Thus, we find that the event parameter is a means to create an intermediate level of intensionality which is restricted to the verb, but not affected by nominal arguments. Now we can modify the Thomason/Stalnaker approach accordingly: (3)
Let manner adverbs be represented as functors which map binary relations between individuals and events onto binary relations between individuals and events.
13
For example, the logical form of "carefully" now is this: sets of pairs of individuals and events are mapped onto sets of pairs of individuals and events again. We may again require that the image set must be a subset of the argument set. Thus (4) [[carefully]]: { I χ climbs in e } —> { I χ carefully climbs in e } { I χ jumps in e } —> { I χ carefully jumps in e } { I χ eats an apple in e}-> { I χ carefully eats an apple in e } In that way, a relational approach can always trivially be turned into an operator approach. This observation shows that Davidson's theory and Thomason/Stalnaker's theory are not automatically different from another. Objections to the use of events can easily be produced, following the pattern of (5) in section 1.4.: "Thomason/Stalnaker failed because they used VP meanings of type . But, if VPs denote relations between times and individuals times, spaces and individuals worlds, times and individuals, then the approach would work." This kind of objection results in suggesting various reconstructions of events from other objects. It will be the task of chapter 2 to refute these reconstructions and to establish an independent level of events.
Chapter 2. Ontology
2.1. Events and Sets of Times Remember the sceptic opponent in section 1.4. who suspected that "event" might just be another term for "time during which a relation holds". This criticism was insofar correct as we find a wide variety of theories, mainly about temporal localization and modification, where the term "event" is used but times would suffice. Let me go through some cases where we write "event" but really only have to assume that verbs denote relations between objects and time intervals (see for instance Taylor[77]). Modifications like "at noon", "on May 15th 1995", "from 2.00 to 4.00" exemplify a simple way in which an event can be further qualified. In order to analyse a sentence like (1), it is enough to know that the SLEEP relation holds between Alma and a time I and that / is a subset, or equal to, the interval [2.00;4.00]. Sentence (2) will state that the marriage time J was during the day 5.15.1995. (1) (2)
Alma slept from 2.00 to 4.00. 3/( SLEEP(A,7) & /e[2.00;4.00] ) Bertha and Clyde married on May 15th 1995. Ξ7( MARRY( & J ç 15.5.1995 )
Eventually, I and J will refer to events with more than only temporal properties, but this is not yet necessary in order to explain (1) and (2). Temporal modifiers are not restricted to limiting the time during which something happened, though. Upon closer investigation it turns out that apart from the running time of events, at least two more temporal parameters are involved in sentence meanings: The time of utterance S, in order to explain the uses of Past, Present and Future tense, and a "reference time" R in order to explain the meaning of constructions like past perfect, or the progressive (see Reichenbach[47], Hinrichs[86], Klein[94]). These parameters R and S differ from the event time in that they are purely temporal "indexical" or "definite" parameters while events and their running times are normally existentially quantified. One can't identify reference times and times of occurrence. Some unwelcome consequences of this malidentification are discussed in Tichy[85], In order to clearly distinguish between indexical and indefinite temporal parameters of an utterance, many recent theories of tense and aspect are formulated on the basis of "events" which are related to "times", although it would be enough for these authors to adopt a purely temporal notion of event. Let me sketch some phenomena illustrating the interaction of R and the temporal constitution of events. Viewing events as sets of points of times, one might worry about the "overlapping intervals question": Assume that I is an interval in time which starts while Alma is sleeping, but where
15 she wakes up in the middle of I. Should SLEEP(A,/) be true or false? If we perceive I as an event, the answer to this question is quite clear: There is a subinterval J of I for which SLEEP(A,J) holds true. However, SLEEP(A,/) is clearly false for the overall intervall I. (Remember that there is no event which is a sleeping of Alma and which has the running time /·) However, we come across another observation which seems to indicate that this simple answer is wrong. Imagine that Alma slept from 2.00 to 4.00 sharp. In this situation we would hesitate to accept the following dialogue as correct: (3)
A: Did Alma sleep between 3.00 and 5.00? - B: No, she didn't.
This answer is correct only if the question is uttered with a clear focus on "between 3.00 and 5.00". However, there also is a sense of A's question for which B's answer does not provide correct information. Should we conclude that the SLEEP relation still is true for Alma and the interval [3.00;5.00], in a certain sense? No, we should not. Examples like (3) show that temporal PPs like "between 3.00 and 5.00" not only can be used to modify the event in question, but also the global temporal frame R under consideration. If the question in (3) is interpreted as a question about the time frame [3.00;5.00], namely the question whether a sleeping of Alma took place somewhere during this time, then the answer should be "yes", because there exists a sleeping interval J in this time. If the question is interpreted as a question about the nature of a sleeping event itself, namely the question if there was such an event with the property of "occuring from 3.00 to 5.00" then the answer correctly is "no". These ideas are elaborated in Klein[94] who gives a systematic survey about modifications of reference and event times. Another phenomenon which demonstrates the necessity of reference times R as a "peep hole" through which events are perceived is the use of the English progressive forms. If we assume that the use of the past tense expresses that "the event in question happened before now" and the use of the future means roughly that "the event in question happens after now" then we will expect that the simple present means that "the event in question takes (is taking?) place now". This is, however, not the case (to the great surprise of many Germans learning English). We obligatorily have to use the progressive form - which of course also exists for the past and future but which one might hope to be able to ignore in a first attempt to represent tense. The "now" seems to be a narrow peep hole through which only events-in-progress can be perceived, while the previously unnoted implicit peep holes in the past or future could be large enough to contain whole events (or their running times). Elaborate theories which spell out the intuitions underlying the above examples can for instance be found in Klein[94], Hinrichs[86], or Kamp/Reyle[93]. Especially, these authors refer to "events" while in fact only using their temporal constitution (for Kamp/Reyle[93], this
16 claim is restricted to their treatment of tense and aspect).1 Thus, focussing on matters of temporality, an opponent of events might come to the conclusion that they are a fancy way to hide away sets of points of time. Although these events will soon be too simple, the idea of handling sets of times has inspired the design of event mereology. We will discuss summation of events and a part-of relation on events in chapter 3. These relations have simple set union and the subset relation as their ancestors. The set theoretic origin not only explains why most authors assume summation to be a total operation on events, but becomes also transparent in works like Egg[94] where an axiomatic treatment of times is given (Egg[94], p.21f) which can equally well be used as an axiomatization of an event ontology.
2.2. Events as Spatio-Temporal Regions
While times may be enough to simulate events in a certain range of phenomena, they will clearly not suffice to get a notion of events of the flexibility we want. This can be demonstrated looking at quite simple examples. Already Davidson suggests that spatial PPs like "in Munich" or "on the Zugspitze" should be interpreted as event modifiers. (1) (2)
Alma sang the Bavarian anthem on the Zugspitze. Sepp prepared potato dumplings (during exactly the same time) in Munich.
If the event ei of singing and the event of preparing dumplings happen exactly during the same time, and if moreover events are reconstructed as sets of times, then ei and C2 must be identical. This means that not only the singing-event but also the dumpling-event has the property of being on the Zugspitze, and that the singing was also in Munich. This is evident nonsense, and we expected this kind of nonsense. Events clearly not only have a temporal but also a spatial dimension. This idea is laid down in the following reconstruction of events: (A)
Events are subsets of the four-dimensional space of time and locations.
Assumption (A) will intuitively lead to a reconstruction of events as little movies in time and space. Think about the event described in (1): We localize ei roughly at the place where Alma is during the time of her singing. If Alma moves, the location of the event will also move, such that (A) should be made more precise as in (A'):
Note that theories on telicity/atelicity in the tradition of Krifka[89] can also be re-written on the basis of time interval parameters of verbs. This is all the more surprising as Krifka makes very elaborate reference to the notion "event" in his works. The reconstruction is beyond the limits of this book, but is explicated in Eckardt[t.a.].
17 (A')
An event is a function / from a set of times Τ into (eucledian) space S / : Τ —» S The domain dom(/)=T is called the running time of / . The value /(f) gives, for each point t in T, the place where f occurs at that moment.
This is a very intuitive idea to reconstruct events. Nevertheless, it will be too extensional in the end, although the problematic cases are more subtle this time than in chapter 2.1. The vague part of the above definition is the spatial component. Where does an event take place? Innocently locating an event as in (3), we will probably claim that it occupies the space of Bertha. (3)
Bertha was brushing her teeth.
However, Bertha might do more than one thing, and we know enough about measuring time to be sure that these things might happen exactly during the same time. (4)
Bertha was brushing her teeth and humming the Bavarian anthem.
As far as our intuitions are concerned, we would clearly want (4) to involve two events - the humming and the brushing of the teeth. In terms of time and space, however, we will have a hard time telling a story why the brushing of the teeth took place somewhere else than the humming, if we stick to the localization of the brushing of the teeth we've given for (3). It seems that we have been too generous. Example (4) suggests that we will rather have to localize the respective events at parts of Bertha. For instance, the brushing of the teeth might exclusively take place in the mouth, whereas for the humming the vocal cord is involved. But then it is clear that we can make the example neater. (5) (6)
Bertha was brushing her teeth and licking a pimple in her mouth. Bertha was brushing her teeth and memorizing her homework.
Does the memorizing involve the brain only, but not the brushing of the teeth? But then, if you say: I saw Bertha memorizing her homework - is it her brain what you observed or would one have to admit that bodily reflexes of the memorizing have to be taken into account as well? The first observation we make about definition (A') is thus the following: (01)
It is not clear whether we have a fully independent notion of "the place where something happens" which allows us to make appropriate event distinctions.
But, why should we want to distinguish events like "Bertha humming" and "Bertha brushing her teeth" if both happened at the same place? After all, inferences like (8) from (7) will be correct:
18 (7) (8)
Bertha brushed her teeth in the bathroom, and at the same time hummed the Bavarian anthem. Bertha hummed the Bavarian anthem in the bathroom.
A first hint that an inference like (8) from (7) should not be based on such a strong premiss as brute identity of the events involved is given by the fact that we count not one, but two things: We will be allowed to say that Bertha did two things at a time, which will be surprising if events are the things we do, and if humming and brushing could literally be the same thing. Another reason to distinguish the two events is that they differ in various parameters which can be further specified. For instance, the brushing the teeth can have an INSTRUMENT but not the singing - or, if it has one, it will certainly be a different instrument than the one used for brushing the teeth. (9) (10)
Bertha brushed her teeth with a tooth brush. ?? Bertha hummed the Bavarian anthem with a tooth brush.
We thus can add a second observation with respect to the reconstruction of events through time and space. (02)
Two co-temporal co-spatial events must be distinguished if they count as two, or they are related to different objects by the same "thematic relation", or for one of them a certain "thematic relation" is defined but not for the other.
(I write "thematic relation" in quotation marks, because I will not use this term in exactly the same sense as other authors in the literature. An extended discussion of the notion "semantic participance relation" will follow in the next section, but for now, an informal use of "thematic relation" will do.) Another classical example illustrating (02) are verb pairs like "to buy'V'to sell". Clearly, the whole economic transaction of buying/selling takes place at the location of the customer, the merchant, and maybe the goods exchanged. Therefore, the events of buying and selling should count as identical event if spatio-temporal manifestations were the decisive criterion. However, we want to say that the "AGENT" of the selling is the merchant, and the "AGENT" of the buying is the customer. This intuition is manifested by observing that, for instance, manners are ascribed to different "AGENTS" in events of selling than in events of buying. (11) (12)
A hectic selling (= the merchants are hectic) A reluctant buying ( = the customers are reluctant)
Examples like (11) and (12) will be devoted more attention in the next section. The classical example for two events which occur at the same time at the same place but nevertheless are different is the one given in (13) to (15), quoted from Davidson[69].
19 (13) ( 14) (15)
The sphere rotated (e¡) and, at the very same time, got warmer (¿2). The sphere rotated quickly. The sphere heated up slowly.
Imagine that the sphere has a hole from north to south pole. Then every point of the sphere will take part in the rotation. On the other hand, all points of the sphere take part in the heating up. Thus e¡ and «2 coincide in time an space. Nevertheless one can be slow but the other quick. Being quick or slow is, however, a matter of comparison. What may be quick as a walking can be slow as a jogging, and nevertheless be the same thing-to-be-done-by-someone (intuitively). Can we claim that (14)/(15) should be analyzed as "the event e was quick-as-arotation, but slow-as-a-heating-up"? This will lead us to the generalization that rotations generally are slower than heatings. Obviously, this is not a solution. Given an "average rotation" and an "average heating up" (whatever that may be) we will not only be ignorant with respect to their relative speeds, but the speed of a rotation is principally incomparable to the speed of a heating up. We measure completely different things: The rising temperature in order to determine the speed of the heating, and the movement of, say, a point on the equator of the sphere in relation to time, in order to get the speed of the rotation. As in the above cases, different parameters are salient to determine the speed of either event. (03)
Two events e7 and «2 are distinct if a modifier α is true for one but not the other, and if this difference with respect to α is due to different parameters being specified by α for event e¡ and ¿2-
Observation (03) bears certain similarities with a criterion for event individuation as used by Parsons (Parsons[90]:156 f.). Parsons' criterion comes down to the first conjunct of the above observation (03). I will discuss in chapter 2.4. why (03) may not blindly be applied in that simpler form. Chapter 2.2. has shown the following: It was useful to relate an event to its running time, and it will be equally useful to relate an event to its spatio-temporal manifestation. Nevertheless, events are more than their spatio-temporal manifestations. First, we have no independent means to tell where an event should take place which would be even precise enough to capture only the simplest cases (see (3) and (4)). Second, we find cases where it looks implausible that we can attribute sensible, but different, spatio-temporal manifestations to two events which we want to distinguish, even without any explicit criterion for event localization. Third, we have nurtured the suspicion that a notion of "semantic participant" or "thematic role" or "parameter" should be the basis of event individuation rather than notions like time and space. The next section will be devoted to this idea.
20
2.3. Semantic Participants
2.3.1. The Polyadicity of Verbs Verbs expect certain arguments. This fact is wellknown in grammar where "verbs govern cases", but also in semantic theory where verbs are represented as η-place relations between objects (and, events). Usually, the arity of the predicate will roughly correspond to the number of grammatically expected arguments. In his classical paper "The Logical Form of Action Sentences", Davidson starts from the observation that an unlimited variety of free modifications can specify a verb, in addition to some finite number of grammatically expected arguments. Looking at sentences like (1) to (3), we note the following: If all the modifiers in question instantiate different parameters of the verbal predicate, then we do no longer know (a) how many parameters a verb (like "butter") must have and (b) whether finitely many parameters will be enough at all. ( 1) (2) (3)
Miller buttered the toast slowly. Miller buttered the toast slowly, in the bathroom. Miller buttered the toast slowly, in the bathroom, with a knife.
Davidson's solution to this dilemma, as we know well by now, consists in postulating an event parameter for the verb and assuming that all free modifiers predicate on events. Parsons[90] discusses the question more deeply, coming to the conclusion that we have to distinguish between semantically optional and obligatory parameters of the verb. Looking at semantically obligatory parameters like time and place, it is difficult to say whether these should correspond to an explicit parameter of the n-ary verbal predicate or not. In either case, we must make sure that events actually do take place at some time, at some place; and whether we say so via meaning postulates about events and their manifestation in time and space, or whether we say so by explicit existential quantification over a verb parameter, will just make a technical difference. The situation is different for semantically optional parameters. Consider the following examples: (4) (5) (6) (7)
Bertha put the ballpen into the fridge. Bertha put the ballpen into the fridge for Clara. Lucy crossed the street. Lucy crossed the street with crutches.
Sentences (5) and (7) in an event free format will indicate that the predicates denoted by "put" and "cross" have a parameter for "beneficient" and "instrument", respectively. Sentence (4) and
21 (6) don't realize these parameters. This is not dangerous in and off itself. We know other cases where a variable simply gets bound existentially, as for instance (8), which in fact means (9): (8) (9)
Clara ate. There is some place p, time t and food/such that Clara ate the f o o d / a t time t and place p.
However, not all "put"-actions have beneficients and not all crossings have instruments. In (4), Bertha might put the pen into the fridge completely without thinking. She doesn't even expect herself to remember where the pen has been put. Thus, this act of putting doesn't have a beneficient. The crossing in (6) might be done on bare legs. Are these legs the instrument of the crossing? If they were, why are they not in a situation where crutches are used in addition? Again, it is more natural to assume that the crossing in (6) was done without an instrument. Therefore, in an event free account sentences (4) and (5) will give rise to two variants of the verb "put", one with a beneficient and one without, and the same holds true for (6) and (7). Meaning postulates will have to ensure that the two "put" predicates are closely related. All this will increase the lexical apparatus of the theory. The alternative solution is, to assume that PPs like "for Clara" or "with crutches" relate an individual with an event via an appropriate semantic relation, here the relations BENEFICIENT and INSTRUMENT (BENE and INSTR for short). This not only accounts for the optionality of these participants - some events simply do have a beneficient, and others don't - but also spares us the meaning postulates which relate trivial lexical variants to one another. Thus we come to the following conclusion: (10)
If a modifier introduces an optional participant of an event, the participant must be related to the event by an appropriate participance relation. Obligatory participants can still be viewed as parameters of the verb, even if the verb thus might have more parameters than we'd expect. (We tacitly assume that there is no infinity of obligatory participants.)
Parsons argues in Parsons[90] that even participants which are apparently obligatory may turn out not to be so, if we allow dream worlds and other unreal scenarios. He claims that we can say things like (11), proving that the role "agent" is optional for a verb like "hit". I leave it to the reader inhowfar she will follow this argument. (11)
In my dream, I was hit but when I turned round I saw that there was nobody who had hit me.
22 Parsons, of course, has a reason to make this claim. He wants to argue in favour of a neoDavidsonian format where verbs denote one-place predicates of events and all arguments are added via thematic roles (Parsons's terminology). A sentence like (3) should, in this format, be represented as in (12): (12)
3 e ( B U T T E R ( e ) & AGENT(e,M) & THEME(e,T) & PLACE(e,ΒΑΤΗ) & Bx(INSTR(e^c) & KNIFED)) )
Logically speaking, this format carrys no more or less logical potential than an n-place representation format like (13): (13)
3e( BUTTER(M,T,e) & PLACE(e,BΑΤΗ) & 3x(INSTR(e,x) & KMFE(x)) )
Differences between (12) and (13) will only arise once we make further claims with respect to the thematic roles used in (12). We could, for example, integrate these roles into a theory of case assignment, claiming that the role AGENT was the semantic representation of the nominative case, in the same way as the role INSTR was more or less the translation of the preposition "with" in its instrumental use. This enterprise will be discussed in the next section. I will define my use of the term "participance relation" in contrast to traditional uses of "thematic relation" and argue that event ontology should not be misused to code grammatical information.2
2.3.2. Thematic Roles or Semantic Participants It is an old idea that there are certain stereotypes in which objects or individuals can be involved in a certain event. Many events have one participant who drives, controls or initiates the event. Many events have one participant which is affected, "done something to". Certain events involve a movement of the affected participant from one locus to another, and so on. The term "thematic role" is meant to capture these stereotype modes of participants. The idea can already found in the work of the famous Indian linguist Panini as early as 2500 years ago. Among more recent authors working on this subject are for instance Fillmore (Fillmore[68]), Gruber (Gruber[65]), Jackendoff (Jackendoff[83]), Dowty (Dowty[92]) and Levin/Rappaport (Levin/Rappaport[95]). While it is roughly clear what kind of content is to be captured by thematic roles like AGENT, THEME, EXPERIENCER and so on, the logical type of thematic roles is still open. One can view them as sets of implications which hold for certain participants of various
The only other difference I know between formats (12) and (13) is this: Format (12) can be used as the starting point for a syntax-semantic interface where nominal arguments can be combined with the event in arbitrary order (see Krifka[92]). However, it remains to be shown that we want this kind of flexible syntaxsemantics interface. Even in languages with "free" word order, like German, there are arguments in favour of a fixed underlying order of nominal arguments.
23 activities (see Chierchia/McConnell-Ginet[90], pp.377 ff.) or as labels which summarize even more subtle differences in prototypical properties of various participants of actions (Dowty[92]). However, one can also pursue the idea to model thematic roles as plain relations between individuals and events. This is in fact the idea underlying Parsons's neo-Davidsonian format which we saw in the previous section. Once one has decided to model thematic roles as relations between individuals and events, these relations may clearly give rise to new event individuations, namely in situations like the following: (14)
If two verbs VI and V2 require a certain individual or object to play distinct thematic roles R1 and R2 in the events denoted by the verbs, then the events denoted by VI and V2 must be distinct (c.g.s.).
How do we know that two verbs require a certain participant to play different roles? On the one hand, this is clearly a matter of intuition about the semantic content of the roles involved. On the other hand, though, this knowledge will also follow from further predictions -we make for bearers of one or the other role. Assuming that we have further linguistic theories that are based on thematic roles, these theories will in part determine our expectations as to what roles are involved for which verb. (These theories will also imply whether one participant may play more than one role in an event, a constellation which is usually unwelcome in practice. This was tacitly understood in observation (14).) Classically, thematic roles are used in theories of case assignment ("grammatical linking") of verbs and perceived as something like the semantic counterpart of grammatical case (Fillmore uses the term "deep case" instead of "thematic role"). The exact nature of the way from thematic roles to the case frame of a verb is still open. A direct theory will assume that cases of verb's arguments and thematic roles correspond to each other directly: If the nature of a certain type of event is known, we can derive the appropriate case frame for a verb which denotes these events, and, knowing the events denoted and the cases involved, we can deduce which NP introduces which participant in an event of that kind. A toy version of this kind of theory can be found in Parsons[90]. Indirect theories, on the other hand, assume that a completely independent mechanism takes care of the appropriate connection of NP denotations to verb parameters on the basis of grammatical information about the verb ("subject NP instantiates parameter no.l, accusative case feeds parameter no.2, etc."). They assume that the range of thematic roles together with other factors, from structural ones up to historical accident, determines the grammatical information included in the verb. Examples of this kind of theory are Dowty[92], and the work of Wunderlich and Kaufmann (Wunderlich/Kaufmann[91]). Clearly, the second type of theory is much less restricted in assumptions both about the logical type and the extension of thematic relations. Especially, there is no need to assume that thematic relations relate individuals and events and give rise to event individuations. Investigating event individuation, we only have to take a stand with respect to the first kind of approach.
24 The problem with the direct approach is that thematic roles tend to code grammatical information instead of semantical information (that is, information about the world). Parsons gives an extreme version of the direct approach in the first chapters of Parsons[90]. (He does it for the sake of simplicity; he would not defend it as a serious linguistic theory.) Individuals are related to events by relations like SUBJ, OBJ, IOBJ, etc. where SUBJ interprets nominative case, OBJ interprets accusative case, IOBJ interprets dative case and so on. Note that these relations will be the basis of a 100% correct theory of grammatical linking without any exceptions. However this theory misses one tacit requirement for sensible semantic modellation: The conceptual basis for relations like SUBJ is grammatical knowledge, not world knowledge. While, for example, a dog should be able, to roughly capture a notion like AGENT, it will (probably) not have the means to comprehend the SUBJ relation. Another observation in the same direction is that thematic roles should be more or less stable across languages, while the SUBJ relation will obviously vary between speakers of different languages. Therefore, we replace relations like IOBJ by more contentful relations like AGENT, EXPERIENCER and the like. However, it is unclear whether the task of emptying thematic roles of all grammatical content can be fulfilled in the end. Again, Parsons unwillingly provides an example for the kind of reasoning which is often involved in defining the semantic content of thematic roles. First, he lists the grammatical realizations of thematic roles (Parsons[90]:73 f.). We find, for example, that GOAL is realized as indirect object, or with "to". Some pages later, the semantic content of thematic roles is characterized by prototypical properties, or by giving examples. With respect to GOAL we are told: "GOAL is the usual term for indirect objects that are paraphrasable with 'to'" (Pasons[90]:75). Thus, the GOAL relation is nothing more than another label for a subset of the TO relation. Parsons does not attempt to give a full characterization of thematic roles, and it would be unfair to blame him for his admittedly short sketch of a role based theory of grammatical linking. It would be equally unfair to blame the whole approach for the deficiencies of one single toy version of the theory. Nevertheless, it makes one suspicious that the discussion of thematic roles has not yet converged into some simple semantic characterizations which would even do for a toy version of the theory. We have highlighted the dangers for thematic roles which are used as part of a direct theory of grammatical linking. A more thorough conceptual survey and critique can be found in Dowty[92], What are the consequences for a sound theory of event individuation? We have, to a certain extend, intuitions about the nature of events. We perceive them as a part of the world, not as part of our grammatical competence. Clearly, once "events" become a term in a linguistic theory, our intuitions will shift slightly, or we will develop firm intuitions in cases where our knowledge about "events" was vague before. Still, it is undesirable to shape events on grammatical grounds. The objects on which our semantic modelling is based should potentially be the same for all languages.
25 I will consequently reject the idea that there are thematic relations which are, on the one hand, criterial for event individuation and, on the other hand, part of a direct theory of case assignment. Note that our linguistic theory does not get any poorer if we refrain from a direct theory of case assignment. We still have the indirect approaches mentioned above, where the insights about the conceptual basis of case assignment can be implemented in a more subtle way. I will assume that an independent mechanism at the syntax-semantics interface organizes the linking of NPs to verbal parameters and that this mechanism relies on grammatical information of the vert). In contrast to thematic relations in the above sense, I will use the term semantic participance relation for relations between events and other objects which arguably can not be replaced by grammatical operations at the syntax-semantics interface. Semantic participance relations must be part of the world knowledge about events. They therefore give rise to event individuations. How do we know about a semantic participance relation, and how do we distinguish it from a thematic relation? In the following, I will show how grammatical and semantic information can be distinguished by the anaphor test.
The Anaphor Test One of Davidson's arguments in favour of the reality of events was that we can refer to them with anaphoric constructions. He assumes that the pronoun "it" in (15) takes up an event in the same way as "he" and "she" can refer to individuals. (15)
Miller did it slowly, in the bathroom, with a knife, at midnight,...
Other constructions which are assumed to be anaphoric to events are the phrases "it happened" and "it occured" as well as the German "es/das geschah" (roughly like "it occurred"). Generally, anaphoric reference to events tends to be not very good style. This might be due to the fact that events in a technical sense are not as predominant in the speaker's mind as solid concrete objects are. I will not speculate further about the deficiencies of anaphoric constructions for events. It turns out that German makes use of the "das geschah" construction more naturally than English of the paralell "this occurred" construction. As the differences are differences in style rather than in grammaticality, I will take data from either language as evidence that a certain construction is possible. How can we distinguish semantic from grammatical knowledge by looking at anaphora? Let me give an example in the realm of nominal anaphor. (16) (16a) (16b)
A husband entered the club. *He was Alma's. * Alma's he ordered a beer.... He was a non-smoker.
26 What is the nature of the combination of "husband" with the genetive NP which introduces the wife of the husband? The noun "husband" is a relational noun, offering a place for the wife who can be introduced by a genitive, like in "Alma's husband". Example (16a) shows that a purely semantic treatment does not make sense. A purely semantic treatment could have the following shape (mimicking something like a "neo-davidsonian introduction of wives of husbands"): Claim that there is a relation SPOUSE(*,y) which is denoted by a certain use of the genetive case. Claim that all male individuals with the property "HUSBAND(*)" are related by SPOUSE to their wives. Claim that the genetive NP means that the individual in question (=Alma) stands in the SPOUSE-relation to an appropriate husband-individual. As mentioned before, this can't be the nature of the interpretation of genetive NPs like in "Alma's husband". Otherwise, the sentences in (16a) would have all information necessary in order to interpret Alma as the wife of the referent of "a husband", let it be Tom. We know that Tom is a husband, and if GEN denoted spouses, Alma should be identified as the wife. However, (16a) are bad sentences. This indicates that grammatical information plays a role in interpreting the genetive NP, grammatical information of the word "husband" which is only available if the actual word "husband" occurs in the sentence. If we pick up the husband later, we retain all world knowledge about husbands, or this special husband, but the grammatical information of the word "husband" is lost. Matters are different in (16b). (16b) also involves certain restrictions, but this time, the information involved is semantical, not grammatical. The predicate "non-smoker" is arguably sortally restricted to adult human beings. Because we know that husbands also are human adults, we know that the referent of "a husband", for example Tom, is human and adult from (16). This kind of knowledge, necessary for the felicitous application of "non-smoker", is not grammatical but world knowledge: It is still available for (16b) which is a perfectly natural sentence. The checking of sortal restrictions is a semantic process, not a grammatical one. Thus, it is possible to separate semantic and grammatical information by looking at anaphors. Semantic properties of objects are still available when we refer to these objects with anaphors, while grammatical knowledge is only available in the sentence where a certain word is actually processed. The proper combination of the genetive NP with the relational predicate "husband" thus is a grammatical process, while sortal information about non-smokers being adults involves semantic properties. (We can argue that (16a) fails due to the fact that the genetive case is not grammatically licenced, although the meaning of the construction might well be derivable in the way sketched above. I acknowledge this argument. What will be used in the following is that (16b) does not involve any kind of grammatical process like the ones that make (16a) fail.) Looking at event anaphora, we note that the same distinction arises for constructions which specify an event. Some constructions are perfectly good with anaphors, while other constructions are restricted to the sentence where the verb (or its nominalization) occurs which introduces the event in question. Thus, the first type of construction can only rely on semantic
27 properties, while the second involves grammatical information about the verb. The German sentences translate the corresponding English examples, unless stated otherwise. (17) (17a)
(18)
(19) (20)
Alma sang the Bavarian anthem. She did it nervously. It was done nervously. It happened nervously. She did it quickly. It happened quickly. Es geschah leise und gefühlvoll, ("quietly", "sensitively") It happened in the opera, on Saturday night. She did it for Harry. Es geschah für Harry. She did it with a microphone. Sie tat es mit einem Mikrofon. Alma öffnete das Fenster. Es geschah nur für wenige Minuten, aber trotzdem entflog Willi Wellensittich. (Alma opened the window. "It occured for few minutes only", but still, Billy the Budgie escaped.) Bertha ate. *She did it apples. Bertha aß. *Sie tat es [einen Apfel](akk)· Bertha carried a cake. *She did it into the fridge. ( "it" = the carrying ) *It occured into the fridge.
(17) to (18) show that the following constructions only rely on semantical information: manners ("nervously", "quickly", "quietly",...) temporal modifiers local modifiers, instruments and beneficients They can be attributed to an event even if the event is refered to with an anaphor. On the other hand, the following modifications involve grammatical factors: objects directional PPs They can only be attributed to the event in the sentence where the event is introduced by the respective verb. This means that there is no need for a mechanism which is based on independent semantic notions. (I will point out the difference between "no need" and "not possible" with respect to directional PPs in section 2.3.7. If the modifications in (17) and (18) can be demonstrated to depend on any further factors, these can only be semantic factors, not grammatical ones. If we, for example, find that "quickly" depends on the temporal duration of an event (in comparison to the duration of others) we can't claim that a verb like "sing" carries a "grammatical parameter: speed" and that "quickly" combines with the verb on the basis of this
28 grammatical information of the verb. This suggestion may look absurd, but manner adverbs provide a more convincing example: I will demonstrate in section 2.3.3 that certain manner adverbs rely on a notion VOLI much similar to AGENT. 3 Example (17) demonstrates that this notion of VOLI, in contrast to AGENT as a thematic role, is a semantic property which can't be neglected in event individuation. Let me summarize the results of this section. - A construction is based on grammatical information (licensing; driving the interpretation) of constituent X iff the referent of X can't be introduced with an anaphor. - A construction is only based on semantic information about the referent of a constituent X iff X can be replaced by an anaphor refering to the same object. Let me finally add two types of constructions which do not give rise to a test for semantic properties. Firstly, nominalization, in contrast to anaphoric reference, is a way to refer to events which carries grammatical information. (21a) shows that nominalized "eat" can combine with a direct object "apples" which comes as an "of-PP. This observation confirms the arguments of Zucchi (Zucchi[95]) in favour of a grammatical argument-relation between nominalized verb and various PPs ("by ...", "of ...",...). We conclude that nominalization doesn't give rise to a test for semantic relations. Another case which is not appropriate to test semantic relations is the "namely..." construction (German: "und zwar...") as in (21b): (21a) (21b)
Alma's eating of an apple... Alma ate, namely an apple. Alma aß, und zwar einen Apfel.
This kind of construction is transparent for grammatical information in the main clause. The next sections will be devoted to investigate various constructions which are based on semantic information. We will trace some semantic participance relations which are crucial in these constructions, and will collect cases where events must differ because one but not the other stands in the relevant participance relation to certain objects. We will also devote a section to a case where the elegance of the syntax-semantic interface component might suggest the use of a semantic participance relation (having consequences for event individuation) while the anaphor test shows that this way to proceed is not compulsory.
The label VOLI is short for "volitional consent". It is borrowed from Wyner[94] who in turn takes it from Dowty[92], I will use it to distinguish the relation from the thematic role AGENT, although I fear that the philosophy underlying my use of VOLI, Wyner's and Dowty's are all three different.
29 2.3.3. Manner Adverbs and the VOLI relation Wyner (Wyner[95]) demonstrates that a certain class of manner adverbs requires that the modified event has a participant which would formerly have been called the AGENT. Among these are "carefully", "reluctantly", "shyly", and other adverbs which specify the attitude of a certain participant in the event. Their behaviour is characterized by the examples in (22) to (24): (22) (23) (24)
Alma carefully/shyly/reluctantly hit Emily. *The stone carefully/shyly/reluctantly hit Emily. *Alma carefully/shyly/reluctantly impressed Emily.
According to Wyner, the contrast between (22) and (23) is due to the fact that "hit" can be used in a reading where the subject is understood to hit the object as a passive projectile (=(23)), that is, as a nonvolitional participant (->VOLI(jc,e)), but also in a reading where the subject does something willingly in order to hit the object (VOLI(x,e), = (22)). If we assume that the adverbs in (22) can only modify the behaviour of volitional participants, we can explain (22)/(23). Of course only sentient beings can show volition at all. On the basis of (22)/(23) alone it would thus be possible that these adverbs are only felicitous if the participant to be modified is human, or living. Sentence (24) shows that this is not the case. Although "Alma" is as human and living as in (22), her involvement in "impressing" lacks the crucial volitional component which is necessary to exhibit care, shyness or reluctance. (24) is marginally possible if "impress" is understood as "doing something in order to impress Emily". This confirms Wyner's hypothesis, because a reinterpretation of this kind introduces exactly the lacking VOLI component. We have tested in the previous section that manner adverbs of this kind can apply to event anaphors. Therefore, the VOLI relation is a semantic relation, not a grammatical one. The VOLI relation also provides new arguments in a classically problematic case of event individuation: the buy-sell pair. We have mentioned the case before. An event of buying is one where a buyer χ gives money m to a seller y and receives a good g in return. An event of selling is one where a seller y gets money m from a buyer χ and gives a good g in return. Nothing can get sold without there being someone who buys it, and nothing can be bought without there being someone who sells. The buying and the selling occur at the same time and place. Are they the same? Of course, the grammatical properties of the verbs differ, "sell" realizes the seller y as a subject, "buy" the buyer x. Direct linking theories will thus conclude that the AGENT of the buying event is different from the AGENT from the selling event and that therefore these two events must be distinct. However, we have seen that the entire enterprise of direct linking is not inevitable, and that the thematic roles in use are in danger of coding notions like "subject" rather than semantic notions. Therefore, this argument in favour of distinguishing events of buying from events of selling can be attacked by pointing out that the AGENT relation is not conclusive for event individuation at all. However, the VOLI relation has been proved to be a semantic relation. What is the behaviour of "buy" and "sell" when it comes to adverbial modification?
30 The data are quite clear. If a manner adverb is combined with a selling event, it is always and only the seller y who's attitude is described. For buying events, matters are the other way round. (25) (26)
Alma reluctantly sold the red bike to Maria. Maria eagerly bought the red bike from Alma.
Do (25) and (26) necessarily mean that the buying e and the selling e ' are distinct? It might be possible that we face one event, that this event has two volitional participants and that the adverbial modification is possible because both Maria and Alma are volitionally involved, but attributed on syntactic grounds, for example by looking for the grammatical subject. This solution is not attractive at least for two reasons. First, the manner adverb can find the correct participant even in anaphoric examples. (27) (28)
Alma sold the red bike. It happened only reluctantly. Alma verkaufte das rote Fahrrad. Es geschah nur zögernd. Maria bought the red bike. It happened hastily/ with big haste. Maria kaufte das rote Fahrrad. Es geschah ganz hastig.
We have argued above that grammatical information like "subjects" are not available in anaphoric examples. The only basis for a correct application of the adverbs in (27) and (28) is the information about volitional participants. The sentences in (27) and (28) are stably understood as expressing Alma to be reluctant, and Maria to be hasty. We do not develop the idea that the one who sold the bike was reluctant, in (27), or get the feeling that (27) is not interpretable because we lack a subject. Therefore, we can conclude three things: - Events of buying are distinct from events of selling. - Events of buying have the buyer as the VOLI participant. Events of selling have the seller as the VOLI participant. - Manner adverbs which specify the behaviour of a certain participant choose this participant on the basis of the VOLI relation. This result matches well with the views developed by Wyner (Wyner[95]). Wyner's starting point are cases of adverb orientation in passive sentences. He also propagates a purely semantic theory of adverb interpretation, thus confirming our conclusions from the anaphor test. Interestingly, his data give rise to more event individuations than only the buy-sell example. I will therefore include a discussion of his theory. Let us first review the data in question. It is wellknown that manner adverbs in English passive sentences allow more readings than adverbs in active sentences. The same data can be reproduced in English and German with the lassen/let construction. Examples are given below.
31 (29) (30) (31) (32)
Dr. Spöck reluctantly examined Rose, (one reading: Dr. Spöck reluctant) Rose was reluctantly examined by Dr. Spöck. (two readings: (a): Rose reluctant; (b): Dr. Spöck reluctant) Rose let Dr.Spock examine herself reluctantly. (two readings: (a): Rose reluctant; (b): Dr. Spöck reluctant) Rosa ließ sich von Dr. Spöck zögernd/gründlich untersuchen. Rose let herself by Dr.Spock
reluctantly/thoroughly
examine
(two readings: (a): Rose reluctant; (b): Dr. Spöck reluctant) Wyner critizises previous theories which treat the attribution of manner adverbs by a mixed syntacto/semantical criterion to the following end: - Manner adverbs modify the AGENT, or the SUBJECT (if the event involved has an agent somewhere else). or, in a slightly different version: - Manner adverbs modify the subject, and: the passive is derived in such a way that the active-subject still counts as a subject in some sense. In special, Wyner argues that the position of the manner adverb in the sentence is not decisive for the reading of the passive sentence. This disqualifies all those accounts which see a correlation between the position of the adverb relative to one or the other "subject NP" and the adverb's orientation (a correlation which could be turned into an argument in favour of orientation towards certain syntactic positions in passive sentences). Wyner's claim that readings arise position-independent is fully confirmed by the German data with "lassen". The examples below show that two readings are possible in all word orders of "lassen" sentences. I chose to use two adverbs, "gründlich" (thoroughly) and "widerwillig" (=reluctantly) in the "lassen"-construction. Due to the semantics of "lassen" (=let), "gründlich" is preferably understood to describe the attitude of Dr. Spöck, while "widerwillig" makes more sense for the subject of "lassen", Rosa. We observe that any word order can express Rosa's reluctance, and Dr. Spock's carefulness. (33)
Rosa ließ sich von Dr. Spöck widerwillig/gründlich untersuchen. Rosa
(34)
let herself by Dr.Spock
reluctantly/thoroughly
examine
preferred readings: Rosa - reluctant; Dr.Spock's examination: thorough. Rosa ließ sich widerwillig/gründlich von Dr. Spöck untersuchen. Rosa
let herself
reluctantly/thoroughly by Dr.Spock
examine
preferred readings: Rosa - reluctant; Dr.Spock's examination: thorough.
32 Wyner suggests that oriented manner adverbs are binary relations between an event and the volitional participant of the event, and are thus interpreted completely independent from syntax: The meaning of "reluctantly" roughly can be paraphrased as "whoever was VOLITIONAL in e also was reluctant about e". Wyner claims that the passive construction can be based on two versions of the verb "be", let me call them "agentive be" and "simple be". Agentive "be" ascribes the subject volitional participance in the event in question. Thus, (35a) means (35b). (35a) (35b)
Rosa was(agentive) examined by Dr.Spock. Rosa gave volitional consent in event e, which consisted of Dr. Spöck examining her.
On the basis of (35b), an adverb like "reluctantly" will find Rosa as a volitional participant in the event, and will be able to modify her behaviour. In contrast, the simple verb "be" does not ascribe volitionality. Thus, the event in (35c) can stand in the VOLITIONAL relation to the active-subject (Dr. Spöck), giving rise to manner attribution to this participant. (35c)
Rosa was( s i mp i e ) examined by Dr. Spöck.
Wyner's approach has the disadvantage that, looking closely into it, we are in danger to get too many volitionally involved participants, a danger which can only very unelegantly be avoided. Let me show why. Wyner is not fully explicit with respect to the question whether it is part of the meaning of verbs like "examine" to have volitional subjects in the active sentence or not. Does it make sense to assume that the verb ("examine") always denotes events where the individual denoted by the subject in active sentences is ascribed VOLITION in the event? This individual will of course remain volitionally involved even if we report the event in a passive sentence. Thus, if we combine the verb ("examine") with agentive "be" we will refer to events which - factually have two volitional participants, the one denoted by the subject of the active sentence ("Dr.Spock") and the one denoted by the subject of agentive "be" ("Rosa"). But, if an event e in fact has two volitional participants, and if manner adverbs simply ask for "anyone who shows volition", then these events should also give rise to two interpretations of active sentences. An active sentence might describe an event where Dr. Spöck volitionally examines Rosa, and where Rosa moreover has shown volitional consent. In this case, the adverb "reluctantly" should be able to modify either one, even in the active sentence. This is a wrong prediction. If we try to avoid this by making the general assumption that every event may only have one volitional participant, we get new trouble. Under this assumption the verb ("examine") may no longer ascribe volition to subjects because combination with agentive "be" will describe events with two volitional participants. If we assume that no event has more than one volitionally involved participant, then passive sentences will describe impossible events - and thus be predicted to be always false. To avoid this prediction, we have two possibilities. Either
33 we claim that the verb ("examine") does not mention volition at all, such that there might be examinations where no participant shows any volition. This, however, means that VOLITIONAL has no intuitive semantic content at all. Or we assume that the verb ("examine") comes in two versions: an active version with volitional agentive subject, and a passive version only to be combined with agentive "be", which is used to describe events like the one in (35a). Wyner, using the example "hit", can in fact explore this latter strategy which is a natural assumption for hitting events. Taken as a general solution, it results in - doubling the verb lexicon, - claiming that these extra verbs only occur in passive sentences, - claiming that events described by passive sentences are different from events described in active sentences. Facing these disadvantages of Wyner's original account, I want to propagate a slight modified solution which is much in the spirit of his original work but seems to capture the data more naturally. I assume that verbs like (agentive) passive "be" and "let" in English, as well as "lassen" in German (but not the passive auxiliary "werden" in German) introduce their own event in addition to the one refered to by the main verb. This event consists in giving volitional consent to the main event in question. The main verb describes an event where the activesubject denotes the volitional participant. Manner adverbs can choose to modify either of these two events. Active sentences never refer to events of giving volitional consent and therefore can only get one reading. Sample representations are given in (36). (36)
Rosa was(agentive) reluctantly examined by Dr.Spock.
= Rosa liess sich zögernd von Dr.Spock untersuchen. (36b) reading 1: 3 e / 3 e 2 ( EXAMINE(SP, R, e2) & VOLI(SP,e 2 ) & AGREE(R,e 2 ,*>/) & (36c)
VOLI(R,e/) & RELUCTANT^/) ) reading 2: 3e]3e2( EXAMINE(SP, R, e2) VOLI(R,ey) & RELUCTANTE) )
& VOLI(SP,e 2 ) & AGREE(R,e 2 ,ey) &
Let me summarize: The anaphor test has shown that manner adverbs are interpreted on the basis of semantic information. As shown by Wyner, the central notion for the interpretation of manner adverbs is the volitional participant. The data so far suggest that all events should have at most one volitional participant - we found no case where it made sense to attribute ambiguities to the presence of two volitionally involved participants. Instead, the most elegant analysis led us to assume events of volitional consent, which are introduced by the embedding verbs "be", "let" and "lassen" (in German). Thus, the VOLITIONAL relation gives rise to new event individuations.
34 Conveniently, this theory allows to represent manners as unary predicates of events, as in (36). The formula RELUCTANT(e) reads as "the VOLITIONAL participant of e was reluctant about e".
2.3.4. Beneficients and Instruments Section 2.3.1 has shown that the most elegant way in order to introduce optional participants like instruments and beneficients is by linking them with relations INSTR and BENEF to the events in question. This means that the proper interpretation of PPs like "with a knife" or "for Bertha" is not left to an independent linking mechanism which looks at lexical information provided by the verb. Instead we interpret the respective PPs as directly refering to the relations in question. This way to proceed is confirmed by the observation that these PPs pass the anaphora test. We can say things like: (37) (38)
Rosa sang a song. She did it for Alma./ This happened for Alma./... Rosa sang ein Lied. Sie tat es für Alma./ Es geschah für Alma.... Alma cut the bread. She did it with a knife. Nun wurde das Schaf geschlachtet. Das geschah mit einem Messer. Now was
the sheep
slaughtered
This happened with
a
knife
Examples (37) and (38) show that it is not only most convenient to link instruments and beneficients to events by using participance relations, but that moreover no further grammatical "licensing" processes operate which would rule out anaphoric sentences like the ones above. Thus, INSTR and BENEF are semantic participance relations. This provides evidence in favour of distinguishing another wellknown puzzling case of event (non-) identity. Do sentences (39) and (40) refer to different events or to the same? (39) (40)
Alma paid the trousers (with a cheque / but not with a ball pen) Alma signed the cheque (with a ball pen)
The nonequivalent behaviour with respect to instruments indicates that these events are different.
35 2.3.5. Telicity and Atelicity Investigation of aspect and aktionsart provides another range of data where the number of events involved is notoriously unclear. I will restrict the discussion here to observations concerning the correct distribution of the temporal prepositional phrases "in χ time" and "for χ time". The former PP is usually taken to be a test for telicity, the latter as an indication of atelicity, while theories do not agree whether these notions refer to "events", to "verbs", to "sentences" or to something else. The crucial contrast is illustrated in (41) and (42); this kind of test goes back to Vendler (Vendler[67]). (41 ) (42)
Clara jogged for 15 minutes / *in 15 minutes Clara jogged to the station in 15 minutes / *for 15 minutes
Sentences (41) and (42) illustrate various things: First, it makes sense to assume that temporal PPs measure the length of an event. Second, measuring with "in" is not possible in the same cases as measuring with "for". In fact, the leading doctrine holds that the PPs are mutually exclusive. 4 Third, it can't be a question of the verb alone which kind of PP is selected, as both (41) and (42) are based on "jog". And, fourth, if the notions "telic" and "atelic" describe events, then (41) and (42) must refer to different events which look overwhelmingly alike in space and time. Do (41) and (42) necessarily refer to different events, a telic one where Clara jogs to the station, and an atelic one where Clara simply jogs? Are sentences like (41)/(42) accepted or rejected on mere grammatical grounds? Or can we explain the contrast in question in a different way? In fact, there is a theory where a doubling of events into a telic and an atelic variant can be avoided: the approach developed by Krifka (Krifka[89], Krifka[92], Krifka[95], but also Eberle[t.a.]). Krifka assumes that PPs combine with sets of events (described by a sentence) and that the properties of these sets of events determine whether an in-PP or a for-PP will be appropriate. These properties of sets are based on the part-whole structure of events, a structure to be investigated in the next chapter. I will only give an intuitive account of the underlying idea. The phrase "Clara jogged" is assumed to denote a set of events where Clara jogs. If Clara jogs for a certain time t and t' is a subinterval of t, then Clara also will jog at t'. Moreover, if Clara jogs at t and also jogs at t', she will jog throughout the overall time f u r T h e phrase "Clara jogged" thus describes events, their parts and sums thereof. In Krifka's terminology, the set denoted by the phrase "Clara jogged" is "homogeneous". The phrase "Clara jogged to the station" is different. Not all parts of Clara jogging to the station are joggings to the station. The sum of two joggings-to-the-station are not a jogging-to-the-station, either. The phrase refers to isolated events, but not to their parts or sums. Sets like this are called "quantized".
This is one main claim of Vendler[67], which was adopted by later authors - for instance those to be quoted below. For a different perspective see Mori/Löbner/Micha[93],
36
Quantized sets of events are telic and must be modified with "in"-PPs. Homogeneous sets are atelic and must be modified with "for"-PPs. Note that this explanation, although looking at a great number of events, does not have to postulate that we have anything like "processes" versus "events" which take place at the same time and location but are nevertheless different (see Bach[86], Pinon[95]). This account is simply based on the assumption that we have to think very carefully about which phrases describe which events. However, it is worth mentioning that in-PPs and for-PPs pass the anaphora test very well. It is possible to take up an event with an anaphor and say something about its duration. (43) (44) (45)
Clara danced. She did it for 15 minutes / *in 15 minutes. Next, the band played. This happened for 15 minutes. Clara jogged to the station. She did it / it happened in 15 minutes.
Sentences (43) to (45) show that the temporal PP can deduce even from the anaphor whether it faces a telic or atelic "construction" - whoever may be telic or atelic. This observation favours the view that "telicity" and "atelicity" are properties of events. We could claim that events, being picked up by anaphors, still have the properties of being TELIC and ATELIC and therefore can determine themselves whether an in-PP or a for-PP should be used. Adopting this view, we will conclude that examples (41) and (42) refer to different events, as do many sentence pairs of the same type ("eat", "eat an apple"; "drink wine", "drink a glass of wine"; "run", "run the race", ... ). Can we save the more parsimonious account? Perhaps we can. Note that Krifka's theory, although being more complex, is entirely based on semantics. All we have to do is looking at certain sets of events which arise during the computation of a sentence. Assume that these sets of events are still accessible when we compute the next sentence. Then we can rely on deriving the appropriate kind of PP on the basis of the set of events which was used to introduce the event in question in the previous sentence. Sentences like the following indicate that there are more cases where properties of individuals and objects in discourse are accessible for later use. (46) (47)
Bertha owns a mouse. It is very big. (Understood: big-for-a-mouse) Alma is a professor. She earns quite well. (Understood: earn-well as a professor. Alma might at the same time be janitor of the faculty buiding, earning little in this second job.)
Example (47) has been treated by Landmann (Landmann[89a], [89b]). Landmann in fact uses a distinction similar to the distinction of TELIC and ATELIC events in the domain of individuals, claiming that there must be a copy of Alma "Alma-as-a-professor" different from "Alma-as-ajanitor". While the use of telic and atelic objects has a firm tradition in the semantics of time and aspect, Landmann's suggestion has not been generally adopted when talking about individuals.
37 I do not know of any formal alternative approach being based on accessible properties of objects. It will be a matter of future investigation to find arguments in favour of one or the other perspective as the ultimately best one. What kind of theory do we rule out, then? The anaphora test rules out all theories which base the appropriate combination of PPs with sentences on a feature matching mechanism. An example for this kind of theory is the one developed by Verkuyl (Verkuyl[93]) which is based on a mere checking of features which are attached to nodes of the sentence structure.
2.3.6. Post States Let us come to another temporally relevant participant of events. The prepositional phrase "for " in the previous section was used in the sense that it gave the duration of the event described in the sentence. However, we find uses of "for η time" (German "n time lang", "für η time") which do not measure the event in the sentence but a state following this event. Sentence (48) does not mean that it took Bertha ten minutes to open the window, but that the window remained open for ten minutes and then got closed by Bertha, again.5 (48)
Bertha opened the window for ten minutes. Bertha öffnete das Fenster für zehn Minuten.
Verbs like "open" denote events which give access to a post state s such that the duration of s can be measured with a "for"-PP. It can be argued that the existence of s, and the respective interpretation of the PP, is not a question of world knowledge. The existence of a post state is part of the meaning of the respective verbs. We can conclude this because we find verbs which should, by all that we know, have a sensible post state reading but are not interpretable in that way. In (49), we use the verb "arrive". Our world knowledge will tell us that if Bertha arrives somewhere, then she is at that location at least for a very short time. Nevertheless, the PP can't state that Bertha's stay in Paris lasted for two days. (49)
*Bertha arrived in Paris for two days.
Thus it is extra information of the verb, or the event, that an accessible post state exists. Does the "for"-PP get interpreted on grammatical or ontological grounds? The anaphora test shows that the interpretation is made on ontological grounds. (50)
Bertha öffnete das Fenster. Es geschah nur für 10 Minuten, trotzdem entflog Willi Bertha opened the window.
It happened only for 10 Minutes,
nevertheless fled
Billy
Wellensittich. the budgie. '
Thus, not any time interval where the window is open, which follows an opening of the window, counts as appropriate post state. This is in line with the observations below: Post states are more than just intervals during which something is true.
38 We thus find certain events which are related to post states by a semantic relation POST. The anaphoric constructions in the second sentence in (50) takes up the event introduced in the first sentence, and post-state "for"-PPs have a meaning which can be paraphrased as: "the object s which is related by POST to e has duration n". Interestingly, post states are more than purely temporal objects, in the same way as events are not simply intervals. We find constructions where post states, or resultant states, can be further modified adverbially. Consider examples (51) and (52). (51 ) (52)
Alma dressed graciously. Bertha painted the chair strangely.
Sentence (51) can either mean that Alma moved and acted graciously while dressing, although she was dressed quite shabbily afterwards. Or, it can mean that Alma's dress was gracious while it doesn't matter how Alma moved while dressing. Example (52) shows the same kind of ambiguity. Bertha can either paint the chair in a strange manner, for example by holding the brush between her teeth. Or, she can paint the chair in a strange colour or pattern, while acting quite conventionally during the painting itself. Thus, these examples show two readings, a manner reading and a result reading. These two readings are independent. They can't simply be done away claiming that different factors were decisive for different events to be "gracious" or "strange". We need two logically independent representations, and I suggest that the manner reading is one where the adverb is predicated on the event, while resultative readings predicate on post states. This is exemplified in (51a), (51b). (51 a) (5 lb)
3e( DRESS(A,e) & GRACIOUS(e) ) (manner reading) 3e3s( DRESS(A,e) & POST(e,j) & GRACIOUS(s) ) (result reading)
Note that more post states are accessible for temporal modification than for resultative adverbials. While it makes sense, for example, to "go to the market for 30 minutes", it is conceptually odd to "go to the market graciously" in the sense that the stay at the market is gracious. It doesn't seem necessary, though, to derive these differences from different types of post state objects. I simply assume that more post states have a duration than we find post states for which properties like "grace" or "strangeness" conceptually make sense.
39 2.3.7. Moved Objects - Grammatical or Semantical Information? The spatio-temporal manifestation of events invites an interpretation of directional PPs on the basis of this movie-like object. A PP like "into the kitchen" in a sentence like (53) seems to state that the event's spatial manifestation at the end of its running time is located in the kitchen: (53)
Otto went into the kitchen.
However, sometimes it is not the overall event, but only certain affected objects which end up in the location indicated by the PP. Moreover, the same scenario can be described in various different ways which carry different implications as to who or what has to end up where. (54) (55)
Otto went into the ice house with a cake in his hand. Otto carried a cake into the ice house.
(54) implies that Otto ends up in the ice house, where the cake will probably also end up there as a side effect. (55) on the other hand only implies that the cake ends up in the ice house, while Otto may remain outside. How do we know which objects are the ones moved around in an event? Extending the idea of semantic participants of the previous section, we might make the following claim: Certain events are related to objects by the relation "MOVED OBJECT". The moved object in an event is always the object which is located by directional PPs at the beginning and end of the event in question. The PP "into the ice house" means that the moved object of the event ends up in the ice house. The PP "out of the cellar" means that the moved object started in the cellar, and so on. Following this idea, we will conclude that the events in (54) and (55) are different: The event in (54) has Otto as its moved object, while the event in (55) has the cake as its moved object. This is a sensible suggestion. Is it compulsory? MOVED-OBJECT is not a compulsory relation between individuals and events. It is not, because directional PPs fail the anaphor test. Sentences like (56) are not well-formed, as we already noted in section 2.2.2. (56)
Otto carried a cake. *He did it into the ice house. *It occured into the ice house.
The grammatical presence of the verb is still necessary in order to interpret directional PPs. Therefore, an analysis via MOVED OBJECTS is technically possible, but not compulsory. A direct grammatical link between the PP and the respective object could be given, leaving the semantic model without the MOVED OBJECT relation. However note that no simple grammatical characterization of MOVED OBJECTS is in sight. Sentences (57) to (61) show all kinds of moved objects which fit no common pattern.
40 (57) (58) (59) (60) (61)
Otto went with a cake in his hand into the ice house. (= subject) Otto carried a cake into the ice house. (= direct object) Alma helped Otto into the bus. (= German: dative object; English?) Alma accompanied Otto into the bus. (= subject, in spite of presence of object) Alma spat into the ice house. (= moved object not lexically expressed; moved object Φ subject )
A grammatical solution will therefore at least involve the following ingredients: The moved object is always a compulsory argument of the verb, but occupies no fixed position (as far as I can see). The verb contains lexical information as to which of its arguments is the moved object. Directional PPs can ask for this information in the combination process, and thereby get access to the correct object. These assumptions will, as far as I can see, result in processes at the syntax-semantic interface which are at least somewhat unorthodox. Nevertheless, I know of no conclusive argument which stands against the grammatical solution, and proves that the semantic solution, including the relation MOVED OBJECT, is the only possible one. Thus, we leave it to the reader's taste which view s/he wants to adopt.
2.3.8. Subtle Parameters At the end of this section, I want to come back to an example which was among the first ones we saw in favour of a richer event distinction than one based on time and space. (14) ( 15)
The sphere rotated quickly. The sphere heated up slowly.
We took the different speed values of the rotation and the heating as evidence that the heating.and the rotating must be different events (and thus, events must be more than regions in space and time). We argued in section 2.1 that it isn't a matter of attributivity that (14) and (15) can possibly be true together, because this would lead to the generalization that average rotations are slower than average heatings. This is not only wrong, but it is even uncomprehensible. The parameters that determine the speed of a heating are quite different from those which determine the speed of a rotation. As long as we regard "quickly" and "slowly" as simple unanalyzed properties, this need not bother us. Things change if we attempt to give a lexical fine analysis of what it means to be "quick". At first sight, the meaning of "quickly" seems easy to grasp, in comparison to, say, "reluctantly" or "ardently". The adverb "quickly" seems to say something about duration in time. Looking at examples like (63), we might conjecture that "quickly" means "event takes shorter than comparable events".
41 (63)
Bertha sang the Bavarian anthem quickly.
Thus we look at the average time it takes to sing the Bavarian anthem, and if Bertha performed in a shorter time, (63) is true. It was already noted by Cresswell (Cresswell[86]) that this notion of "quickly" is too simple. It is not always the event described by the overall sentence which is the event of which we measure the duration in order to say something about the speed. Consider example (14). It is of no relevance at all how long the sphere rotates in order to determine whether it was a quick rotation. A rotation of 5 seconds may be a slow rotation, while a rotation over 5 hours may be a quick rotation. A relevant event for rotation speed might be this: A point on the equator of the sphere making the round once. If we assume rotation at constant speed and take this kind of test event, "quickly" and "slowly" for rotations means that the test event took shorter/longer that the average. This meaning is independent of the question how long the objects rotate. Measuring the speed of a quick rise of temperature is based on yet another kind of event. No matter how long an object's temperature rises, what counts is the speed of a subevent of the following kind: "the object's temperature rising for a fixed difference on the temperature scale, for example: for one degree Celsius". If this subevent is shorter in e than in the average heating event, we talk about a quick heating. Generally, "quickly" has the meaning "a certain speed-sub-event takes shorter in this event than per average". The "speed-sub-event" may differ for all types of events. Is there any hope to find a clever uniform overall characterization of the speed-sub-event which allows us to look at an event, moving through time and space, and determine what parameters have to be measured in order to know its speed? The example of the sphere rotating and heating up indicates that there isn't. We can't confront someone with this sphere and ask, well what is the speed? Or: is this quick? Although "quickly" in fact only refers to duration, we have to know which of the events that happen at this very place and time is refered to, in order to know what the speed-event must be. (Once we know the relevant speed-event, "quickly" is of course a pure matter of the stop-watch.) Thus if we want to do a lexical fine analysis of "quickly" in terms of comparing durations, we will be formally allowed to distinguish the heating event and the rotation event in that the sub-events relevant to measure the speed of either one are different: HEAT(e;) & SPEED(c/,er) & ROTATE(e2) & SPEED(e2,e") & e'*e" This example provides just a first glance into the multitude of semantic parameters6 that are necessary in order to do lexical fine analysis and reveal the concepts we implicitly master in structuring the world around us. We are now at the end of chapter 2.3, which was devoted to
Note that these parameters are in fact the same as the additional arguments of the verb which McConnellGinet proposes in McConnell-Ginet[82], Thus, the only difference between her perspective and ours is that we avoid a multitude of meaning postulates which connect the meaning of a verb in arity η to the meaning of the same verb in arity m. For an extended discussion see Wyner[94],
42 the discussion of semantic participants of events. I argued that there is a distinction between semantically motivated participants, and merely grammatical thematic relations. We found that event individuations should be based on semantic grounds, not on grammatical grounds. In sections 2.3.3. to 2.3.8., I gave a number of examples for semantically motivated parameters and participants and pointed out where these participant relations give rise to event distinctions.
2.4. Summary
Chapter 2 provided an overview over various individuation strategies for events, and the linguistic applications tied to each of these individuation strategies. We started perceiving events as purely temporal objects. Next, we found that events also must have a spatial dimension. However, even sets of times and space were not appropriate to reconstruct events. Our discussion of criteria converged into the set of phenomena and the corresponding criterion in chapter 2.3. - semantic participance. Chapter 2.3. also located the role of manner adverbs in event individuation: Adverbs provide evidence for one prominent semantic parameter which we called VOLI ("volitional participant") and which has much in common with the former notion of AGENT. Chapter 3 lists three challenges to the event ontology which we developed so far. We will discuss various reasons which might lead one to refine the event ontology even more. However, we will be able to maintain the present picture of event individuation.
Chapter 3. Events and Their Names
Chapter 2 set the range of phenomena which naturally fit into event based semantics. Apart from temporal and spatial modifiers, I argued that specifications of various semantic participants and parameters should be treated with reference to events. A notion of events was developed which fits the needs of these applications. Chapter 3 will be devoted to phenomena which go beyond the limits of naive event semantics. We will list three kinds of data which seem to suggest that the events provided in chapter 2 are not yet enough: - Attribution - Evaluative adverbs - Causation In all three cases, we will eventually come to the conclusion that - a theory of the phenomenon which is based on "more" events than those in chapter 2 is not desirable - an alternative account should be given which can be based on the notion of event of chapter 2 - the phenomenon in question therefore does not give reason to postulate more events than those in chapter 2. The title of this chapter quotes a standard reference for one of the three topics to be disucssed, namely the book of Bennett (Bennett[88]) on events in philosophic literature. Bennett's book provides more material with respect to causation theory and can be taken to complement the considerations in section 3.3. The title "events and their names" reflects the observation that all three cases in the present chapter seem to give reason to use language-driven event individuation: Different descriptions always refer to different events.
3.1. Attribution
T. Parsons, in his book "Events in the Semantics of English" (Parsons[90]), discusses the question of event identity and opts for a purely linguistic notion of events. He takes up the Davidsonian idea to represent manner adverbs as predicates over events, and observes the following: "When a verb modifier appears truly in one source sentence and falsely in another, the events cannot be identical" (Parsons[90], p. 157).
44 We are presented a number of uncontroversial examples to illustrate his assumption: "If I both brush my teeth and sing, but if I brush my teeth loudly without singing loudly, then the theory entails that the singing is not the same as the tooth-brushing." (Parsons[90], p. 157) Another case is given in the following argument: Assume that we want to test the hypothesis "Every murder is a killing" It will lead us to the following conclusions: "If χ murders y with a knife, then χ kills y with a knife." "If χ murders y in the hallway, then χ kills y in the hallway." "If χ murders y violently, then χ kills y violently."
Parsons, who adopts the above hypothesis for the sake of illustration, then continues: "I suggest that the above pattern holds for any verb modifier applied to "murder" and "kill". If so, then these data are explained by the additional hypothesis that every murder is a killing. (...) (If I am wrong about the data, then the theory can be used in conjunction with the counterexample to show that some murders are not killings.)" (Parsons[90], p. 156) Formally, the above criterion to identify events can be summed up as: (P)
Two events e¡ and C2 are distinct if a modifier α is true for one but not the other.
Parsons' criterion (P) thus consists in the first conjunct of observation (03) in chapter 2.2. (03) contained the refinement that the observed distinctive behavious must be due to a difference in event parameters. I will now come to cases where (P) as it stands is too strong. Let us look at another example to test (P). Imagine you are told the following story: (1)
Last night Alma sat in the restaurant "Waldhorn" and ate Sauerbraten. She did it greedily, so she was really devouring her meal.
We potentially have two events under debate: (2)
e¡: Alma's eating Sauerbraten e2: Alma's devouring Sauerbraten
Are e j and e 2 identical? In analogy to the above example in Parsons, we might propose that (3)
Every devouring is an eating
45 which sounds plausible enough. But now we have that Alma's eating was done in a greedy manner, which gave reason to call it devouring. However, if you consult a dictionary of English, you will find that "devouring greedily" is still something more than "eating greedily". Synonyms for the former but not the latter are "wolf' or "bolt". Thus, it could be the case that one but not the other of the above events was greedy: (4)
GREEDY(ebut NOT GREED Y(e2)
If we adopt (P) and assume that adverbs are predicates over events, we will be bound to say that the two events can not be identical. However, looking at the present example and comparing it with the phenomena we know from the domain of individuals, we feel that criterion (P) is misleading in this case. The predicate "devour" selects the greedier ones among the events of eating. The adverb "greedily", on the other hand, compares the degree of greediness of an event with that of other events of a contextually given class. And something which is greedy-as-an-eating need not also be greedy-as-a-devouring. Similar cases are wellknown from the domain of individuals, where for example lap-dogs are small dogs, but not every lap-dog is a small lap-dog. The first conclusion to draw is that criterion (P) has to be specified. We can't simply talk about natural language modifiers α without specifying how they are formally represented. Only those modifiers which are unary predicates over events are interesting in the light of (P). However, we, and Parsons, have assumed that manner adverbs like "greedily" in fact are represented as unary predicates over events. Should we thus maintain that the eating and the devouring are distinct? In analogy to our observations about adjectives we will probably refrain from this distinction. In face of example (1) we might simply conclude that manners (tendentially) are attributives for events. In order to implement the respective semantic refinements into the representation of manners, we can rely on much work done in the field of adjectives and comparative constructions. This solution moreover is respectably backed up by the following quotation from Austin: "No modification
without aberration:
when it is stated that X did A, there is a temptation to suppose
that given some, indeed perhaps any, expression modifying the verb, we shall be entitled to insert either it or its opposite or negation in our statement: that is, we shall be entitled to ask, typically, D i d X do A Mly or not Mly ? ' (eg. "Did X murder Y voluntarily or involuntarily ?'), and to answer one or the other. Or, as a minimum, it is supposed that if X did A there must be at least one modifying expression that we could, justifyably and informatively, insert with the verb. In the great majority of cases of the use of the great majority of verbs ('murder' perhaps is not one of the majority) such suppositions are quite unjustified. The natural economy of language dictates that for the standard case covered by any normal verb (...) no modifying expression is required or even permissible." (Austin: "A plea for excuses", in Philosophical Papers [1961])
This observation confirms the comparative character of most adverbial modifiers.
46 On the other hand, we thoroughly discussed sensible individuation criteria for events in the previous chapter, such that it would be somewhat unsatisfying to answer the present puzzle by mere reference to naive intuition. Therefore I want to spend some more time on the question whether attribution should be accounted for by more complex predicates ("greedy-as-a-P") or by more objects ("eating * devouring").
3.1.1. Spreading and Non-spreading Predicates In this section, I want to make an argument by analogy. Let us explore the consequences of Parson's position in a domain where we have comparatively stable intuitions about the nature of things - the domain of individuals. It will turn out that the difference between intersective and attributive predicates still has to be taken care for, just in a different way. Thus, taking care for attributive predicates by increasing the ontology (of individuals in the present section, of events in the last section) doesn't spare us distinguished treatments for the two kinds of modifications. Let us start exploring the domain of individuals, then. What can we say about an object like Fido, a lap-dog? We will find sentences like (5) (6)
There was a dog (under the sofa). It was a lap-dog. There was a brown lap-dog under the sofa.
(7) (8) (9) (10)
—» There was a brown dog under the sofa. Fido is a big lap-dog, but not a big dog. The dog is not a lap-dog. There are two dogs. There are two dogs which are not lap-dogs.
What will happen if we take sentence (5) as a reason to divide the real dog Fido into at least two abstract objects: Fido Fp, the Pinscher and Fido FD, the dog? One result will evidently be, as in the above case for events, that we will be able to predicate BIG over Fρ independently from FD and thus directly account for (3). However, the fact that multiple abstract Fidos will model the one real dog Fido can't be left unaccounted for. Let us adopt a further new equivalence relation Ä which will store the information as to which abstract objects belong to each other. We will see from examples (5) to (10) that we will eventually have to make use of this information, which means that relation Ä is not an ontological luxury without linguistic relevance. Using this idea, how will we represent the sentences (5) to (10)? Sentence (5) will roughly say something like (5,)
3*(DOG(*))
47 Dogs are animals, as we all know. If a dog is present, we want to be able to infer that an animal is present, too. However, we will not be allowed to actually say that all dogs are animals, but have to translate this into the equivalence model as axioms of the form (Axl)
Vx( PINSCHER« -> 3y( Ä(x,y) A D O G « )) Vx( D O G «
3y( Ä(x,y) A ANIMALCy) ))
To ensure that our equivalence classes do not collapse into one element, we will probably want to add that all these abstract Fidos are inequal, and assume (Ax2)
VxVy( PINSCHER« Λ DOGCy) D O G « A ANIMALCy) -» ^
x*y ) ) -
A first introduction of our dog Fido like in sentence (5), translated as (5t), will have the result that we open up an equivalence class (which you might imagine to look like a little box), and fill it with all kinds of abstract objects: an object a, which has the property of being a dog, an object b, which will have the property of being an animal, maybé an object c which is a pet, and so on. If we pick up the same object anaphorically as in the second sentence of (5), this then means the following: Let us assume for the moment that it is unproblematical to find the object a in our Fido-box which corresponds to what we were talking about in the first sentence, that is, the dog-object. Now (5) will continue: (5t-)
a = "the dog" / "it" 3y ( Ä(a,y) Λ PINSCHER« )
In order to verify or say that a dog is a pinscher, we must be allowed to search in the whole box in question, because the dog-thing itself will never be a Pinscher. The same way to proceed will also suggest itself for intersective predicates. Assume we want to say: "The dog was brown." Again if only one element in the equivalence class representing the real Fido is a brown object, this will be enough to recognize Fido as a brown Pinscher, a brown dog, a brown animal, and so on. We will say something like a = "the dog" / "it" 3y ( Ä(a,y) A BROWN« ) With the assumptions made so far, the conclusion in (6) will indeed come out as true. The "brown Pinscher" sentence will open up a box which includes an object a which is a Pinscher, an object b which is a dog, and an object c which is brown. The second sentence will be true if there is a box which is shared by a dog-object and a brown-object: and such a box does exist.
48 Example (7) was the first reason for our multiplication of Fido into a multiplicity of abstract Fi'do-objects. Assume that Fido's box includes an object interpreting the constant F in the representation language. Now (3) will be represented as: (7t)
3y ( Ä(F,y) λ PINSCHER^) a BIG(j) ) a 3y ( Ä(F,y) a DOG(y) a -.BIG(y) )
Thus, whereas "brown" spreads all over the equivalence class, "big" predicates over the individual. Formally, this corresponds to the following two kinds of predicates. "be big" = Xjc.BIG(jt) "be brown" = hc.3y (
a BROWNS) )
If it comes to negate that some individual has an intersective property, as in sentence (8), we therefore will have to negate the existence of an object which has the relevant property anywhere in that equivalence class. Literally speaking, no dog in the model will ever be a Pinscher. So (8) has to be represented as (8t)
a = "thè dog" / "it" ->3y ( Ä(a,y) a PINSCHER^) )
When we want to count dogs, like in sentence (9) and (10), we have to say something like (9t)
3x3y( DOG(jc) a DOG^y) a ->A(;c,;y) )
(10t)
3x3y( [DOG(jc) a ->3y ( Ä(a,y) a PINSCHERCy) ) ] a [DOGOO a ->3y ( Ä(a,y) a PINSCHER^) )] a ~>Ä(x,y) )
Thus, we will count equivalence classes, not abstract objects. Let me summarize: In this dry run, done for objects that we know comparatively well, we have tried to get a feeling for what a modelling of objects will look like in which attributive adjectives can be treated as predicates over objects. We have seen that the idea of "refer to the same real object out in the world" has to be coded somewhere, and that intersectives ("brown") and attributives ("big") still remain different types of predicates. The crucial observation with respect to events is that we get the same range of data. A devouring gives rise to an eating, like lap-dog individuals give rise to dog-individuals. These two events not only coincide in time and space, but they also share all the parameters we have discussed in section 2.3.: Agents, objects eaten, instruments, beneficients, etc. Like in the adjectival case, we get a distinction between "spreading" intersective predicates and "nonspreading" attributive predicates. While features like location, duration, time,
49 participants, speed, instruments, etc. are the same for both the eating and the devouring in our initial example (1), their degree of greediness differs, for example. Thus, we conclude that we will have to adopt something like the theory sketched in this section as soon as we committ ourselves to manner adverbs as unary predicates, with all consequences. We will have to provide the ontological basis the axiomatic background the appropriate semantic representation of each word in agreement with this representation format for manner adverbs. The next sections will show two technical and one conceptual problem which come along with this kind of approach.
3.1.2. Partiality One open issue with respect to an ontology of the kind above is that classical two valued logic is not appropriate in such a framework. We will find formulae where it makes neither sense to claim that they are true nor that they are false. We have seen that Fido-the-dog was small, while Fido-the lap-dog was big. What about Fido-Grandma's-best-friend? What if there is an object Fido-with-both-properties, dog and lap-dog? While the characterization as "Grandma's best friend is open to both, being small as a dog, and being big as a lap-dog, the conjunctive characterization as "being a dog and moreover a lap-dog" does not allow either, "small" or "big". 1 Although it might be possible to implement semantics in partial logic, this step is compulsory in order to turn 3.1.1. into a working semantic theory.
3.1.3.
Representation of Predicates
We saw that intersective and non-intersective predicates have to be treated differently, even in a setup like in 3.1.1. How do we decide whether a predicate in natural language is intersective or not? In principle, it is of course enough for a predicate to behave non-intersectively in at least one case to get a representation like "big", not like "brown" (see above). However, if we refrain from giving an appropriate fine analysis of non-intersective predicates, our semantic analysis will not be as precise as desirable, in many cases. Let me give an example. The predicate "big" clearly was non-intersective. For example, "Fido the dog" and "Fido the lap-dog" can differ with respect to "big". Intuitively, dogs and lap dogs differ in average
I discuss this option because there are two theories which model "event objects" in a naive sense by lattices of "abstract event objects" which are ordered according to the informational increase: Link[87] and Wyner[95]. An analogous ontology for Fido, the dog, will be such that the dog-object and the lap-dogobject give rise to a sum object of these two where both properties hold.
50 size, and "big" refers to size. Therefore we expect differences for these combinations. Another pair of non-intersective predicate are "old" and "young". For example, someone might be an old student but a young president. We will have to give "old" and "young" a semantic representation which refers to individual objects (as for "big" in 3.1.1.), not to equivalence classes (as for "brown" in 3.1.1.). If we do not say any more about "being old", we will therefore predict that Fido-the-dog and Fido-the-lap-dog might differ with respect to the predicate "old". They are different individuals, and "old" refers to individuals. Dogs and lap-dogs, however, do not generally differ in average age (as far as I know). This is reflected by the fact that we will never call an animal an "old dog, but a young lap-dog" or vice versa.2 With the simple-minded treatment of attributivity sketched in section 3.1.1., it comes as an unexpected surprise that "lap-dogs" and their twin "dogs" never differ with respect to the predicates "old" or "young". A more traditional account will represent attributive predicates as comparing a certain value (size, age) of one individual to the values of other, contextually given individuals. In such an account, we are able to predict that lap-dogs are as old as ordinary dogs, although lap-dogs differ in some properties from dogs, and although "old" is attributive for some individuals under a given description. A more traditional account will moreover avoid inconvenient multitudes of individuals and events. This observation provides a further argument against the treatment of attributivity outlined in 3.1.1.
3.1.4. A Greediness Parameter? Let me come back to the "quickly" example in section 2.3.8. We argued that a rotation of a sphere must be distinct from a heating up of a sphere because speed was measured on the basis of different "speed" parameters for either event. We stressed that these different speed parameters have nothing to do with any standard or average speeds for rotation or heating, but only with the question which subevents are relevant in measuring quickness. Nevertheless, we might try to take up this idea with respect to our "eating-devouring" example. What is the value of a hypothetical argument like in (11)? (11)
The event e¡ of eating and the event e2 of devouring differ in the value of their greediness parameter. Therefore, they are distinct.
Assumption (11) in fact takes up the idea of S. McConnell-Ginet who suggests that additional modifiers have access to further parameters of an extendable verbal predicate (McConnellGinet[82]). The modifier GREEDY says something about the greediness parameter; its nonintersectiveness might be accounted for by appropriate values of that parameter. We could assume that an event e is "greedy" iff its greediness parameter is above a certain value. Thus, the greediness parameter of an eating event will tendentially be higher than the one of a "lap-dog" is used as a cover term for "small, handy kind of dog", not as a kind of profession for a dog. In the latter use, a dog might be old-as-a-dog but young-as-a-lap-dog.
51 devouring event, because it takes more for a devouring to be at the greedy end of the scale than for an eating. With respect to example (1) we get (12): (12)
EAT(ey) Λ DEVOURS) GREED-VALUER /) = η > m = GREED-VALUE(e2) GREEDY(ey) λ - ~>0(e). ii. c is a causal factor for e iff there is a series of events c, c/,...,
e (for n>0)
such that each member of the series depends causally on the previous member. iii. c CAUSE e is true iff - c is a causal factor for e and - for all other c ' such that c ' is a causal factor for e: for all worlds w where -0(cO in w, there is some world w ' which is equally or more similar among the ->(0(c) A 0(cO) - worlds than w and -O(c) is true in w'. As a formula: Vw ( w 1= --(0(c) A 0 ( c 0) A w l=(-iO(c 0) 3W'(W'I=-I(O(C)AO(C0)
Λ W' 1= 1 0 ( C ) A d(w0,
W 0 0(β4) A d(wo, w') < d(w0,w) )) On the other hand, the same does not hold true for e2 as a cause for Joe's death. (11) is FALSE.
Apart from adapting Powty's definition for the event case, I have included some refinements with respect to Dowty's definitions while keeping the idea that we compare weightier causes and lighter causes. The differences are of a rather technical nature. The definitions will have to be refined even more, if we want them to work in cases where more than two potential causes for an event are to be compared. They should result in a partial order among a finite set of potential causes. I omitt these details for the moment.
63 (11)
Vw ( w 1= ->(0(e2) Λ 0(e4))
Λ w 1= ->0(e4)
3w' (w' 1= ι ( 0 ( € 2 ) Λ 0{e4)) Λ w' 1= ->0(e 2 ) Λ d(w 0 , w') < d(w0,w) )) (The places where the second formula differs from the first are given in boldface.) The closest worlds are those in which e4 does not occur, but still does. Thus, in comparison to the stronger e4 (Pat's shooting), e 2 (Dr. Spöcks first aid) is ruled out as a possible cause. We can derive this without any unnatural assumptions about the essential properties of the event of Dr. Spöck giving first aid to Joe. Note that the Dowty-style analysis (D.II) does not solve the problem of having too many events. It avoids certain misuses of the lavish richness of the domain of events. However, examples like (3) and (4) above still motivate a rich event ontology. Interestingly, (D.II) in and of itself does not provide an independent indentity criterion, either. The framework is flexible enough to host any kind of events. It is our raw intuition about sentences like (3)a. and (3)b., (4)a. and (4)b. which calls for a fine grained individuation. Let us reconsider (3) in the modified framework. (3) (3a) (3b)
Pat came home late last night, due to a traffic jam. She started cooking spaghetti at 11pm which caused the neighbour to call the police. "The traffic jam caused Pat's cooking spaghetti." (false) "The traffic jam caused Pat's late spaghetti cooking." (true) => The event of Pat cooking spaghetti is not the same event as the event of Pat cooking spaghetti late.
If we ask ourselves which of the two cooking events, the cooking spaghetti (cs) or the cooking spaghetti late (csl), is the one which causes the neighbour to call the police, (D.II) will give a clear answer: Although both events qualify as potential causes, it is only the more fragile (csl) which actually stands in the CAUSE-relation to the neighbour's call (d): (csl) CAUSE (d) A —»(cs) CAUSE (d) This in and of itself does not violate against our intuition. However, it will lead to serious problems in the analysis of sentences of the following kind: (12)
If Pat's cooking had occured earlier, it would not have caused the neighbour to call the police.
The first sentence refers to an event of Pat cooking spaghetti which can occur at earlier times in other possible worlds. It might, for instance, refer to (cs). It may, however, not refer to (csl) because wherever (csl) occurs, it occurs late. The second sentence takes up the event with the anaphoric pronoun "it". The sentence counterfactually states that "it" would not have stood in a certain causal relation. This presupposes that "it" actually does stand in that causal relation in
64 the actual world. (Otherwise we would be allowed to say: "...it would not have caused the neighbour to call the police either", which is certainly odd in the above scenario.) However, the event that causes the call in the actual world is (csl), the essentially late event of Pat cooking spaghetti. But, exactly this event was not a possible denotation for "it". Definitions in the tradition of (D.I) and (D.Ü) have the effect that an event can hardly occur without its causes and effects, in other worlds. However, sentence (12) demonstrates that we quite naturally talk about events that occur without causing what they actually do cause in the real world. This observation can't be accounted for in frameworks which are designed in such a way that an event essentially occurs with all its causes and effects. The conclusion to draw from examples like (8) is one which is not breathtakingly new at all: (13)
There are sentences of the form "the A caused the B", where "the A" and "the B" refer to events a and b, which are NOT equivalent to "a CAUSE b".
Let us call these sentences "pseudo-causal sentences" in contrast to "real causal sentences". The existence of pseudo-causal sentences was already noted by Davidson, in Davidson[69]. A more recent discussion to the same end can be found in Bennett (Bennet[88]).6 Remember that the definition of CAUSE in (D.Ü) was such that it could be used with both, linguistic events or philosophic events. In the rest of this section I will elaborate (13) in such a way that CAUSE is understood as a relation between coarser events like those in the beginning of chapter 2, and all alleged counterexamples are understood as pseudo-causal sentences, i.e. not referring to a causal statement "a CAUSE b". I will interpret pseudo-causal sentences as counterfactual statements involving a focus. This will motivate my distinction between real causal sentences and pseudo-causal sentences on independent grounds, and at the same time give a comparatively precise answer to the question "what do pseudo-causal sentences say?".
3.3.3. Real Causal Statements and Pseudo Causal Statements I claim that there is an essential difference between real causal statements and pseudo-causal statements. Let me repeat two examples: ( 14) (15)
The delayed departure caused Bob's rescue. Dr. Spöck's first aid caused Joe's heart to start beating again.
Sentence (15) will qualify as a real causal statement. We refer to the event ej of Dr. Spöck's first aid and the event e2 of Joe's heart starting to beat, and the world is such that non-
Dowty[79] doesn't have to face these questions because he defines causation as a relation between propositions, not events. He can leave the linguistic notion of "event" untouched, but has to pay the price: Events do not cause any longer.
65 occurrence of the first would have resulted in non-occurrence of the second. In (14), however, matters are different. We refer to the event d of the delayed departure and the event r of Bob's rescue, but, if the departure had not occured the rescue would have taken place all the more. I assume that (14) rather expresses that "if the departure had occured in a different manner (e.g.: earlier, in time), then the rescue would not have occurred". The difference between real causal statements and pseudo-causal statements is this: (16)
Real causal statements talk about the non-occurrence of certain events. Pseudo-causal statements talk about the occurrence of events in a different manner. There is no event so fragile that it could not occur in a different manner.
I assume that pseudo-causal statements are formally distinguished from real causal statements in that the former carry a focus on the crucial property. As in the previous section, I will sketch that recent theories about focus can be used straightforwardly to derive the meaning of pseudocausal statements. I will not go into a formal definition of the "focus semantic value of a sentence". Again, my informal discussion will be based on the interpretation of focus developed by Rooth (Rooth[85], Rooth[92]); and my terminology will be coherent with his theory. If we manage to elaborate (16) in a satisfying manner, then a Wyner-style ontology of events (in order to treat causation) will again be avoidable. We will distinguish real causal statements from pseudo-causal statements in that the former do not contain a focus, and are interpreted along definition (D.II), whereas the latter are distinguished by having a focus feature on some property in the description of the events involved, and consequently are interpreted in a different manner. In the following, we will interpret some sentences as real causal statements and see what the result will be. (17)
Pat's cooking spaghetti (cs) caused the neighbour's calling the police (d).
Sentence (17), if interpreted as a real causal statement, will be true if and only if the relation cs CAUSE d holds true, due to defintion (D.II). We noted above that event (cs), the cooking spaghetti, will in any case be a potential cause for (d), the call. However, we still maintained the picture that there was another potential cause, (csl), the cooking-spaghetti-late event. We have given up this assumption in the meantime. We assume, in agreement with the beginning of chapter 2, that there is only one event of cooking, which is a cooking spaghetti, which occurs late, which is performed by Pat, etc. Event (cs) no longer has to compete with other potential causes, apart from very weak ones like the invention of the telephone, Pat's birth, etc. We will thus find that cs CAUSE d holds true, due to defintion (D.II). Sentence (17) is true in our world. (18)
Pat's late spaghetti cooking caused the neighbour's calling the police.
66 Sentence (18) will be trae for the same reasons as sentence (17). In fact, sentence (18) expresses the same causal statement as (17). The expressions "Pat's cooking spaghetti" and "Pat's cooking spaghetti late" denote the same event. (19)
The (late) departure (e) caused the rescue (r).
We assume that there is only one departure event (e), which is late, which is the departure of the train etc. If we interpret sentence (19) as a causal statement, we predict that it will be true if and only if the relation e CAUSE r holds true, due to definition (D.II). However, we have noted already that the non-occurrence of (e) means that the train does not leave the station at all. In worlds where the train does not leave the station, Bob will be rescued in any case: Pat will come by, will find him lying on the rail, and free him. Therefore, (e) does not even qualify as a potential cause for (r). Sentence (19) will predicted to be false if analyzed as a real causal statement. (20)
The traffic jam (t) caused Pat's late spaghetti cooking (cs).
Sentence (20) will also be false if interpreted as a real causal statement. As in (17)/(18), there is only one event (cs) of Pat cooking spaghetti. If the traffic jam would not have occured, then Pat would still have cooked spaghetti, just somewhat earlier. Therefore, the relation (f) CAUSE (cs) does not hold. Sentence (15) is another example which becomes true, if interpreted as a real causal statement. I will not go through the computations. We have seen some sentences and whether they are true or false if interpreted as a real causal statement. The results were acceptable for sentences (15), (17) and (18). However, for (19) and (20) we are left with a problem. After all we had the feeling that they should be true in some sense, if they are uttered in the scenarios I gave in (3) and (4). It turns out that they are true, if they are interpreted as pseudo-causal statements (of a certain focus structure). Let me start from the by now wellknown sentence (19) in order to give an outline of the interpretation of pseudo-causal statements. I will repeat (19) for convenience. (19)
The delayed departure caused Bob's rescue.
I think that we take (19) to say something like: "If the departure had not occured delayed, then Bob would not have been rescued." In order to evaluate this sentence, we will have to check in "nearby worlds where the departure did not occur delayed" whether the rescue takes place. However, when looking for these nearby worlds we take into account that the departure must
67 have a property contrasting with the delay. The departure should not be omitted altogether, but should occur "early", or "in time". It has been shown that focussing an expression in a sentence always results in evoking a set of logically and contextually appropriate alternatives to the focussed expression (see Rooth[92], Krifka[91], among others). Focus associates with operators which compute the value of the sentence by taking these alternatives into account. For a thorough discussion of these mechanisms, the reader is refered to Rooth[92]. Let us for now denote the set of focus alternatives for an expression α by llallf. Focus alternatives for "delayed" will be, for example, lldelayedllf = {early, in time}. As the alternatives may depend on the context where a sentence is uttered, the set of alternatives is not determined uniquely. However, this notion of "context" is not circular, in contrast to the one used by Lewis (see above). We know this use of "context" from examples which had nothing to do with causation, and we will rely on the same kind of "context" in the present examples. The set of alternatives is furthermore restricted on logical and linguistic grounds. There is no danger of misusing it as a wastebin for unresolved cases of causation. I assume that (19) is uttered with a focus on "delayed". The set of focus alternatives will be used in order to compute the set of possible worlds we quantify over in a counterfactual statement. In short, sentence (21) will be interpreted as (22). I spelled out Lewis' symbol •-> of counterfactual implication as quantification over "nearby" possible worlds. The world parameter consequently has to turn up at all places where a world-dependent property is ascribed. It is given as an index w. (21)
The [delayedjfocus departure caused Bob's rescue.
(22)
Vw( [ ->3e( departure w (e) Λ delayed w (e) ) Λ 3Q3e'( Q e lldelayed departurellf Λ Ôw(0 ) ]
ί 3 / ( rescue w (/, BOB) )
The important point is, that the set of relevant alternative worlds is restricted by the focus alternatives of the property in focus. Let me comment on this point. It was already pointed out by Dretske (Dretske[72]) that focus may restrict the choice of possible alternative worlds. He discussed the following examples giving a scenario where, due to the foci, (23) becomes false, whereas (24) is true. (23) (24)
If Clyde hadn't married BERTHA, he would not have qualified for the inheritance. If Clyde hadn't MARRIED Bertha, he would not have qualified for the inheritance.
The effect which focus has in those cases is similar to the effect outlined in (22). Thus we do not stipulate a completely new mechanism. On the other hand, it is wellknown that focus helps to determine the domain of quantification, both for nominal and adverbial quantifiers (see Rooth[95a], Eckardt[96a]). This is again in line with the use of focus postulated in example (21)/(22).
68 If we represent the pseudo-causal statement (21) as in (22), then sentence (21) will be true in our world. We have to look at counterfactual worlds where the departure occurs with different properties, and where Bob consequently will not be rescued. When (21) was evaluated as a real causal statement, it was false. I think that this analysis reflects the data appropriately. Let us look at some more examples. (25)
The traffic jam caused Pat's cooking [latelfocus.
In sentence (25) we face a focussed property in the description of the second event. Nevertheless, the effect of focus again will be, to restrict the set of possible worlds refered to in the example. The focus alternatives of "late" might be something like {at a normal time, early}. We can use this set of alternatives of a single item to compute the resulting alternatives for the more complex parts of the sentence, as developed in Rooth[85]. This will result in a set of alternative properties of events: HPat's cooking spaghetti latefllf = {Pat's cooking spaghetti early, Pat's cooking spaghetti at a normal time} I assume that sentence (25) states something like this: "In all nearby worlds where no traffic jam occured, there would not have been a cooking by Pat which was late, but instead a cooking by Pat which was at a normal time or early. This is formalized in (26): (26)
Vw( [ ->3e( traffic-jam w (e) -iBfi cook w (PAT, spaghetti,/) Λ late w (/) ) Λ 3Q3e '( Q E HPat's cooking spaghetti latefllf Λ
") )
We have now seen example (21), a pseudo-causal statement with a focus in the description of the subject event, and example (22), a pseudo-causal statement with a focus in the description of the object event. Are there any examples where both, subject and object event description involve a focussed property? Yes, I think we can find such examples. The following sentence is of that kind: (27)
Assume that the coal prices are always low in summer and somewhat higher in late autumn, because everybody has to buy coals then. Assume further that this year, the winter came surprisingly early, and people were more eager than normally to get coals in time. Then we can say: The [earlylfocus beginning of the winter caused the [quick]focus rising of coal prices.
If we interpret (27) as a sentence involving two foci, as indicated in the example, then we will predict the following meaning:
69 Vw>( [ ~3e( begin w (e, winter) Λ early w (e) Λ 3 ß 3 e ' ( Q e llearlyf beginning of
(28)
the winterllf A Qw(e') ) ] -3ft, risew(/, coal price) Λ quickw(/) ) A 3Q3e XQe
llquickf rising of coal pricesllf λ ß w ( 0 ) )
Thus, (28) says something like: "In all worlds where there isn't an early beginning of the winter, but a normal or even a late beginning of the winter, in all these worlds (nearby) there isn't a quick but only a normal or even maybe slow rising of coal prices." This reflects our intuitive understanding of (27) in the given scenario. Sentence (27), if understood as a pseudocausal statement involving only one focus, would not match these intuitions as well as the formula in (28) does. Assume, for example, that there was only a focus on "early". I give the respective structure in (29): (29)
The [early]f beginning of the winter caused the quick rising of coal prices.
Sentence (29) will now be treated like sentence (25) and result in the following formula: V w ( [ ->3e( begin w (e, winter) Λ early w (e) A 3 Q 3 e ' ( Q € llearlyf beginning of
(30)
the winterllf A ß w ( e ' ) ) ] —» -3ft rise w (/, coal price) A quickw(/) ) ) However, (30) only states that without an early beginning of the winter, there would not have been a quick-rising-of-the-coal-prices. (30) leaves the possibility open that the coal prices remain constant in those worlds. That is: Either the winter starts early, and the coal prices raise quickly - or the winter starts normally, and the coal prices remain constant. This is not what (27) was originally meant to express. Can we give a systematic interpretation scheme for pseudo-causal statements? The examples discussed in (21) to (30) have shown that we have a certain pattern to follow. All pseudo-causal statements are of the kind (31)
The A caused the Β
where A and Β are event descriptions, and at least one, A or Β contains a focussed property. These statements are interpreted as universal quantifications over worlds: (32)i.
Vw( i 3 e ( A w ( e ) ) A 3Q3e XQe
IIAIIf A Q w (e 0 ) - > ^3ft B w ( / ) ) )
-.3f( B w ( / ) ) A 3Q3e'( Q e IIBIIf A Q w ( e 0 ) )
ii.
Vw( - d e ( A w ( e ) )
iii.
Vw( ->3e(A w (e)) A 3Q3C '( Q e IIAIIf A Q w (e 0 ) ~3ft B w ( / ) ) A 3Q3E '( Q e IIBIIf A QW(C 0 ) )
70 (32)i. corresponds to the statement with only a focus in the subject, (32)ii. to the statement with only a focus in the object, and (32)iii. to a double-focussed statement. (32) shows that the effect of focus is the same in all three cases, namely, the restriction of the set of worlds in question by reference to the focus alternatives of the original NP. The verb "cause" must be understood as a doubly-focus sensitive operator with the following meaning: (33)
cause(A, B) = Vw( -n3é?(Aw(e)) Λ 3Q3e '( Q e IIAIIf Λ Q w (e 0 ) -a/(B w(/))) if IIAII 0 and IIBII f= 0 = Vw( -3e(A w (e))
-3\f[ B w ( / ) ) A 3Q3e'( Q e IIBIIf A Q w ( 0 ) )
if IIAIIf = 0 and IIBIIf* 0 =Vw( -i3e(A w (e)) A 3Q3e '( Q e IIAIIf A Q w (e 0 ) -> B w ( / ) ) a 3Q3eX Q e IIBIIf a Q w ( 0 ) ) if IIAII pt 0 and IIBIIf I will not go any further into the consequences this definition will have for the syntax-semantics interface. Chapter 3.3. was devoted to a classical topic in event theory, namely the treatment of causation. How do we get evidence about causation? We started using natural language sentences which contain the verb "to cause" in order to sharpen our intuitions about abstract causation. This strategy confronted us with the need to postulate very fine events. However we took the time to show the following: If we assume that causation has something to do with counterfactual reasoning ("what might have happened if other things hadn't"), and if we spell out this assumption along the lines set by Lewis and Dowty, then we end up with a notion of "event" where each event necessarily comes along with all its causes and effects. We can, however, easily imagine an event occuring without certain of its causes and effects. Therefore, I claim, the hope to treat all sentences including "to cause" on the basis of sufficiently fine grained events was in vain. Instead, I suggested to distinguish real causal statements (based on a Dowty-Lewis style analysis and few events) from pseudo causal statements (based on a focus driven interpretation). We took some time to spell out this idea because the focus effects in pseudo causal statements are not as easily accessible as those in section 3.2. The common observation in section 3.2. and the present section is this: Theories where event individuation on the basis of single properties was required can at least in two cases be replaced by (better) theories where variation of properties occurs focus driven, instead of event-based. This observtion motivates the decision to adopt the strategy of event individuation by semantic parameters in the remainder of the book.
Chapter 4: Mereological Structure
4.1. Evidence in Favour of Complex Events All events that we discussed in the previous chapter were comparatively simple. We were concerned, for example, with the selling of a car by Alma to Bertha or a singing of the Bajuwarian anthem by Alma. We did not investigate whether these events might still have smaller parts, or might be used to form complex events. However, chapter 1 has shown that a thorough understanding of the nature of adverbial semantics will involve adverbial predication over non-simple events. Therefore, the present chapter will be devoted to the study of event mereology. As we saw in chapter 2, events can be perceived as a generalization of a spatio-temporal verb argument, telling where and when in time and space some action took place. Sets of points of time, and also sets of points in time and space, are objects which come naturally equipped with the set theoretic operations ç (subset relation) and υ (set union). Moreover, these notions were of relevance in the discussion of certain temporal and aspectual phenomena (like the progressive, telicity, processuality etc.), as long as these were treated on the basis of temporal arguments of the verb. As these applications are naturally transferred to the event argument in an event based verb semantics, we will have the need to find their generalizations in the event ontology. In some sense, the part-whole relation < and the summation operation Θ on events can be viewed as generalized subset relation and union. The set theoretic origin of < and Θ becomes evident when we look at axiomatic characterizations of the structure. Summation Θ usually is assumed to be a total operation which is reflexive, symmetric, associative and idempotent. There certainly are restrictions of θ which are conceptually more sensible. We might restrict summation to temporally or temporally-spatially adjacent events, we might assume that e®e' is only defined if e happens before e' etc. (see for example the summation operations suggested in Asher[93]). However, propagators of total unrestricted summation will point out that all those notions can be defined afterwards on the basis of unrestricted summation, but not vice versa. The following set of axioms was suggested by Krifka (Krifka[89]) to characterize the mereological structure of events on the basis of summation Θ, parts
]
Examples are the following: If Alma smokes in e¡ and Emma smokes in e 2 , then the sum individual "Alma and Emma" smoke in e/®e 2 · If Alma kisses Lukas in e ¡ , and Bertha kisses Lukas in e 2 , then "Alma and Bertha" kiss Lukas in ei®e2. If Alma kisses Lukas in e¡, and Bertha kisses Otto in e 2 , then "Alma and Bertha" kiss "Lukas and Otto" in ei®e2. Evidently, summativity is a property of certain relations R, and not a generally valid prnciple for all properties of events P.This can easily be seen, going back to our first example with Bertha eating apples. For e¡, e2 and we will sensibly assume: ΕΑΤ(Β,α;,^/)
EAT(B,a2,e2)
ΕΑΤ(Β,α^)
-ιΕΑΤ(Β,α /( «2) -iE AT (Β, ay, e^)
-ΈΑΤ(Β ,a2,e¡) -.EAT(B,a 2 ,ej)
-ιΕΑΤ(Β,α 5 ,β;) ->EAT(B,a5,e2)
75
Now for R=EAT and e/.ej, the summativity principle will allow us to say: EAT(B,a / ®a2,e/®e 2 ) If we were allowed to set R=-iEAT and would then look at ej,e 2 again, we could also derive that ->EAT(B,a ¡®a2, e ¡®e2) This will yield a contradiction. It is not Krifka's aim, though, to develop general principles for event ontology. Investigation of aspect and aktionsart classically concentrate on the behaviour of verbs, and thus principle (S) (as well as divisivity (18)) is an empirical observation about the meanings of certain verbs. A full theory of telic and atelic predicates is then developed on the basis of these observations about events, their sums and their parts. Summativity and divisivity are discussed more thoroughly in that full theory of telicity and atelicity in Krifka[89a], Krifka[89b], Krifka[95], as well as Eberle[91].
4.1.2. Temporal Modification of Complex Events Considerations like those in the previous section will naturally only concern sums and parts of events which are rather similar. We summed up eating events where single apples were eaten, to get eating events where sums of apples were eaten. We looked at parts of sleeping-events which turned out to be sleeping events, again. Generally, we were only summing and dividing events which were all described by the same verb. What will happen if we sum arbitrary events? We argued that the summation operation had to be total because set union on sets of times was, or because the resulting structure was more elegant. However, it will be interesting to see that we can actually make use of this concept in natural language representation. The first kind of predication over complex event arises with temporal PPs which measure the precise length of an event. The temporal PP "from χ o'clock to y o'clock" is of that kind, another example are "in"-PPs ("in exactly two hours"). Consider examples (20) and (21). (20)
From 2.00 to 4.00, Lukas cleaned the bathroom and washed the dishes.
(21)
Lukas cleaned the bathroom and washed the dishes in exactly two hours.
These sentences do not imply the ones in (22) to (25). (22) (23)
From 2.00 to 4.00, Lukas cleaned the bathroom. From 2.00 to 4.00, Lukas washed the dishes.
76 (24) (25)
Lukas cleaned the bathroom in exactly two hours. Lukas washed the dishes in exactly two hours.
Therefore we can't assume that the temporal predicates "from 2.00 to 4.00" and "in exactly two hours" apply distributively to both, the event ei of Lukas cleaning the bathroom and the event 62 of Lukas washing the dishes. Both events in isolation are shorter than two hours. Only their sum ei®e2 has a running time of two hours. This fits our above assumptions that event summation is homomorphous to union of running times and insofar brings nothing new. However, (20) and (21) are examples where a sentence of English refers to a complex event which is not formed by summing similar events. We need to know how the expression "Lukas cleaned the bathroom and washed the dishes" refers to the event e/θ«2·
4.1.3. Spatial Modification of Complex Events A second case of reference to arbitrary sums of events arises with respect to local modification. (This, again, is not surprising because we found in chapter 2 that much predication over events can be mimicked by predication over a spatio-temporal verb parameter.) Examples are given in sentences (26) to (28). (26) (27) (28)
A Porsche was driven and a bike was pushed into garages no. 1 and 2. There were priests walking and little children running out of the church and the neighbouring school. In the cages of our zoo, tigers are yawning, guinea pigs are running and monkeys are jumping around.
All three sentences are such that a plurality of objects is related to a plurality of locations by a plurality of events of different kinds. (26) reports about an event of movement, summed from an event of driving a Porsche and an event of pushing a bike, with the collective target location "garages no.l and 2". The sentence as such does not tell which vehicle ends up where. The same holds true for sentence (27), where we might only have expectations about the sources of various people due to our world knowledge. Sentence (28) finally locates the sum of events of tigers yawning, guinea pigs running and monkeys jumping around in the plurality of cages of our zoo. As before, we find that it doesn't make sense to distribute the local modifiers over the various events mentioned. Thus, (26) is not equivalent with (29). (29)
A Porsche was driven into garages no.l and 2. A bike was pushed into garages no.l and 2.
77 (29) will be understood such that the Porsche and the bike both passed the two garages, being brought there successively. Similarly, sentence (28) can not be replaced by (30) to (32), these implying that all cages were inhabited by all three species - which would not exactly be sensible zoo management. (30) (31) (32)
In the cages of the zoo, tigers are yawning. In the cages of the zoo, guinea pigs are running. In the cages of the zoo, monkeys are jumping around.
4.1.4. Causation and Complex Events Can we find examples for non-temporal, non-local collective predication over events, that is, predication which can't be reduced to single events by distribution? One clear case where nondistributive predication is the case arises in causation sentences. We have, of course, to take a stand then as to what kind of objects are causes and effects. It was discussed in chapter 3.3. that classical theories of causation gave rise to another type of event objects than the ones we use. However, it was also argued that this connection of causation and fine-grained objects was not inevitable. I showed that causation can be maintained as a relation between events in the naive sense, giving an alternative analysis to those sentences which refer to isolated properties of (naive) events. Adopting this perspective, real causal sentences are still in the realm of event semantics. Among these, we naturally find many examples where only complex events, but not their parts in isolation, are causes. A case in question might be (33). (33) (34)
Alma turned on the stove and Bertha, at exactly the same time, switched on the dish washer. THIS caused the fuse to blow. Alma's turning on the stove and Bertha's switching on the dishwasher caused the fuse to blow.
First, it is clear that none of the events mentioned caused the fuse to blow in isolation. We assume that they occur cotemporally such that there is not a first and a later event and the later event alone could be blamed. Thus, what we want to say is that there is a causal connection between the complex event and the blow. Next, we do not face pseudo-causal statements. The question is not whether the complex event could have occured in a different manner, without having an effect on the fuse. The complex event would have to be omitted altogether in order for the fuse to remain intact. Thus, both the anaphor in (33) and the nominalization in (34) refer to a complex event summed up from parts of different shape and relates its referent causally to the event of the fuse blowing.
78 4.1.5. Complex Events in Lexical Semantics Let me only mention without further elaborations that a lexical fine analysis of verbs might also require that we perceive the events denoted by simple verbs as sums of smaller parts following each other in the correct order. Imagine that Clara first puts the pan on the stove (e\), then heats some fat in it (¿2). then carefully drops an egg into the fat («3), lets it fry for some minutes (e¿¡) and removes the pan from the Stovetop (es). The complex event glued together from these smaller ones can be described as : Clara prepares a fried egg, sunny side up. 1 This is an accidential property of e 1
, of course. None of the lower events
has this property and passes it up to the higher event. (This observation will become important in chapter 4.3.) It even is not the case that any sum of pan-puttings, fat-heatings, etc... can be joined to a successful preparation of fried eggs. Only the other way round, the stereotype preparation of fried eggs, sunny side up, always involves these steps. Meaning postulates which state such relations will be the event semanticist's answer to prototype, or cognitive, or lexical semantics wherever linguistic phenomena arise which can only be explained on the basis of those prototypes 2 .
4.1.6. Manners òf Complex Events Do manner adverbs also give rise to collective predication over events? A first example was discussed in chapter 1, where it seemed to make a difference whether a manner adverb had scope above or below a quantificational NP (see chapter 1.3., examples (1) to (3)). Another, more complex example was already discussed by Thomason and Stalnaker (Thomason/Stalnaker [1973]). They note that we can say something like (35). (35)
Reluctantly, Jones bought gas and had the oil changed.
Thomason/Stalnaker observe (35) can be used in a situation where Jones is not reluctant about the buying of the gas alone nor about the changing of the oil alone. It is the complex event of both being done together (and forcing him to spend a lot of money) which was the target of his reluctance. The adverb "reluctantly" relates Jones to the complex event of "Jones buying gas" θ "Jones having the oil changed". This is not the same as distributing the predicate over the two subevents, which would yield the prediction that Jones reluctantly bought gas, and Jones reluctantly had the oil changed.
= "Spiegelei" The following might provide an example. Imagine that you want to explain the definite use of "the fat" in the sentence: "Bertha prepared fried eggs. The fat was rancid". Using the standard shape of events of frying eggs, you will be able to derive that fat is involved in subevent no.2 and that it was this fat which was rancid. It is a matter of debate whether such phenomena are still within the range of linguistic investigations or whether they have to do with knowledge about cooking, though.
79 4.1.7. Anaphoric Reference to Complex Events Whenever we know that a certain construction has to be treated on the basis of events, we will find corresponding anaphoric constructions on the basis of events. On one hand, all phenomena above give rise to examples of anaphoric reference to complex events. An explicit example was already given in (33). More examples are listed in (36) and (37). (36)
(37)
Alma shoveled the walk and built a big snow woman. It took her two hours. She did it from 3.00 to 5.00. This happened from 3.00 to 5.00. Bertha was baking a surprise cake for Clara's birthday, and Alma was occupied finishing a present. As this happened in the kitchen and the living room, Clara herself had to stay in her bedroom.
As the original sentences were shown to involve collective predication over complex events, this is also the case for the anaphoric examples (33), (36) and (37). However, event anaphors provide further evidence that predication over complex events takes place. The argument is much similar to the argument with respect to the anaphor test in chapter 2.3. Consider the sentences in (38) and (39). (38) (39)
Bertha repaired and cleaned the Porsche in the garage. Bertha repaired and cleaned the Porsche. She did it in the garage.
Faced with sentence (38), we might still maintain the claim that it isn't necessary to allude to a complex event of Bertha repairing the Porsche and cleaning the Porsche. After all, (38) is in fact equivalent to stating that Bertha repaired the Porsche in the garage and Bertha cleaned the Porsche in the garage. Simple distribution of the local PP over both simple events would yield the correct semantic representation. If this distribution, however, is to take place without making reference to any complex events then it must be some kind of syntactic (or "LF") process copying the PP and attaching it to the appropriate places close to the verbs "repair" and "clean". These two structural places, however, are no longer available in the anaphoric example (39). All we are given in the second sentence in (39) is anaphoric reference to an event. The anaphor "it" can't be used up twice, first referring to the repair and then referring to the cleaning, and passing both to the PP "in the garage" in succession. Therefore, the second sentence can only talk about the complex event, even if the formulae in (40a) and (40b) are equivalent. (40a) (40b)
IN( e i ® e 2 , GARAGE) IN( e ι , GARAGE) & IN( e 2 , GARAGE)
80 Thus, anaphors to events show that we can refer to complex events and make statements about their properties. In examples like (39), the complex event will inherit the relevant property from its simpler parts. In examples like (36) and (37), like in all previous cases in this section, we find that complex events can have more properties than their simple parts. We will make use of this observation in chapter 4.3. The aim of chapter 4.1. was to introduce a basic mereological structure on events, and to list a number of examples where sentences refer to complex and simple events. Up to now, we have said nothing with respect to the question how exactly a sentence of the form "A and B" has to be represented semantically in order to denote a property of events which will be true of the sum of the Α-event and the B-event which we want to refer to. This question will be investigated in chapters 4.2. and 4.3.
4.2. Persistent Event Structures
Chapter 4.1. did not lead us to any general predictions about the properties of summed events, given that we know the properties of the parts. Essentially, question (#) remains unanswered: (#)
Given that P(e) and Q(e 0. what do we know about e®e ' ?
Let me illustrate this question with an example. Assume that we have the events in (1). Do we know anything about the event e®e'? (1)
e\ Clara has a nap e ': Dorothea repairs the Porsche
It seems reasonable to claim that e@e ' is an event where both happens: Clara has a nap and Dorothea repairs the Porsche. This observation motivates the following answer to question (#). (PP)
If P(e) and Q(e 0 then ?{e@e 0 and Q(e®e ")·
(PP) states that information persists from lower to higher events, and will therefore be called "persistency principle". The idea of information persisting from smaller units to larger ones has been discussed in situation theoretic approaches (see Muskens[92], Kratzer[89], also Zucchi[95]). The conceptual borderline between situations and events is not always very clear (although I will refute the idea that events and situations are the same objects, as one result of the overall book; see Appendix I). Therefore, it looks like a natural idea to adopt (PP) for event mereology.
81 Clearly, adopting (PP) as a general principle of information flow in event structure will have the side effect that event semantics takes place beyond the borders of classical two-valued logic. Let me repeat (PP) in a more general form: (PP)
If P(e) and e and the corresponding infimum operation *. Events and propositions are linked by the "holds-in" relation l=. This relation is used, on one hand, like the "models" relation in
82 classical model theory. On the other hand, it is implicitely understood to replace our previous integration of events as a parameter of the verb: Instead of "P(xi,...,xn,e)" we are invited to write "e 1= P(xi,...,x n )"· Lasersohn, not starting from any well developed account of traditional event semantics, leaves it open inhowfar other insights about events will translate from the event-as-argument notation to the event-as-true-maker notation. The main question to be studied at present is this: When does a proposition hold in an event? e\=p I will now give the definition of a Lasersohn structure Ξ from Lasersohn[92]. Definition 0.: Let Ξ := n . Each Pn is assumed to have a least element 0n with respect to =>". For each subset X of Pn, let *X be the greatest lower bound of X with respect to =>". By applying them to an individual argument, n+l-ary relations turn into n-ary relations in the usual manner. • 1= is the support relation between events and propositions. • Τ is the set of sets of times with the subset relation < j and the "temporarily precedes"-relation «γ. • τ is the function mapping events onto their running times, and is a homomorphism with respect to < E and < T • F is the interpretation function that relates some natural language to the model. In addition, the following axioms are required to hold true (I adopt the enumeration given in Lasersohn[92] and will give a short motivating comment for each axiom, in the spirit of Lasersohn[92]): (13^)
Forall e,e'e E;pe P^: \ie\=p and e< E e then e'\=p. This is the aforementioned persisteny principle.
83 (14^,)
For all X^E; eh...,ene
ρ e P°:
if
Θ Χ 1= p,
then for some q¡,..„
X: e¡ 1= qj,...,e„ 1= qn, and q¡ * ...
qn e P°,
and
*qn=p.
This principle is introduced by Lasersohn to ensure that "nothing more happens" in complex events than what they could inherit from their simpler parts. ( 1 5 ] ^ . For all ee E; p,q e PO·. if e 1= ρ and e 1= ρ then e 1= p*q. (15L a )b. For all ee E; p,q e pO; if e 1= ρ and p=>q,
then e 1= q.
These two principles guarantee a sensible connection of the event lattice and the propositional algebra. ( lÓLa)
For all ee E; there exists some ρ e i^such that e 1= p. This requirement ensures that no "empty" events exist. Something should happen in each event.
(17^)
For no ee E; is it the case that e 1= ( f i . N o event should support the contradiction.
(18^)
Definition: For all ee E; ρ e PO·. e is a simple p-event iff e 1= ρ and for all q e pO, if e I - q then q = p. e is a simple event iff e is a simple />-event for some ρ e pO. Simple events will be of great importance in what follows. Lasersohn has to use them in order to develop a semantic representation for "alternately". However it seems plausible in any case to assume that there are events in which not much happens.
(19La)a. For all ee E; there exists some set X of simple events, such that e = ΘΧ. ( l Ç ^ b - I f X and Y are distinct subsets of E whose elements are all simple events, then (19) ensures that the event lattice is freely generated over the set of simple events. Any structure Ξ which fulfills requirements (13La) to (19La) will be called a LasersohnStructure. (The definitions and theorems in the rest of Lasersohn's paper are concerned with the proper analysis of the adverb "alternatingly"; we need not consider them here.) I will sometimes write "e is simple for p" instead of "e is a simple p-event". M y aim is to develop a characterization theorem for an important subclass of LasersohnStructures. Let me start by giving some lemmas about Lasersohn-Structures Ξ. Lemma 1.: Let Ξ be a Lasersohn structure and e a simple /»-event. Then the proposition ρ cannot be the infimum of further propositions.
84 Proof: Assume that e\=p and that e is a simple /»-event. Thus for all propositions q: if e 1= q then q=p. Assume that p=p\*p2- Thus ρ => p\ and ρ => ρ2· So with (15La)b.: « '= ρ ι and e 1= ρ2- But then p\=p
and P2=P- Thus {p\,p2)
= [ρ}» and ρ is not a
conjunction of further propositions. Lemma 2.: Let Ξ be a Lasersohn structure and let e\ and e2 be simple for p\ and p2, respectively. Then the complex event e\@e2 supports only the following three propositions: e\®e2 1= p\*pi, e\®e2 1= pi, e\®e2 1= P2, and nothing else. Proof: Assume that e\@e2 1= q for some proposition q. But e\@e2 = ©(«1,^2} and thus (14La) applies. There are q\,...,qn e P° and e'\,...,e'n e {e ι ,«2} such that e'\ 1= q\ «Ì
«V (i) e'i=ei:
e'\ I = qn and q\*...*qn
=q. However we don't have much choice for
Then q\ = p\ (due to Lemma 1) and thus q = p\-
(ii) e 'i= e2 : Then q\ = p2 (due to Lemma 1) and thus q = p2(iii) e'i=ei and e'\= e?. Then q\ = p\ and q2 = P2 and thus q = p\*p2Lemma 2. easily generalizes to sums of η simple events e = e\® ... ®en. It shows that we have comparatively good control over the properties of simple events and their finite joins. Moreover lemma 1. indicates that not only simple events, but also the propositions which have a simple event, have a distinguished status. Let us make the following definition: Definition 3.: Let Ξ be a Lasersohn structure and ps
P° a proposition such that there is an
event e which is a simple p-event. Then call ρ a simple proposition in Ξ. Let SP° be the set of simple propositions in Ξ. The following lemma tells us something about simple propositions. We will moreover make use of it later. Lemma 4.: Let Ξ be a Lasersohn structure and {qi,...,qn}*
{Pl.—.Pm} finite sets of
simple propositions. Then the respective suprema are inequal, too: *{qi,...,qn}
*
*{pu.,.lPm}
Proof: As the q\, p\ are simple, they must be simple somewhere. Let e\ be simple for q\, and fi simple for p-y Let us assume, without loss of generality, p\ i { ι ,...,