Ewald Lang, Claudia Maienborn & Cathrine Fabricius-Hansen (eds.) Modifying Adjuncts. Berlin-New York: Mouton - de Gruyter (2003), 511-552 Flexibility in adverbal modification: Reinterpretation as contextual enrichment* Johannes Dölling Abstract This paper is concerned with the fact that a number of adverbal modifications in- volve a reinterpretation of at least one of the expressions connected by the operation in question. It investigates the question of to what extent such meaning transfers can be viewed as emerging from operations of semantic coercion. Unlike other proposals, the paper offers an approach in which systematic reinterpretations turn out to be a result of contextual enrichments of an underspecified, yet strictly com- positional, semantic representation of the given utterance. The approach proposed takes into account reinterpretations in both temporal and non-temporal modificati- on. Moreover, it allows so-called secondary predications to be handled as a particu- lar kind of adverbal modification. In addition, it opens up the possibility of also treating reinterpretations beyond those triggered by direct semantic conflict. The paper outlines a general formal framework of meaning variation that inter alia explains the semantic flexibility we observe in adverbal modification. 1. Introduction It has been observed that modification by adverbials sometimes seems to involve more than a simple composition of the meanings of the expressions combined. There are cases which suggest that, in addition, they require a reinterpretation of at least one of the syntactic constituents connected by adjunction. This is illustrated by sentences like (1) and (2), which contain durative adverbials as modifiers of verbal expressions with which, in a strict sense, they should not be compatible.1 Nevertheless, the sentences can be interpreted without much difficulty. (1) Eva nieste zehn Minuten lang. ‘Eva sneezed for ten minutes.’ (2) Udo las den Roman zwei Stunden lang. ‘Udo read the novel for two hours.’ 512 Johannes Dölling Sentence (1) is not interpreted as referring to a single sneezing but to re- peated sneezing events which together last ten minutes. Sentence (2) is not construed as describing that it took Udo two hours to read a certain novel. Rather, it conveys how long Udo was reading the novel without finishing it. Thus, in both cases, the adverbial is not taken to specify the situation de- noted originally by the expression it modifies but, rather, the duration of a sortally different situation that is connected with the situation at issue in a specific way.2 Evidently, such modifications can be performed only if the given verbal expression undergoes a suitable meaning adaptation.3 Firstly, I will argue that the observation described above is not confined to occasional occurrences of modifying adjunction but rather that reinterpre- tation in adverbal modification is a ubiquitous phenomenon in utterance understanding. In particular, I will show that the proposal developed in Moens and Steedman (1988), according to which the situation type of ver- bal expressions can be coerced by temporal adverbials, does not cover all of the cases where such operations influence the interpretation of the given utterance. The flaw we are facing is twofold. On the one hand, when we take a closer look at such meaning transfers, we encounter numerous occur- rences of non-temporal adverbials that have to be taken into account as well. Moreover, from the point of view I am adopting here, so-called secondary predications can also be understood as a special kind of adverbal modifica- tion. On the other hand, meaning adaptations can be observed not only in the modified constituents but also in the modifiers. Thus, many modifying adverbials and all secondary predicates turn out to be subject to operations of meaning transfer. Secondly, I will demonstrate how systematic reinterpretations of the type considered here can be analyzed within a multi-level model of meaning rep- resentation. Starting from the idea that in understanding an utterance the conceptual information it conveys has to be elaborated step by step, the model rests on a distinction between two types of operations: (i) operations of computing a strictly grammatically determined, context-independent and, hence, underspecified meaning; (ii) operations of subsequently specifying by contextual enrichment the meaning that results from (i). As a conse- quence, one and the same expression can receive several readings depend- ing on the context of use. Unlike other partly similar approaches, my pro- posal builds on the claim that in a compositional derivation, the variation potential of meaning can be systematically extended by obligatory semantic operations of structural enrichment. The strategy followed by this model has several advantages. First of all, in contrast to the proposals of Jackendoff (1991, 1997, 2002) and Pustejovsky (1991, 1995, 1998), the principle of semantic compositionality is entirely maintained in its validity. Second, the Flexibility in adverbal modification 513 approach opens up the possibility of explaining reinterpretations in adverbal modification as triggered not exclusively by an immediate semantic conflict between the verbal head and its adjunct but also by other factors. And fi- nally, within the model chosen here, the phenomena under discussion turn out to be instances of a more general kind of meaning transfer insofar as the operations underlying them account for the semantic flexibility we observe in utterance interpretation beyond adverbal modification. The paper is organized as follows. Section 2 gives a survey of relevant reinterpretation data with temporal adverbials as modifiers. Section 3 exam- ines the question of to what extent such meaning transfers can be considered a result of more or less concrete operations of type coercion. Section 4 of- fers, as an alternative, a general outline of the multi-level model of meaning representation. Section 5 applies it to the analysis of the problem area by way of example. In Sections 6 and 7, the approach proposed is extended to further phenomena of meaning transfer. Section 6 delivers an explanation for reinterpretations that come with adverbials of manner, location and di- rection and Section 7 does so for constructions with depictive and resulta- tive predications. 2. Temporal modifiers that trigger reinterpretation Let me begin with a closer look at sentences (1) and (2) in which an achievement, or more specifically, a semelfactive4 and an accomplishment, respectively, are modified by a durative adverbial. The deviation from pri- mary meaning we observe in sentence (1) is based on an iterative reading of the verb niesen (‘sneeze’). Originally, this verb denotes only a property of momentaneous eventualities, or simply, of moments. By way of reinterpre- tation, however, it can denote a property of a process that consists of im- mediately successive acts of sneezing.5 Suppose that p and m are variables for processes and moments, respectively, TH and CONST are predicates for the relations ‘the THeme of’ and ‘CONSTituent of’, respectively, and that τ denotes the temporal trace function that maps eventualities to the time points with which they coincide. Then, ignoring the semantic contribution of tense, mood, etc., the content of (1) can be identified by the structure given in (1a).6 (1) a. ∃p [TH(eva, p) & ∀m [CONST(m, p) → SNEEZE(m) & TH(eva, m)] & τ(p) ≥ 10min] 514 Johannes Dölling Thus, sentence (1) indicates the duration of a sneezing process the Theme of which is Eva. Conceptually, if the case of (2) is seen in analogy to (1), an iterative interpretation of den Roman lesen (‘read the novel’) referring to a chain of immediately repeated events, during which one and the same novel is read, would of course be possible. But in view of the time that is usually needed and, according to (2), available for reading through a novel, this interpretation is very unlikely. Instead, in order to meet the condition of the adverbial, we will focus on the internal structure of events and thereby limit ourselves to their so-called developmental phase. Given this, in the contin- uous interpretation of (2), the VP denotes the set of processes which make up events of reading a novel irrespective of their possible completion by culmination.7 Using e as a variable for events, AG as a predicate for the relation ‘the AGent of’, and COMPL as a predicate denoting the COMPletion relation between events and processes, respectively, the information con- veyed by (2) can be represented as follows: (2) a. ∃p [AG(udo, p) & TH(novel, p) & ∃e [COMPL(e, p) & READ(e) & AG(udo, e) & TH(novel, e)] & τ(p) ≥ 2hours] Here, Udo is the Agent and a certain novel is the Theme of a process which lasts at least two hours and which forms a constituent part of a potential reading event, the Agent and the Theme of which are likewise Udo and the novel, respectively.8 For a sentence like (3) in which, again, an accomplishment comes in combination with a durative adverbial, a process-related interpretation is also possible. (3) Anna öffnete das Fenster fünf Minuten lang. ‘Anna opened the window for five minutes.’ While a continuous reading of the VP das Fenster öffnen (‘open the win- dow’) seems to be adequate only under particular contextual conditions, an iterative interpretation is quite feasible. If, however, such an understanding is not explicitly suggested by the context, sentence (3) will have a clear preference for a third kind of interpretation on which the adverbial specifies the duration of the state brought about by the event of opening the window. In this case, (3) conveys that Anna opened the window and the resulting state of its being open lasted at least five minutes. This reading is repre- sented in (3a), where s is used as a variable for states, RES and HD as predicates for the relations ‘RESulting state of’ and ‘the HolDer of’, respec- tively.9 Flexibility in adverbal modification 515 (3) a. ∃e [AG(anna, e) & OPEN(e) & TH(window, e) & ∃s [RES(s, e) + & OPEN(s) & HD(window, s) & τ(s) ≥ 5min]] Unlike the cases considered so far, (3) in this understanding requires the adverbal modifier to be shifted in its meaning in order to meet the conditi- ons of the verbal expression. The use of durative adverbials such as fünf Minuten lang (‘for five min- utes’– a literal equivalent is missing in English, see below) has hardly been mentioned in the literature, which mostly deals with English.10 A first pro- posal for explaining such resulting state-related reinterpretations was made in Dölling (1998), which served as a starting point for Piñón (1999) who, however, argues against the necessity of a meaning transfer. Instead, Piñón assumes that the argument structure of a resultative verb like öffnen (‘open’) contains an additional state variable with which the durative adverbial can immediately link up in modification.11 As a consequence, it seems that the adverbials under review here might be treated analogously to adverbials like für fünf Minuten or their English equivalents for five minutes but have to be restricted to specifying the duration of resulting states. For several reasons, I consider such an approach unacceptable. First, this approach does not take into consideration the fact that the be- haviour of a resultative verb may vary when modified by different result- oriented modifiers. For example, the verbs zerbrechen (‘break’), essen (‘eat’) und zerstören (‘destroy’) are compatible with wieder (‘again’) in its restitutive reading, although the resulting state induced by these verbs can- not be temporally restricted by other durative modifiers. Since, according to Piñón's proposal, such verbs thus cannot have an additional state variable, it remains unclear how the modifier wieder could be linked up with such a variable.12 Second, Piñón’s proposal should also account for cases in which the putative state variable in the argument structure is not required for adverbal modification. In fact, it includes the assumption that whenever the variable remains unused in this sense, then a special semantic operator takes over the linking. However, since this is evidently the standard case, the proposal requires an additional step in the compositional derivation of these verbs that can scarcely be motivated. Third, Piñón’s approach reveals the need for a more general procedure that enables us to capture another type of occurrence of modifiers for which no verbal linking site exists. As we shall see, modifying expressions such as unauffällig (‘unobtrusively’), elegant (‘elegantly’) or korrekt (‘correctly’) can be used in such a way as not to characterize the resulting state associ- ated with the verb. Rather, these adjuncts relate to objects which emerge as 516 Johannes Dölling a result of the respective event. Thus, it appears to be mistaken to assume that such resultative verbs have a further argument position for resulting objects. It will be shown later that the solution based on Dölling (1998) not only avoids the problems mentioned above but that it also has the advantage of being an instantiation of a more general approach. In sentence (4), the adverbial drei Wochen lang (‘for three weeks’) cer- tainly neither implies that Jutta arrived three weeks late nor does it specify the duration of a single arrival of Jutta. (4) Jutta kam drei Wochen lang zu spät an. ‘Jutta arrived (too) late for three weeks.’ However, contrary to the cases of reinterpretation adduced above, in (4) it is highly improbable that the modifier is used to characterize a process of ar- riving late on any occasion within three weeks. Rather, in view of our stan- dard experience, in the given use the achievement zu spät ankommen (‘ar- rive late’) should be understood in the habitual reading.13 Thus, (4) refers to a habitual state of Jutta which lasted at least three weeks and which was realized by repeated, but not immediately successive, situations of arriving late. (4) a. ∃s [HD(jutta, s) & ∀b [REAL(b, s) → ARRIVE_TOO_LATE(b)] & τ(s) ≥ 3weeks] Here, b is a variable for borderline situations, or more simply, borders14, as characterized, for example, by the verb ankommen ‘arrive’ while REAL stands for the relation ‘REALization of’. Let me now turn to the analysis of cases where time-span adverbials occur as modifiers of achievements, states or activities. Since, for example, den Gipfel erreichen (‘reach the summit’), as well as ankommen, denotes a property of borders, in a sentence like (5), the adverbial in zwei Tagen (‘in two days’) cannot serve to modify the VP in its original meaning. (5) Ede erreichte den Gipfel in zwei Tagen. ‘Ede reached the summit in two days.’ However, sentence (5) can be understood such that Ede was the Agent of an event which ended within two days by Ede’s reaching the summit and thus culminating in it. Using CULM as a predicate for the relation ‘the CULMina- tion of’, the content of (5) can be represented as in (5a). Flexibility in adverbal modification 517 (5) a. ∃e [AG(ede, e) & ∃b [CULM(b, e) & REACH(b) & TH(summit, b)] & τ(e) ≤ 2days] The core of this event-related reading is that the VP den Gipfel erreichen changes from a predicate of borders into a predicate of events that terminate in such borderline situations.15 A sentence like (6) can be treated in a similar way. (6) Sarah war in fünf Minuten wach. ‘Sarah was awake in five minutes.’ + (6) a. ∃e [TH(sarah, e) & ∃s [RES(s, e) & AWAKE(s) & HD(sarah, s)] & τ(e) ≤ 5min] As follows from (6a), Sarah is characterized as the Theme of an event which results in her being awake within five minutes. This interpretation of (6) requires the expression wach sein (‘be awake’), which originally denotes a property of states, to be changed into a predicate of events that may have the pertinent resulting state.16 It is somewhat more complicated to assign an event-related interpre- tation to a sentence like (7). (7) Peter rannte in fünfundvierzig Sekunden. ‘Peter ran in forty-five seconds.’ Here, it would be necessary to construe the process predicate rennen (‘run’) as a predicate that can describe an event, the developmental phase of which is formed by a quantum of the running process. Then, the content of (7) can be identified with (7a) where the predicate SUBST denotes the relation ‘SUBSTratum of’ between processes and events. (7) a. ∃e [AG(peter, e) & ∃p [SUBST(p, e) & RUN(p) & AG(peter, p)] & τ(e) ≤ 45sec] Obviously, such an understanding is justified only in contexts from which a suitable culmination can be drawn − in (7) by way of identifying a certain running distance. 518 Johannes Dölling 3. Reinterpretation by sort coercion? Meaning transfers that occur in connection with modification by durative or time-span adverbials have already been documented more or less exten- sively in the literature, and various proposals for their explanation have been advanced. Basic deliberations can be found in Moens and Steedman (1988), where a first systematic, albeit informal, analysis of reinterpreta- tions in temporal modification is presented. Moens and Steedman analyze temporal adverbials (as well as aspectual auxiliaries) as functions which, under particular conditions, induce changes in the meaning of the verbal expressions they modify. The change involves that the verb’s reference to situations of one sort gets transformed into a reference to situations of an- other sort. Such meaning adaptations based on a correspondingly differenti- ated network of ontological relationships is referred to as type coercion.17 How the relevant shifts are to be accomplished in detail, however, still re- quires explication. It can be assumed that such adverbials trigger semantic operations by means of which the verbal expressions are directly reinterpreted in a suit- able way thereby creating the prerequisites for suitable modifications. So, if a conflict arises between the sortal selection restrictions of an adverbal modifier and the semantic sort of its argument, a specific operator applies to the verbal predicate to achieve sort matching. For example, the reinterpreta- tion that takes place in (2) can be explained simply thus: utilizing a special coercion operator which meets the requirements of the adverbial, the mean- ing of the VP den Roman lesen gets transferred from a predicate of events to a predicate of processes. However, this mechanism of direct semantic adaptation leaves a number of questions unsettled. As discussed with respect to (2) above, the occur- rence of a sortal conflict between temporal adverbial and verbal expression does not at all predetermine the form of its solution by the underlying con- ceptual ontology. A first problem is how, out of the set of conceptually pos- sible operators, and in both systematic and economical a way, we can pick out exactly those operators that will each time provide the adequate reinter- pretations. It is obvious that this choice cannot be made without resorting to world and discourse knowledge and without allowing for specific pragmatic restrictions. A second, and more serious, problem follows from the fact that by inserting such adaptation operators, contextually determined parts of meaning are introduced into an otherwise compositional derivation. Obvi- ously, under this condition, the general validity of the principle of semantic compositionality cannot be upheld any longer.18 In view of the fact that we so far lack any convincing alternative to this principle, renouncing a strictly Flexibility in adverbal modification 519 regulated calculation method of context-independent meaning of expres- sions is not acceptable. As a possible way out, some authors have offered a two-step approach according to which necessary reinterpretations are realized as follows: In a first step, a semantic representation of a given utterance is constructed in terms of compositionality. If, in the derivation of the context-independent meaning, a sortal conflict arises, it is resolved by inserting an underspecified coercion operator. In a second step, attempts are made to justify this hypo- thetic sortal adaptation by exploiting world and discourse knowledge in order to contextually specify the semantic representation. Thus, it is only in this step that a proper meaning transfer, if it is possible, gets realized.19 Taking up this idea for the reinterpretations discussed above, it seems that only two underspecified adaptation operators are required: one for con- structions involving durative adverbials and one for those involving time- span adverbials. Given the sortal selection requirements of adverbials, the first operator should allow transferral of predicates of events, borders or moments to predicates of processes or states; the second operator transfers predicates of borders, processes or states to predicates of events. These con- ditions are largely met by the operators proposed in (8) and (9) where e/b/m, p/s and b/p/s are provisional variables for situations of the respective supersorts, Q varies over the quantifiers ∀ and ∃, C varies over the connec- tors & and → and R is a parameter for relations between situations of vari- ous sorts. (8) λPλp/s. Qe/b/m [R(e/b/m, p/s) C P(e/b/m)] (9) λPλe. Qb/p/s [R(b/p/s, e) C P(b/p/s)] Now, if, for example, (8) is used in the compositional construction of the semantic representation of (10), the structure given in (10’) can be assumed to be the result of this derivation (further provisionals included). (10) Ilse spielte die Sonate einen Tag lang. ‘Ilse played the sonata for one day.’ (10’) ∃p/s [AG/HD(ilse, p/s) & Qe [R(e, p/s) C PLAY(e) & TH(sonata, e)] & τ(p/s) ≥ 1day] Then, conceptually possible specifications of (10’) will result in (10a) to (10c). 520 Johannes Dölling (10) a. ∃p [AG(ilse, p) & ∀e [CONST(e, p) → PLAY(e) & TH(sonata, e)] & τ(p) ≥ 1day] b. ∃p [AG(ilse, p) & ∃e [COMPL(e, p) & PLAY(e) & TH(sonata, e)] & τ(p) ≥ 1day] c. ∃s [HD(ilse, p) & ∀e [REAL(e, s) → PLAY(e) & TH(sonata, e)] & τ(s) ≥ 1day] Which of the alternatives eventually provides the specified content of an utterance of (10), i.e. whether it refers to a process of successively repeated playing the sonata concerned (iterative reading), to a process being only part of an individual playing event (continuous reading) or to a state realized by repeated but not uninterrupted playing the sonata (habitual reading) has to be decided depending on world knowledge and other contextual informa- tion.20 However, an approach like this, in which semantic sort adaptation and context-related reinterpretation are separated, will also lead to difficulties. First, it has to meet the condition that meaning transfers can proceed only in one direction.21 The inadequacy of this condition emerges from sentences like (3) where in addition to the reinterpretation of the verbal expression, reinterpretation of the modifying expression is also possible. Therefore, the starting point of a required meaning transfer is not at all clearly determined a priori. Thus, it has to be decided to which of the involved expressions an adaptation operator is to be applied. However, decisions of this kind are not compatible with a strictly compositional semantic derivation because they require extra-linguistic knowledge to be taken into account. Second, under this approach, the fact that not every meaning transfer in adverbal modification has to result from a direct conflict of sorts is left out of consideration.22 For example, in (11), joggen (‘jog’) fulfills the sortal selection restriction of durative adverbials insofar as this verb represents a predicate of a process. (11) Renate joggte zehn Jahre lang. ‘Renate jogged for ten years.’ Accordingly, (11) can mean that Renate’s activity of incessant jogging lasted at least ten years. (11) a. ∃p [AG(renate, p) & JOG(p) & τ(p) ≥ 10years]