| (50) | *Sie hofft die Ankunft von Herrn Meier | (ungrammatical) |
| (51) | *Sie verdächtigt ihn des Herrn Meier. | (semantically incompatible) |
| (52) | ? Sie hofft auf Herrn Meier. | (semantically incompatible, unless Herrn Meier is understood to refer to an event, such as his arrival or help, etc.) |
| (53) | *Es dauert um 7 Uhr. | (ungrammatical and semantically incompatible) |
| (54) | *Sie sitzt auf das Dach. | (ungrammatical and semantically incompatible) |
As was shown above, the assignment of Spez-Sel values turns on this differentiatiation, something that is not always possible (cf. Eisenberg 1994:297). For verbs requiring autosemantic complements such as expansive (53) or situational complements (54), grammatical and semantic incompatibility can coincide, which is arguably an additional difficulty for Spez-Sel.
(55) Sie aß einen Apfel.
aus Hunger is excluded from the commutation class based on einen Apfel because it represents a different construction. (One cannot avoid using judgements on the identity of constructions by pointing out that aus Hunger is anaphorised differently from einen Apfel because establishing the commutation class has to precede identifying its anaphora — otherwise one would beg the question.)
In defence of Spez-Sel it has to be said that one need not overlook the whole commutation class in order to qualify a clause element as a complement. The phrase in question can be combined with different verbs until a verb is found that does not combine with the phrase on grammatical grounds. It is then assumed that all the other members of the commutation class have complement status (commutation classes are consistently complement or adjunct). The next step involves determining the anaphora of commutation classes in order to be able to identify commutation classes of different verbs with each other, as was shown above. (Commutation classes can have complementary distribution apart from their anaphora: e.g. eine Person einstellen vs. eine Summe Geld veruntreuen.)
I have demonstrated that the identification of adjuncts is more problematic as it is not feasible to combine a clause element with all verbs. An informed guess is required. Also, the identification of adjunct classes is less straightforward. Apart from anaphora, syntactic behaviour has to be used to establish classes. If the chosen classification is too fine, adjunct classes turn out to be +Spez (Engel 1994:101).
3.8 Syn-Sel
There are more relations that differentiate between Jacobs’ prototypical cases (cf. (3) and (4) in section 2.):
(56) (weil) Peter Herrn Meier rasierte
(57) (weil) Peter auf der Wiese schläft
Herrn Meier in (56) is "synsemantic", i.e. the phrase cannot be assigned full meaning out of context, while auf der Wiese in (57) is "autosemantic" (i.e. has full meaning out of context). Zifonun considers "synsemanticity" (Syn-Sel) as a "complement indicator" ("Komplement-Fürsprecher", 1995:185). Obviously, not all complements are +Syn:
(58) Die Aufführung beginnt um 19 Uhr. -Syn
4. Scenario constitution (Szen-Sel)
The relation that is currently favoured by valency grammarians, and the one that, apart from Info-Sel, performed best in our analysis, is Arg-Sel. The problem with Arg-Sel was that it depends on a semantic representation. A relation which is related to Arg-Sel avoids this, however. Participation in a "core scenario" (Szen-Sel) is based on Tesnière’s original insight that complements refer to entities that participate in the action, process or state denoted by the verb. "Scenario" corresponds to Sachverhalt and covers actions, states and processes. "Core" is added to indicate that the term is used restrictively, excluding any situational elements. The relation is not to be understood as extralinguistric: the division between participants and circumstances strictly depends on the meaning of the verb in question (which might be influenced by its construction). As a semantic relation, Szen-Sel is informed by pretheoretical talk about events, participation in them, circumstances of events, etc. This is the reason why linguists can have intuitions about Szen-Sel. For more or less prototypical cases, these intuitions are clear and tests are in my view unnecessary:
(59) Wahrscheinlich hat sie die Adresse im Dunkeln nicht gefunden.
In order to constitute a finden-scenario, two entities are needed: one that finds something (sie) and one that is found (die Adresse). All the other clause elements say something directly or indirectly about the core scenario: that the attempt at achieving a situation designated by the finden-scenario was unsuccessful (nicht), the condition under which the attempt took place (im Dunklen) and the speaker evaluation of the truth of the rest of the statement (wahrscheinlich).
The division between participants and non-participants looks clear-cut from examples such as the above: either an "entity" contributes to a core scenario or it does not. The idea intimates a qualitative difference between participants and non-participants. While the sphere of non-participants is differentiated (as indicated above), participants appear as equal in their function to constitute the scenario in question. Hence the insistence in many valency representations that all the complements are treated in the same way.
Though Szen-Sel is based on the same intuition as Arg-Sel, it is not exposed to theoretical arrangements outside valency theory. For example, there is no problem in classifying the referents of predicatives as +Szen. In
(60) Er nannte ihn einen Überlebenskünstler.
the referent of einen Überlebenskünstler is a participant as it clearly contributes to constituting the scenario of calling somebody something. Indeed, Szen-Sel sorts all the clause elements discussed up to now correctly (in accordance with intuition and valency dictionaries). One might say: Szen-Sel is the valency intuition. Thus Szen-Sel provides counter-evidence to Jacobs’ claim that valency has no conceptual content.
Szen-Sel can also, at least in principle, be extended to the valency of nouns and adjectives. Scenario constitution is replaced in such cases by object/quality constitution:
(61) Frau Meiers gestriger, mündlicher Antrag auf Entlassung aus dem Schuldienst
(62) sehr wild auf Erdbeeren
To constitute the entity "Antrag", an applicant and an objective are needed, but not a particular time and arguably not the mode of application. The object of desire is a constituent part of the "wild" quality while the degree to which this quality applies is not (neither is the person to whom it applies although he or she is an argument of the quality!).
The result that all there is in essence to the valency concept is a slight elaboration of Tesnière’s metaphor might be disappointing to those who have proposed that there must be a consistent syntactic reflex of this semantic differentiation. But why should this be? Possible scenarios are varied and so are their participants. They comprise participants (e.g. agent, patient, experiencer) which can hardly supply circumstantial information in any scenario and are encoded synsemantically (subject, direct object), but also participants (e.g. time, location) which typically supply circumstantial information and are autosemantically encoded in most instances, having the same form in both functions. Because of this alone, the chance that a simple/uniform syntactic reflection of Szen-Sel could be found that could be used in an operational definition — be it formal or distributional — is negligible.
The quest for an operational definition might never have received the urgent attention that it has in valency research, if Szen-Sel allowed a fairly clear division for all clause elements. This, however, is not the case. There is a considerable number of clause elements for which it is simply not clear whether they are +Szen or -Szen:
(63) Er hat ihr gestern ein Omelette gemacht.
(64) Sie hat die Flasche mit ihren Zähnen geöffnet.
Does the core scenario of (63) include the person for whom the omelette was made? Does the core scenario of (64) include the instrument with which the action was achieved? Such questions cannot be answered definitely, as one could argue either way. This, however, does not make Szen-Sel a vague concept with + and - values fading into each other as colours do in a spectrum. The relation is indeterminate for a number of cases but this does not invalidate the qualitative functional difference between participants and non-participants. If, for example, the core scenario of (63) is seen as ‘making something for somebody’, the beneficiary becomes a participant of equal footing and like the other participants it contributes to the formation of this core scenario. Thus Szen-Sel works for indeterminate cases like a toggle (the intuition switches). This means that attempts to replace the complement-adjunct distinction with several categories, for instance, by creating the category of "middles" for the indeterminate cases (Somers 1987:25-8) are — from the perspective of Szen-Sel at least — notionally problematic (apart from the fact that this simply replaces one delimitation problem by several). (Somers’ observation (1987:26) that middles are "a bit of both" (complements and adjuncts) is compatible with the interpretation that they could be seen as either.)
Szen-Sel can also be held responsible for the continued search for an operational procedure: as scenario constitution is so intuitively clear in the prototypical cases, it was perhaps not unreasonable to assume that all clause elements should attract a Szen-Sel value which would be determined by a criterion or procedure yet to be discovered. This assumption was mistaken, however: there is not even a criterion that has the same value as Szen-Sel for the clear cases. Thus Szen-Sel does not only lend semantic content to the chemical metaphor but also explains why no operational valency definition can be found.
This point is important: the one semantic relation that decides all the prototypical cases correctly allows contradictory classifications in a number of non-prototypical cases — not because it is not clear but because of its notional content. Partial indeterminism is inbuilt in valency as scenario constitution, and no operational procedure can remedy this (it can only move away from Szen-Sel by forcing a decision, as will be shown below).
How should a line be drawn between complements and adjuncts for practical purposes? Obviously in concurrence with the prototypical cases and anywhere through the indeterminate cases, depending on the aims of the investigation. If selected verb fields are to be studied, it is desirable that more clause elements be included than is the case in more conventional lexicographical studies (i.e. those which feature large numbers of alphabetically listed lexemes with different readings). Contrastive studies will include anything of interest.
To summarise, participation in a core scenario (Szen-Sel) has the following characteristics:
- it treats all participants in a core scenario equally, assigning them complement status
- it does not lead to counter-intuitive classifications
- it decides more cases correctly than any of the other relations
- it leaves a considerable number of cases undecided
- it explains discrepancies in different valency applications and the failure to find a sharp definition.
4.1 Scenario constitution and morphosyntax
The perception of elements as participants is not only decided by what is being said but also by how it is said:
(65) She lied to her parents about her grades.
The core scenario is lying to somebody about something: both the addressee and the "topic" of the lie are felt to participate in the lying-scenario. This is not necessarily so in the German translation:
(65G) Sie hat ihren Eltern gegenüber hinsichtlich ihrer Noten gelogen.
There is a greater hesitation in acknowledging the addressee and the "topic" of the lie as participants. The core scenario could be seen as one of simply lying, with addressee and topic as something that is said about this core scenario.
The difference between the English and the German sentences is that addressee and topic of the lie are more grammaticalised in the first. ihren Eltern gegenüber and hinsichtlich ihrer Noten are autosemantic, meaning the same in different contexts. They are easily converted into separate statements (Sie hat gelogen, und das geschah ihren Eltern gegenüber und hinsichtlich ihrer Noten.). Therefore they are -Arg according to Jacobs’ test (though Steinitz considers such phrases to be arguments and calls them Argument-Adjunkte (1992:37)).
The elements to her parents and about her grades are less autosemantic. Although the two prepositions have meaning, their application is much wider and thus their meaning less specific:
(66) She gave her report to her parents.
(66G) Sie hat ihr Zeugnis ihren Eltern/*ihren Eltern gegenüber gegeben.
(67) She told a story about school.
(67G) Sie hat eine Geschichte über die Schule/*hinsichtlich der Schule erzählt.
My assessment is supported by Engel & Schumacher (1976) who classify lügen as monovalent, though the gegenüber- and hinsichtlich-phrases are +Spez (as the above examples show).
Beneficiary phrases are another example of the form plane influencing the perception of certain roles as participants or non-participants:
(68) Er hat ihr einen Pullover gestrickt.
(69) Er hat einen Pullover für sie gestrickt.
Intuitively, (68) is more conducive to assuming a core scenario jemandem etwas stricken than (69) is to assigning a core scenario of etwas für jemanden stricken. Again, the reason is that cases are more grammaticalised than prepositions.
Thus the morphosyntactic encoding is not only the effect of verb meaning but it contributes to the definition of the event that is constituted.
4.2 A test procedure for Szen-Sel?
Valency grammarians who favour Arg-Sel might be happy to subscribe to Szen-Sel instead. This step was explicitly taken by Zifonun who introduced scenario constitution (Ereignisbeteiligung: Bet-Sel) as a replacement for Arg-Sel (1995:183). Her definition of Bet-Sel can be paraphrased in the following way:
A clause element X is a ‘participant’ (= "beteiligt") in relation to a valency carrier Y and a sentence S if and only if the referent of X has the role of a participant in the scenario/event devised by S because of the predication/meaning of Y.
Zifonun emphasizes that Bet-Sel is, contrary to Jacobs’ Arg-Sel, not linked to a formal representation but to semantic knowledge only (1995:183).
However, Zifonun introduces Bet-Sel as a "sharp" relation that is determinate in all cases, offering the battery of tests used in Verben in Feldern (Schumacher 1986:21) as an operational procedure:
- Elimination test: obligatory clause elements are participants. (+Real)
- Inference test: one optional clause element is omitted. If an expression containing the omitted phrase as an indefinite element can be inferred from the reduced expression (in certain contexts), the clause element is subjected to the next test. If not it is a non-participant. For example:
Gestern hat er sich über das Essen beschwert.
Reduced expression. Er hat sich über das Essen beschwert. Inferred expression Er hat sich irgendwann über das Essen beschwert. (The inference is intuitively valid, relying on world knowledge: complaining is a speech act. Speech acts take place in space and time.) Result: The adverbial phrase is tested further. - Specificity test. This test is different from Spez-Sel. Two (related) versions are offered:
A. A semantic subclassification of verbs is presumed (e.g. verbs of evaluation). Roles that are specific for the field in question (e.g. evaluator, addressee of evaluation, evaluated entity, perhaps also means and utterance of evaluation) are participants. (Cf. Schumacher 1981:145)
B. Paraphrase test: the meaning of the verb in question is paraphrased. If a paraphrase has to make reference to the clause element in question, it is a participant. For example:
sich beschweren paraphrase: x intentionally causes through uttering k, that y learns that x finds z not acceptable. result: the complainant, the addressee, the utterance and the entity complained about are all participants
The following criticism of the tests is not meant to show that any of the tested phrases are either participants or non-participants (they are all ±Szen).
4.2.1 Inference test
I will first investigate the inference test and show that
- the test can receive at least two different interpretations (1),
- that under either interpretation there is reason to question the Bet-Sel values assigned by Zifonun (1),
- the outcome is not always clear (2),
- the outcome can be seen as contradicting intuitive perception with good reason (2).
1. The outcome of the inference test depends on how the phrase "in certain contexts" is interpreted:
(70a) Die Jacke passt schlecht.
(70b) Die Jacke passt schlecht zu etwas.
(71) Die Jacke passt jemandem schlecht.
Obviously, a zu etwas-phrase can only be inferred from (70a) if the context of something matching something else is given. Otherwise (71) could be inferred instead of (70b). (70b) and (71) involve semantically closely related, but nevertheless different, readings of passen. If the proviso "in certain contexts" is meant to guarantee that the same reading applies, it is uncontroversial. However, there is reason to believe that the proviso is meant to be more restrictive, differentiating between usages within one reading with the effect that more phrases pass the inference test. Rather surprisingly, Zifonun classifies the für-phrase in
(72) Er bekommt viel Leistung für wenig Geld.
as +Bet (1995:188). The phrase is optional and cannot be inferred:
(72a) Er bekommt viel Leistung.
could for example be a comment on somebody souping up an engine. Somebody else might add Und das für wenig Geld. The value ought to be -Bet. Obviously, Zifonun had a different context in mind, possibly the field of buying and selling, but the meaning of bekommen is arguably the same in both. (Admittedly, there is no natural cut off point for the differentiation of readings but the problem should not be overestimated.) If, however, the proviso "in certain contexts" is used to exclude non-transactional contexts, the same manoeuvre should result in mir in
(73) Mir ist die Vase zerbrochen.
passing the inference test (and the paraphrase test) by restricting the context to talk about responsibilities. Zifonun classifies mir as a non-participant (-Bet) (1995:187).
In my view, the inference should be based on verb meaning to make sense, not on features of a particular situation in which the verb is used. But in both interpretations of the test, the Bet-Sel values listed by Zifonun can be questioned: under the more restrictive proviso mir in (73) ought to be +Bet instead of -Bet, under the less restrictive proviso für wenig Geld in (72) ought to be -Bet instead of +Bet. In the following, I will assume the less restrictive interpretation of the test.
2. Consider the following two sentences:
(74) Er hat ihr den Tee/schlechte Nachrichten gebracht.
(74a) Er hat den Tee/schlechte Nachrichten gebracht.
It is arguable whether a recipient/beneficiary can be inferred from (74a). If no inference is assumed, this does not necessarily have to mean that the recipient/beneficiary in (74) cannot be perceived as a participant. The core scenario of (74) might simply be different from the core scenario of (74a) (Valenzerhöhung).
If one assumes the inference of a recipient/beneficiary to be valid (as every act of taking something involves taking something to somebody or for the benefit of somebody, often oneself) then the test would assign a beneficiary to most verbs: every act of doing something is for the benefit of somebody.
As was pointed out above, Zifonun classifies mir in
(73) Mir ist die Vase zerbrochen.
as (-Bet), presumably as Die Vase ist zerbrochen. does not imply that somebody caused it or is responsible for it. But even under the less restrictive test interpretation, -Bet might not be the appropriate value. As with (74), we might wonder whether two different scenarios are not being played off against each other: the dative is introduced to describe a quite different event, one in which the speaker sees somebody as responsible for what has happened. It is certainly feasible, and in my view intuitively more convincing, to judge the dative phrase as encoding a participant. Compare with a possible English translation:
(73E) I broke the vase.
Zifonun’s split of (73) into an event of the vase breaking and an entity for which the event has a certain function (responsible person) is problematic. Such an analysis could be carried out for many verbs:
(75) Er rollt den Ball an die Wand.
could be analysed into an event of the ball rolling to the wall and an entity that causes it.
(76) Ihr fehlt ein guter Computer.
Likewise, (76) could be broken up into an event of a computer lacking and this event having an adverse effect on her. However, what can be divided up extralinguistically is of less interest than what is drawn together by a particular verb (usage).
The application and the results of the inference test are not always clear. The test might exclude clause elements that refer to intuitive participants. Being not only dependent on semantics but also on general (and possibly contextual) world knowledge, the inference test moves scenario participation towards a relation between extralinguistic objects.
4.2.2 Specificity test
The decisive test is in many cases the "specificity test". In the A version, the test of proof is shifted towards establishing verb fields. This means that it is not so much an individual verb but rather a whole scene that determines participation. The different perspectives of verbs on a scene are levelled and Szen-Sel is again pushed towards an extralinguistic rather than a semantic relation.
The paraphrase test (B) has one clear advantage over the A version – field constitution can be motivated in this way. As an explication of verb meaning, it aims exactly at Szen-Sel. However, a number of factors might interfere with the result:
- the chosen format for paraphrases
- the chosen semantic vocabulary, in particular the valency of verbs.
The chosen format for paraphrases insures consistency of application, especially across fields. The format of the paraphrase of sich beschweren above (cf. beginning of 4.2) results in four participants. However, equally valid paraphrases can be suggested with different results:
(77) x says that z is not acceptable to him
(78) x expresses his anger about z to y by uttering k through the medium m
(77) indicates two participants, (78) five. Though the differences concern only non-prototypical complements, the test might suggest a certain precision about Szen-Sel which, in fact, it does not possess.
The tests favoured by Schumacher and Zifonun are only of heuristic value. Studying verb meaning without set procedure yields no worse if not better results.
4.3 Scenario constitution as a gradable relation
It was established that participation (Szen-Sel) has the following characteristics as a non-gradable relation (cf. 4):
- it treats all participants in a core scenario equally, assigning them complement status
- it does not lead to counter-intuitive classifications
- it decides more cases correctly than any of the other relations
- it leaves a considerable number of cases undecided
- it explains discrepancies in different valency applications and the failure to find a sharp definition.
Scenario constitution/participation as a gradable concept can either be seen as replacing Szen-Sel and thus abolishing the complement-adjunct-divide or as complementing Szen-Sel by ranking complements only according to degree of participation (Szeng-Sel). There are various possible bases for the gradation. An obvious way to grade participation is to use speaker expectation. This is captured in Info-Sel. Heringer’s association test aimed to show the validity of the valency concept, in particular that valency is semantic, relational and gradual (1986:81). He describes the semantic phenomenon of valency thus: a verb and its complements (as arguments) form a propositional core which is predicated by adjuncts. This is virtually identical to Szen-Sel. Heringer tested 20 verbs (most of which allowed only one semantic reading) on 100 speakers. Speakers were given the verbs in the infinitive and were asked to associate questions or question words. Those with low frequency were excluded from the result; those that could be considered as obvious variants were sometimes pooled (e.g. warum? weshalb? etc. under warum?). Frequency, mean rank and mean latency of the question words were calculated and their associative verb proximity was arrived at through the formula: 10 x mean latency ÷ frequency. For example:
verkaufen |
was .5 wer 1.2 an wen 1.6 wem 2.3 |
wo 4.8 wie 7.3 wieviel 7.4 warum 8.5 |
wann 9.4 wofür 10.8 wohin 13.9 wen 15.0 |
wozu 15.2 |
kaufen |
was .7 wer 1.3 warum 4.1 |
wo 4.4 wann 6.5 wem 7.0 |
wozu 7.1 wieviel 7.7 wie 15.3 |
womit 15.8 weshalb 20.0 |
schulden |
wem .9 wer 1.1 |
was 1.6 wieviel 2.4 |
warum 4.2 weshalb 14.4 |
für wieviel 20.0 |
heißen |
wer 1.0 wie 1.1 |
wen 4.2 warum 6.9 |
was 8.4 wann 15.0 |
Heringer assumed that the association test measured semantic proximity. As traditional complements (with the exception of the genitive complement) are more closely associated with verbs than traditional adjuncts, and as there is a noticeable jump in proximity values between the two categories, Heringer concluded (a) that valency is indeed based on meaning and (b) that the valency approach is vindicated in separating complements from adjuncts. However, as proximity values are on a scale, Heringer insisted that valency is a graded phenomenon:
Figure 1 [cf. the table above] shows that the essential questions with verkaufen are wer? was? wem? They aim at just the slots that are traditionally postulated in valence theory. In addition, our satellite system [= graphic representation of test results] shows a jump between complements and supplements, i.e. wer? was? wem? on the one hand and the remaining questions on the other. This jump, a gap, is more or less wide for different verbs and does not justify two different syntactic categories, because that would presuppose an absolute distinction where we find a transition, as has been expected within the theory of associative meaning. (Heringer 1986:86/89)
Certainly, we do not expect a categorial distinction, but this is a clear criterion of degree for the different semantic functions of complements and supplements in sentences. Hence valence theory has not been chasing a chimera, but represents an adequate intuition with respect to linguistic facts. (Heringer 1986:97)
The different values of question words like wie? with different verbs show that valency is relational (the proximity values are verb-specific).
Heringer’s results can be used to grade Szen-Sel. Szeng-Sel and Infog-Sel would merge and be extended to adjuncts. However, the marked difference in speaker association of traditional complements and traditional adjuncts could — with Jacobs (1994:29) — be taken as evidence for a binary divide with complements and adjuncts, each graded within their categories.
Another way to introduce gradation is to base it on morphosyntactic information: on morphological case or syntactic functions. Intuitively, the subject might be felt to have a greater stake in the description of an action than objects and the direct object a greater stake than the indirect object as the traditional terminology suggests. This intuition can be backed up with the hierarchy of "grammatical accessibility of clause elements" (subject > direct object > indirect object > prepositional object/adverbial for nominative languages) (Comrie 1981:148-57, Primus 1993) if the hierarchy is understood as semantically relevant (comparing referents that the statement is about with referents that are mentioned for the sake of the former; cf. Ickler 1990).
I should emphasize that I mention the grammatical accessibility hierarchy only in order to distinguish it from the valency approach. No in-depth assessment or discussion of the hierarchy is intended, and I will concentrate on its presumed semantic effects. These are best illustrated by an example:
(79) Das Reisebüro hat dem Hotel die Exkursion bereits bezahlt.
(80) Das Reisebüro hat das Hotel für die Exkursion bereits bezahlt.
(81) Das Reisebüro hat dem Hotel bereits 10 000 Mark für die Exkursion bezahlt.
The claim is that by virtue of the respective construction, different clause elements are focused upon: first and foremost the subject in all three sentences, then the accusative object (the goods in (79), the vendor in (80), the price in (81)). (79) is mainly about the travel agency and the excursion and arguably to a lesser degree about the hotel:
(79a) Das Reisebüro hat die Exkursion bereits bezahlt.
is possible, but not:
(79b) *Das Reisebüro hat dem Hotel bereits bezahlt.
This could be taken as evidence that dem Hotel is not mentioned "for its own sake" — (80) is about the travel agency and the hotel (though the claim cannot be backed up through the omission test), etc. The existence of such syntactically induced semantic nuances has been disputed (cf. Wegener 1998:80-3). The problem is that the semantic effect of the accessibility hierarchy is easily overridden by changes of stress, order of clause elements, context and situation and perhaps also by the meaning of individual lexemes.
The language used to describe the semantic effects of the grammatical accessibility hierarchy is very similar to the language that could be used to describe a graded version of Szen-Sel. Thus, the hierarchy could be seen as reflecting the degree of participation (Szeng-Sel). Alternatively, it could be seen as a separate relation that introduces a "perspective" on participants in addition to their participation in constituting a scenario (denotative vs. signifikative Bedeutung). The question is whether we are we dealing with one semantic phenomenon or two. (Zifonun introduced "perspective" (PerspZ-Sel) as a separate relation from Bet-Sel (1995:183f). Her concept of perspective is directly semantic and therefore independent of the grammatical accessibility hierarchy. It always coincides with Bet-Sel.)
Though one might feel that both scenario constitution (Szen-Sel) and the semantic effect of the grammatical accessibility hierarchy are feeding on the same semantic intuition, there is an independent reason why the two should not be equated. The accessibility hierarchy is based on categorial information and the semantic effect must be the same in different context. But Szen-Sel can have different values for the same category (e.g. adverbials), depending on the governing verb: valency cannot be based principally on the grammatical accessibility hierarchy! Therefore, semantic intuitions aside, a separate relation has to be established (Persp-Sel). Needless to say that Persp-Sel would not separate complements from adjuncts satisfactorily.
Having established a second gradable semantic relation, it must be asked what exactly Heringer’s test measured and whether it necessarily showed that valency is a graded phenomenon. Leaving possible interference aside (connected with, for example, the infinitive form of the tested verb, or the order of question words learnt at school) either Szen-Sel or Persp-Sel might have led to the results: Szen-Sel might be responsible for the proximity gap (the tested speakers might be trying to construct a core scenario), Persp-Sel for the ranking (among the complements). In that case Szen-Sel could be defended as an absolute relation defining valency as Persp-Sel does not lead to a correct separation of complements from adjuncts. Persp-Sel would join the other inadequate relations in being used for the description of valency but not for its definition.
To summarise: participation in constituting a scenario can — but does not have to — be seen as graded, arguably extending to (part of) the traditional adjuncts. Syntactically induced semantic ranking of clause elements, though largely coinciding with graded participation, has to be analysed as a separate relation.
4.4 Verb valency
Valency partners are best justified semantically through their participation in scenario constitution (Szen-Sel). The fact that Szen-Sel is indeterminate for a number of cases should not be seen as a defect to be amended by a superior relation which is yet to be found but as an explanation for the failure of the valency research community to agree on a sharp division between complements and adjuncts.
Szen-Sel (whether graded or not) draws together heterogeneous clause elements that are not selected by any of the other relations that have been considered. For some of these relations there is a principled reason why they should fail to replicate Szen-Sel: valency is a relational concept that applies to clause elements regardless of their categorial status. Rel-Sel, Spez-Sel, Syn-Sel and Persp-Sel are category-based and therefore must select a different set of clause elements from valency. The reason why these categorial relations perform reasonably well is that most valency-selected clause elements belong to commutation classes (categories) that cannot be used for adjuncts as they are synsemantic. Thus categorial classifications and valency overlap.
Form selection itself is relational, but to the extent that phrases with shared form features have to receive the same Form-Sel or Rekt-Sel value, a categorial classification is admitted through the back door. In any case, none of the non-categorial relations produces a satisfactory description of valency partners even for prototypical cases.
The argument has led to the following understanding of valency: Szen-Sel (whether graded or not) defines valency and gives the term conceptual content. A large number of undecided cases are acknowledged and these will be decided pragmatically, using the aim of the description as a criterion. The other relations are used to describe valency. They are independent in that none of them picks out the same set of clause elements but their co-occurrence is of course not accidental but due to the way in which scenarios can be expressed through language in general and are expressed in a particular language (cf. Zifonun 1995:184f): +Szen elements are in most cases +Arg because, without reference to the entities that play a role in scenarios, nothing much could be said. They are also +SortR as most entities can play different roles in different scenarios and the particular role in a specific scenario must be assigned. Not all entities are eligible to play every role that verbs require; therefore +Szen elements are +SortM in most cases. As events can be described through different scenarios, some participants are more focused than others and are +Persp (to varying degree). These semantic relations can be subsumed under the familiar term semantic valency if the particular quality of a relation is considered (e.g. not just the fact that a semantic role is assigned, but which role it is). To some extent no particular encoding of these relations is required, as verb meaning and autosemantic complements secure the identity of the scenario. But formal differentiation of synsemantic complements makes language more situation-independent and facilitates processing. Thus Form-/Rekt-Sel encode aspects of semantic valency, keeping a balance between the form effort and the semantic differentiation achieved. Form-/Rekt-Sel, Real-Sel and Spez-Sel are usually described as syntactic valency.
It is a commonplace that syntactic and semantic valency are not in a one-to-one-relationship. The statement is based on a comparison of SortR-Sel and Form-/Rekt-Sel: semantic roles cannot be mapped onto morphosyntactic classes. The matching improves significantly, however, if combinations of roles and complements are considered. For example with sentences of the structure sub verb dat akk, an agent can only be mapped onto the subject, a recipient or "donor" only onto the dative complement, a patient only onto the accusative complement:
(82) Sie wirft ihm den Ball zu.
(83) Sie nimmt ihm den Ball ab.
Not only SortR-Sel but the other semantic relations also determine syntactic valency. SortM-Sel influences the choice between competing encodings:
(84) Sie hat dem Freund/an das Finanzamt geschrieben.
Persons are more likely to be encoded as dative in (84), institutions as a prepositional an-phrase.
Persp-Sel is by definition linked to morphosyntactic classes which can be seen to encode or effect Persp-Sel or both. Unless Persp-Sel is rejected, it cannot be claimed that the assignment of morphosyntactic classes is ever independent from or neutral to meaning. Even ignoring Persp-Sel, syntactic valency is largely an expression of the remaining semantic relations. Therefore I cannot see much sense in playing syntactic valency off against semantic valency. Obviously one can define a syntactic relation and call it valency but then it should be applied consistently and not lead to the traditional valency classification.
The different selectional relations turn out to be much more closely linked (both conceptually and factually) than Jacobs originally assumed when he relegated valency to the status of a mere cover term. (In the 1993 epilogue to Kontra Valenz, he admitted that the valency concept does not just label a loose collection of phenomena but that it refers to the frequent and prototypical co-occurrence of these phenomena (1994:71).)
5. Verb valency — a multi-faceted concept?
Given the above, should we not take advantage of the links between the selectional relations and declare valency a multi-faceted concept, a "family resemblance" term (cf. Wittgenstein 1958: para. 66f)? Like the members of a family, the referents of the valency concept would be linked through overlapping features, without any one feature applying exclusively to all of them. A positive value for any relation would increase the valency rating for a clause element, a negative one decrease it. The most prototypical complements would only have positive values. Zifonun championed such an approach in her response to Jacobs’ attack which outlines the concept of valency used in the new Grammatik der deutschen Sprache of the Institut für deutsche Sprache (Zifonun, Hoffmann & Strecker 1997). She differentiates between morphosyntactic and semantic relations in the usual fashion and attributes particular importance to the co-occurrence of the two for assigning complement status to a clause element:
Wesentlich jedoch für die Zuerkennung oder Aberkennung des Komplementstatus ist die Bündelung von Relationen jeweils auf einer der beiden Seiten und der Bezug zwischen den Relationen bzw. Bündelungen auf der Form- und auf der Bedeutungsseite. (Zifonun 1995:181)
Zifonun uses a number of form relations. Firstly there is Rekt-Sel (cf. 3.4.1) which includes the controversial adverbial complements (wohnen in ...). In order to differentiate fixed prepositions from these, a sub-relation "Konstanz" is introduced (Konst-Sel: preposition fixed by a paradigm category of the verb). Other form relations are: "Fixiertheit" (= Real-Sel) which has two different positive values, and "Autonome Kodierung" (Autokod) which corresponds to -Syn. The latter is introduced as a one-place predicate and applies if the kind of encoding implies a "sentence-semantic information" that is also valid in other contexts. Although Autokod addresses the form-meaning relationship, it is claimed for the form side. (However, the form of a phrase does not necessarily decide whether it is synsemantic or not. For example: Er erinnert sich des Abends gut. (+Syn) vs. Des Abends kamen ihm Zweifel. (—Syn). The commutation class which a phrase belongs to has to be considered. This is why I introduced synsemanticity as a relation.)
The plane of meaning features just three relations. Bet-Sel is considered to be a sharp relation based on the tests discussed above. Though this does not alter the fact that +Bet (like +Szen) elements are always complements, it might have a bearing on the indeterminate cases (where it is intuitively unclear what the core scenario is). The other two are "Perspektivierung" (with two positive values; PerspZ-Sel) and "Ereigniskontextualisierung" (Kont-Sel), the complementary relation to Bet-Sel.
Only the relations of the form plane are independent of each other; the relations of the meaning plane are not:
| +Bet | iff | ++/+PerspZ |
| +Bet | iff | -Kont |
Thus Bet-Sel suffices to represent the meaning plane.
A number of regularities pertain between the planes of form and meaning:
| +Rekt / +Konst / +Real | prototyp | +Bet |
| -Rekt / -Konst / -Real | prototyp | -Bet |
| +Autokod | prototyp | -Bet |
| -Autokod | prototyp | +Bet |
| +Autokod | markiert | +Bet |
| +Rekt | markiert | -Bet |
These regularities still exist if Bet-Sel is replaced by Szen-Sel though it is doubtful whether there would be any instances of the last one (given my claim that the referents of grammaticalised clause elements tend to be perceived as participants). By listing the above regularities, Zifonun has responded to Jacobs’ suggestion that valency theory should address the links between the various "valency relations". Zifonun offers the following explanation of the observed regularities (1995:184-5):
Sachverhaltsentwürfe sind in der Regel sprachlich nur dann herstellbar, wenn Prädikatsausdrücke, die Charakteristika denotieren, kombiniert werden mit n (n ³ 1) Ausdrücken, die irgendwelche beteiligte Gegenstände unterschiedlichster Art entwerfen: Auf diese Weise wird spezifiert, was (welches Charakteristikum) wofür (für welche Gegenstände) gelten soll. Vereinfacht gesehen, konstituiert nun der Verbkomplex den Prädikatsausdruck, eine NP in einem regierten Kasus konstituiert im prototypischen Fall einen Beteiligten- und Argumentausdruck. Eine solche kasusmarkierte NP hat also zweierlei Funktionen: den Entwurf eines Gegenstands und die Markierung einer relationalen Bedeutung. Mit letzterer wird indiziert, daß überhaupt eine semantische Relation vorliegt; es wird jedoch nicht angegeben, welcher Art diese Beziehung ist. In diesem Sinne kodiert die NP nicht auf autonome Weise eine satzsemantische Information.
I should make clear that my reservations about Zifonun’s approach apply neither to the observed regularities nor to the quoted explanation. Indeed, scenario constitution is seen as the cause of the various relations and regularities (which is not borne out in the way Zifonun "calculates" valencies below).
+Rekt, +Konst and -Autokod (+Syn) are valency indicators of the form plane, +Bet (+Szen) is the one on the content plane. Convergence of form and content indicators results in core complements or supplements (adjuncts), lack of convergence results in peripheral complements. Strong convergence means that all four indicators support valency status, weak convergence that one of the form indicators supports supplement status.
In the following adaptation of Zifonun’s table (1995:188) the abbreviations used in this article have been put in square brackets (where they are different) with ± representing indeterminate cases. (The numbers in square brackets are the ones used by Zifonun.)
(Two example sentences that were obviously under the wrong heading have been exchanged. I assume that +Rekt under [4a] has been bracketed to express reservations about the dative being verb selected.)
The actual complement classifications shown are all within the range of values given by Szen-Sel and thus acceptable in this respect. What is problematic is the precision and definiteness of allocation that the table suggests. Does seeing valency as a multi-faceted concept really entail that, for example, the peripheral complements cannot be classified as adjuncts?
The assignment of individual values for relations is problematic, as was indicated above: +Rekt for [2a] but -Rekt for [3b] (cf. 3.4 sentence (30) in conjunction with discussion in 3.4.1 and 3.4.1 (31)), +Bet for [3a] (cf. 4.2 sentence (14)) and -Bet for [1b] (cf. 4.2 sentence (15)).
My reservations are also of a conceptual nature: the fact that Bet-Sel has a positive value for all the complements but one, and that it was chosen as the relation for the meaning plane (rather than PerspZ-Sel), suggests a special status for Bet-Sel — also over and above the form relations — which is not acknowledged. This assessment receives additional support from the controversial status of the one -Bet value for a complement [1b], controversial both in relation to Zifonun’s operational procedure and in comparison with Szen-Sel.
As expected, the Szen-Sel values show this relation as the only one not to contradict the achieved complement-adjunct divide.
The point of Zifonun’s table is to show convergence. Apart from Bet-Sel, Rekt-Sel is a strong performer showing only one negative value for a complement [3a]. However, Rekt-Sel’s positive values for [2a] and [4a] are very problematic as Zifonun herself indicated. (Zifonun 1995:179. +Konst for [4a] is not controversial. If Konst-Sel is seen as a subrelation of Rekt-Sel, the +Rekt value can be justified in that way.) Had a negative value been given, there would have been no convergence in these cases. This would have increased the extent to which the values of the form relations and Bet-Sel do not coincide.
My main reservation concerns the kind of concept that valency becomes under Zifonun’s approach. She admits that valency is not an absolutely unitary concept under her interpretation ("kein in absolutem Sinne einheitliches Konzept", Zifonun 1995:189). Given how the relations are defined and given the values that are assigned, valency looks like a true "family resemblance" concept. This kind of concept is better than none — much of our daily dealings rely on them — but science aims at moving from such concepts to "classical definitions" (through genus proximum and differentia specifica). In the case of valency, such a definition is possible using Szen-Sel. Though partially indeterminate, the relation leaves valency an absolutely unitary concept. The other relations introduced by Jacobs and Zifonun can be used to describe instances of valency, and the implicational relations shown above can be established on that basis. Description and implications can be summarised as "valency phenomena/effects" and kept separate from the valency definition. There is also no reason to stop talking about syntactic, semantic, logical and pragmatic valency as long as it is clear that this is meant as description and not as definition.
Though Zifonun strongly stresses that her concept of valency is structured with the different valency indicators being prototypically (and also causally) linked, she differs from Jacobs’ conclusions in emphasis rather than in principle: the individual "partner binding relations" are the ones that are real and relevant, their summative labelling as valency (when strong, weak or no convergence occurs) is little better than a cover term. As Bet-Sel is not given a special status in determining valency (-Bet can be compensated) the "calculation" of valency values from the individual relations might provoke the misunderstanding that valency was something separate from them and responsible for their values.
The main attraction of Zifonun’s approach is the matching of the planes of form and meaning. But to the extent that the form plane expresses scenario constitution in a unitary fashion, using the form relations for the definition of valency means doubling up. To the extent that participants do not receive prototypical form encoding, the form plane is irrelevant. Only if the form plane encodes semantically indeterminate elements like prototypical complements, does Zifonun’s approach go beyond a semantic concept of valency. However, if my argument is accepted that the form of encoding influences the semantic perception, those cases are also catered for by Szen-Sel.
6. Conclusion
Clearly, we cannot prove that one valency concept is right and another wrong. However, if the practice of valency grammarians is taken as the arbiter, only the concept of scenario constitution based on Tesnière’s original metaphor can motivate their choices. A syntactic valency concept based on form determination would either have to exclude adverbials from complement status or promote adverbial adjuncts to it. Thus, the definition of valency must be kept separate from its description in which syntactic notions feature very prominently.
Scenario constitution is a two-valued, semantic concept that is partially indeterminate and thus explains the borderline cases that have caused so much controversy. It disproves Jacobs’ claim that valency is void of intuitive conceptual content. It is also preferable to a multi-faceted definition of valency which necessarily turns valency into a hybrid concept. We should acknowledge the advances in the understanding of valency which have come from Jacobs’ listing and definition of individual valency relations and from Zifonun’s study of the regularities which exist between them. However, doubts remain about the rationale and consistent application (and applicability) of various suggested tests. Given that a clear-cut decision in each case is only something that has been asserted and that test results might be counter-intuitive, it is all the more defensible to forego operational procedures in the assignment of valency values and to rely on intuitive (but expressible) semantic knowledge.
The indeterminate cases that scenario constitution allows are not considered a weakness of the criterion as no alternative criterion exists to amend this indeterminacy: it is part of the concept of valency as scenario constitution. Any forced decision is arbitrary, mixing valency with something else.
Bibliography
Bianco, Maria T., 1996. Valenzlexikon deutsch-italienisch (= Deutsch im Kontrast 17). 2 vols. Heidelberg: Groos.
Comrie, Bernard, 1981. Language Universals and Linguistic Typology. Oxford: Blackwell.
Eichinger, Ludwig M. & Hans-Werner Eroms (eds.), 1995. Dependenz und Valenz. Hamburg: Helmut Buske.
Eisenberg, Peter, 1994. Grundriß der deutschen Grammatik. 3rd rev.ed. Stuttgart: Metzler.
Engel, Ulrich, 1970. Die deutschen Satxbaupläne. In: Wirkendes Wort 20, 361-92.
-- 1988. Deutsche Grammatik. Vollständige Darstellung der deutschen Gegenwartssprache. Heidelberg: Groos.
-- 1994. Syntax der deutschen Gegenwartssprache. 3rd rev.ed. Berlin: Schmidt.
-- 1996. Semantische Relatoren. Ein Entwurf für künftige Valenzwörterbücher. In: N. Weber (ed.), Semantik, Lexikographie und Computeranwendungen. Tübingen: Niemeyer. 223-36.
Engel, Ulrich, Mikai Isbasescu, Speranta Stanescu & Octavian Nicolae, 1993. Kontrastive Grammatik deutsch-rumänisch. Heidelberg: Groos.
Engel, Ulrich & Pavica Mrazovic (eds.), 1986. Kontrastive Grammatik deutsch-serbokroatisch. 2 vols. München: Sagner.
Engel, Ulrich, Emilia Savin et al., 1983. Valenzlexikon deutsch-rumänisch. Heidelberg: Groos.
Engel, Ulrich & Helmut Schumacher (eds.), 1976. Kleines Valenzlexikon deutscher Verben. 2nd rev.ed. 1978. Tübingen: Narr.
Engelen, Bernhard, 1975. Untersuchungen zu Satzbauplan und Wortfeld in der geschriebenen deutschen Sprache der Gegenwart. Vol 2: Verblisten. München: Hueber.
Eroms, Hans-Werner, 1981. Valenz, Kasus und Präpositionen. Untersuchungen zur Syntax und Semantik präpositionaler Konstruktionen in der deutschen Gegenwartssprache. Heidelberg: Carl Winter.
Fillmore, Charles, J., 1968. The case for case. In: E. Bach & Robert T. Harms (eds), Universals in Linguisic Theory. New York, Chicago and London: Holt, Rinehart & Winston. 1-88.
Fischer, Klaus, 1997. German-English Verb Valency. A Contrastive Analysis. Tübingen: Narr.
Helbig, Gerhard & Wolfgang Schenkel, 1969. Wörterbuch zur Valenz und Distribution deutscher Verben. 2nd rev. ed. 1973. 6th ed. 1982. Leipzig: Bibliographisches Institut. 8th ed. 1991. Tübingen: Niemeyer.
Heringer, Hans J., 1984. Neues von der Verbszene. In: G. Stickel (ed.), Pragmatik in der Grammatik. Jahrbuch 1983 des IDS. Düsseldorf: Cornelsen. 34-64.
-- 1986. The Verb and Its Semantic Power. Association as a Basis for Valence Theory. In: Journal of Semantics 4. 79-99.
-- 1987. Was lange währt. Gedanken zum Mannheimer Valenzwörterbuch. Review. In: ZGL 15.3. 311-7.
-- 1996. Deutsche Syntax. Dependentiell. Tübingen: Stauffenburg.
Hudson, Richard A., 1993. Approaches to a Theory of Syntax III: Dependency Syntax. 14. Recent Developments in Dependency Theory. In: J. Jacobs et al. (eds.), Syntax. An International Handbook of Contemporary Research (= Handbücher zur Sprach- und Kommunikationswissenschaft vol. 9.1). Berlin: de Gruyter. 329-38.
Ickler, Irene, 1990. Kasusrahmen und Perspektive. Zur Kodierung von semantischen Rollen. In: Deutsche Sprache 18. 1-37.
Jacobs, Joachim, 1994. Kontra Valenz. (=Fokus vol. 12). Trier: Wissenschaftlicher Verlag.
-- 1992. Syntax und Valenz. In: L. Hoffmann (ed.), Deutsche Syntax. Ansichten und Aussichten. Berlin: de Gruyter. 94-127.
Primus, Beatrice, 1993. Syntactic Relations. In: J. Jacobs et al. (eds.), Syntax. An International Handbook of Contemporary Research (= Handbücher zur Sprach- und Kommunikationswissenschaft vol. 9.1). Berlin: de Gruyter. 686-705.
Radford, Andrew, 1988. Transformational Grammar. A First Course. Cambridge: CUP.
Schumacher, Helmut (ed.), 1986. Verben in Feldern. Valenzwörterbuch zur Syntax und Semantik deutscher Verben (= Schriften des Instituts für deutsche Sprache vol.1). Berlin: de Gruyter.
Schumacher, Helmut et al., 1981. Konzeption eines Wörterbuchs deutscher Verben. Zu Theorie und Praxis einer semantisch orientierten Valenzlexikographie. Tübingen: Narr.
Sommerfeldt, Karl-Ernst & Herbert Schreiber, 1974. Wörterbuch zur Valenz und Distribution deutscher Adjektive. Leipzig: Bibliographisches Institut. 3rd ed. 1983 Tübingen: Niemeyer.
-- 1977. Wörterbuch zur Valenz und Distribution der Substantive. Leipzig: Bibliographisches Institut. 3rd ed. 1983. Tübingen: Niemeyer.
-- 1996. Wörterbuch der Valenz etymologisch verwandter Wörter. Verben, Adjektive, Substantive. Tübingen: Niemeyer.
Somers, H.L., 1987. Valency and Case in Computational Linguistics. Edinburgh: Edinburgh University Press.
Steinitz, Renate, 1992. „Modern": Argumentstruktur, „Traditionell": Valenz — Versuch eines Brückenschlags. In: Linguistische Berichte 137. 33-44.
Storrer, Angelika, 1992. Verbvalenz. Theoretische und methodische Grundlagen ihrer Beschreibung in Grammatikographie und Lexikographie. Tübingen: Niemeyer.
Tesnière, Lucien (1966): Éléments de syntaxe structurale. 2nd rev.ed. Paris: Éditions Klincksieck.
Varnhorn, Beate, 1986. Nochmals: Ergänzungen und Angaben. Forschungskritik und ein neuer Vorschlag. In: Deutsche Sprache 14. 1-11.
Vater, Heinz, 1978. On the Possibility of Distinguishing between Complements and Adjuncts. In: W. Abraham (ed.), Valence, Semantic Case, and Grammatical Relations. Amsterdam: Benjamins. 21-46.
Wegener, Heide, 1998. Die Kasus des EXP. In: M. Vuillaume (ed.), Die Kasus im Deutschen. Form und Inhalt. Tübingen: Stauffenburg. 71-84.
Welke, Klaus M., 1988. Einführung in die Valenz- und Kasustheorie. Leipzig: Bibliographisches Institut.
Wittgenstein, Ludwig, 1958. Philosophical Investigations. Oxford: Blackwell.
Zifonun, Gisela, 1995. Wieviele Valenzrelationen braucht eine Grammatik? In: L.M. Eichinger & H.-W. Eroms (eds.), Dependenz und Valenz. Hamburg: Helmut Buske. 177-89.
Zifonun, Gisela, Ludger Hoffmann & bruno Strecker et al., 1997. Grammatik der deutschen Sprache I-III. Berlin: de Gruyter.
Return to The Issue 4-5 Contents Page | The WJMLL Home Page