Verb valency — an attempt at conceptual clarification


Klaus Fischer

London Guildhall University


(revised October 1999)

Copyright Notice:
First published in Web Journal of Modern Language Linguistics. © 1999 Klaus Fischer.
The moral rights of the author(s) to be identified as author(s) of this work are asserted in accordance with §§.77 and 78 of the Copyright, Designs and Patents Act 1988. This work may be reproduced without the consent of the author, in part or in whole in any manner and in any medium subject only to the two following conditions:
(a) no charge shall be made for the copy containing the work or the excerpt,
(b) a copy of this notice shall precede the work or the excerpt.


0 Introduction
1 Developments before Jacobs
2 Jacobs’ attack on valency
3 Which relations are valency relations?
3.1 Real-Sel
3.2 Rel-Sel
3.3 Arg-Sel
3.3.1The tun/geschehen test
3.4.1 Rekt-Sel
4.Scenario constitution (Szen-Sel)
4.1 Scenario constitution and morphosyntax
4.2 A test procedure for Szen-Sel?
4.2.1 Inference test
4.2.2 Specificity test
4.3 Scenario constitution as a gradable relation
4.4 Verb valency
5 Verb valency — a multi-faceted concept?

6 Conclusion


0 Introduction

Valency has come of age on both sides of the former inner-German border. It now forms the basis of a considerable body of lexicographical and grammatical work. The table below shows a selection of valency dictionaries and contrastive grammars based on valency:

valency dictionaries for verbs:

Wörterbuch zur Valenz und Distribution deutscher Verben by G. Helbig and W. Schenkel (1969, 2nd rev.ed. 1973, 8th ed. 1991)

Kleines Valenzlexikon deutscher Verben by U. Engel and H. Schumacher (1976, 2nd rev. ed. 1978)

Verben in Feldern by H. Schumacher (ed.) (1986)

valency dictionaries for adjectives:

Wörterbuch zur Valenz und Distribution deutscher Adjektive by K.-E. Sommerfeldt and H. Schreiber (1974, 3rd ed. 1983)

valency dictionaries for nouns:

Wörterbuch zur Valenz und Distribution der Substantive by K.-E. Sommerfeldt and H. Schreiber (1977, 3rd ed. 1983)

valency dictionaries for related words:

Wörterbuch der Valenz etymologisch verwandter Wörter. Verben, Adjektive, Substantive by K.-E. Sommerfeldt and H. Schreiber (1996)

contrastive valency dictionaries:

Valenzlexikon deutsch-rumänisch by U. Engel, E. Savin et al. (1983)

Valenzlexikon deutsch-italienisch by M.T. Bianco (1996)

contrastive grammars:

Kontrastive Grammatik deutsch-serbokroatisch by U. Engel and P. Mrazovic (eds.) (1986)

Kontrastive Grammatik deutsch-rumänisch by U. Engel, M. Isbasescu, S. Speranta and O. Nicolae (1993)

Valency has played a central role in dependency grammar but largely equivalent concepts (subcategorisation, complementation, lexical government) also feature in grammars that use constituency to represent syntactic structure (Standard Theory of generative grammar, Government and Binding Theory, Generalised Phrase Structure Grammar, Construction Grammar, etc.). Those concepts gained greater weight as the importance attributed to the lexicon grew, bridging to some extent the theoretical gap with the valency approach (cf. Hudson 1993). Indeed, it is difficult to see how a syntactic theory could do without encoding information about the syntactic surroundings of lexemes (or groups of lexemes) which differentiate their valency partners — the complements — from the non-valency elements — the supplements or adjuncts. Radford, for example, discusses complementation at length, using a number of the criteria investigated in this article to differentiate English verb complements from adjuncts (1988:230-41).

Despite this success, it was felt that the concept of valency needed further conceptual clarification: Heringer was puzzled by the fact that practical applications had been developed before the concept of valency was sufficiently grounded in linguistic theory and researchers had come to a shared understanding of valency (1987:311). Vater (1978) suggested that, in the absence of a valid criterion to differentiate complements from adjuncts, the differentiation should be dropped. This article anticipates many of Jacobs’ criticisms discussed below but the supporting evidence is not always convincing. Jacobs coined the provocative term "Valenzmisere" (1994:5) and questioned the validity of the concept altogether. And in the preface to Dependenz und Valenz, the proceedings of the 1992 symposium of the same name, Eichinger and Eroms drew attention to a deficit which, they claim, marks out dependency and valency theory in comparison with other grammatical models, and which the symposium had started to address (1995:9).

Taking Jacobs’ attack on the concept of valency as its point of departure, this article aims to defend valency as a valid concept based on semantic intuitions — albeit one with very fuzzy edges. I will suggest that the definition of valency should be separated from the description of valency partners, and will argue that a definition that does not contradict prototypical valency and non-valency cases, but is partially indeterminate, is preferable to a sharp multi-faceted definition that relies on criteria which individually do contradict prototypical cases. I will also try to show why no satisfactory distinguishing criterion can be found.

1 Developments before Jacobs

This necessarily brief and selective account attempts to highlight the main developments before Jacobs’ criticism of the valency concept. It is generally recognised that the credit for introducing the term valency into linguistics has to be attributed to Lucien Tesnière. As is well known, it was originally used as a technical term in chemistry to denote the number of atoms of hydrogen an atom or chemical group is able to combine with to form a chemical compound. Tesnière puts it like this:

On peut ainsi comparer le verbe à une sorte d’atome crochu susceptible d’exercer son attraction sur un nombre plus ou moins élevé d’actants, selon qu’il comporte un nombre plus ou moins élevé de crochets pour les maintenir dans sa dépendance. Le nombre de crochets que présente un verbe et par conséquent le nombre d’actants qu’il est susceptible de régir, constitue ce que nous appellerons la valence du verbe. (Tesnière 1966:238)

The chemical metaphor establishes valency as a relationship between unequal elements: one element has "hooks", is "unsaturated" and can bind other (complete) elements, the complements. The relations between the elements are thus part of the unsaturated element. This is of course exactly how (mathematical) functions work but the parallel between the valency approach and the predicate calculus is of limited use, as will be shown below.

Tesnière introduces valency long after the dichotomy "actant - circonstant" which he bases on the idea of a simple sentence as a slice of dramatic action ("un petit drame"). The verb "expresses" a process, the "actants" actors and the "circonstants" circumstances. The key sentence for my analysis is:

Les actants sont les êtres ou les choses qui, à un titre quelquonque et de quelque façon que soit, même au titre de simples figurants et de la façon la plus passive, participent au procès. (Tesnière 1966:102)

Thus Tesnière introduces "actants" semantically through the identification of participants in the "process" (event/scenario).

Tesnière’s remarks on the relationship of the planes of form and content ("plan structurale" vs. "plan sémantique" (Tesnière 1966:40-8)) are more complex and not entirely consistent, but in view of the semantic basis of Tesnière’s distinction, it is remarkable that the more influential followers of Tesnière in Germany (U. Engel, G. Helbig, H.J. Heringer) initially opted for a syntactic concept of valency.

This did not mean rejecting the idea that valency was grounded in the meaning of the verb:

Natürlich war die Orientierung beim Versuch der syntaktischen Kategorisierung nicht oberflächlich syntaktisch. Man wollte ja doch die semantischen Phänomene am Ausdruck festmachen, an dem sie tatsächlich allein festgemacht sind. (Heringer 1983:35-6)

The hope was to define valency adequately through its expression in syntax: the reasons for a syntactic concept of valency were methodological and in tune with classical structuralist thinking and the then contemporary approach in generative grammar. The search was thus on for a criterion that would separate complements from adjuncts in given sentences, and thus identify the complement classes with which a lexeme combines. Even as late as 1988, Welke wrote that the need to distinguish complements from adjuncts was the most discussed issue in valency research (1988:21). Despite great efforts no satisfactory criterion was found.

It is important for my argument to realize that valency grammarians were not content simply to construct different concepts of valency and apply them consistently. There was by and large agreement about prototypical instances of valency and non-valency against which criteria could be assessed: Heringer finds it astonishing that, during the historical discussion on the difference between complements and adjuncts, researchers always presumed the difference that they wanted to establish in the course of their work. He concludes that a notion of the difference was available to researchers but not a definition (Heringer 1984:35).

Also, criteria were not always strictly adhered to in applied work: Eisenberg remarks that the entries for individual verbs do not differ between the Leipzig (Helbig & Schenkel) and the Mannheim (Engel/Schumacher) valency dictionaries to the extent that could be expected from the different valency concepts. In their applied work neither of the two teams of authors had exactly adhered to their respective theoretical claims. He comes to the same conclusion as Heringer: in most instances, researchers would know quite well which prepostional phrases were valency bound and which were not (Eisenberg 1994:298).

For example, the adverbial phrases of verbs like wohnen, stehen, liegen, and sitzen were considered as complements in the "Kleines Valenzlexikon" of 1976 although Engel’s subclassification criterion (see 3.7) does not cover those. This is justified by their obligatory status or a high level of speaker expectation of their occurrence. Engel & Schumacher emphasize that there cannot be a "proof" of the valency status in such cases, nor could frequency counts decide the classification (1976:69f).

Engel later gave greater weight to obligatoriness as a second criterion, allowing only obligatory adverbial phrases as complements. However, both bleiben and stattfinden are listed with an adverbial complement in his Deutsche Grammatik (1988:215):

(1) Sie blieb (in der Stadt). in der Stadt is contextually optional

(2) Die Veranstaltung findet (im Hubertussaal) statt. im Hubertussaal is truly optional

This approach is sensible. If adverbial phrases can be complements at all, why should there not be optional complements among them? It is the criteria that are at fault.

Though good valency criteria initially had to be syntactic (syntactic valency), semantic restrictions of lexemes on their surroundings were soon integrated into the model (both in terms of semantic features and semantic/thematic roles = semantic valency). Pragmatic conditions on changes in valency (pragmatic valency) and differences between logical representations (logical valency) and the valency structure of actual lexemes were also observed. Thus valency took on the appearance of a layered cake with syntactic valency as the base. Still, the different "valencies" were considered as aspects of a single valency phenomenon, rather than as different phenomena altogether, and, despite some doubts, the hope persisted that a criterion could be found that would correctly distinguish complements from adjuncts — not by accident, but because it reflected the essence of valency.

2 Jacobs’ attack on valency

These assumptions were challenged by Jacobs whose typescript "Kontra Valenz" had been in circulation since 1986. His criticism centred on the lack of a consistent definition

Seit über 20 Jahren versucht man, den Valenzbegriff zu präzisieren, ohne daß sich einer der vielen Vorschläge durchgesetzt hätte. [...] Eine gemeinsame Grundidee dieser Typen von Präzisierungen ist nicht zu erkennen [...] (Jacobs 1994:5)

Die traditionellen inhaltlichen Charakterisierungen der Valenz [...] beschreiben zwar sicher plastisch und in sehr anregender Weise eine Intuition, die man bezüglich dieser Beispiele hat, aber sie können ebenso sicher nicht direkt als Grundlage einer linguistisch brauchbaren Definition dienen. Sie sind ja reine Metaphern, die das zu erfassende sprachliche Phänomen durch Vergleich mit nicht-sprachlichen Phänomenen dingfest zu machen versuchen. (Andererseits ist es natürlich gerade das metaphorische Schillern solcher Bestimmungen von "Valenz", das eine Präzisierung dieses Begriffs als unerläßlich erscheinen läßt.) (Jacobs 1994:9)

By emphasizing that none of the attempts to arrive at a precise definition had succeeded (i.e. none were accepted by the majority of valency grammarians), Jacobs accuses the valency research community of having got into a philosophical tangle: the term valency and the metaphorical images that are evoked by it made valency grammarians look for an object, a unified phenomenon, where there was none. Jacobs’ position can be summarised as follows:

  1. There is no widely accepted precise definition of valency.
  2. Pretheoretical intuition about valency is non-existent or limited (i.e. restricted to prototypical cases).
  3. Each relation between valency carrier and dependent elements that differentiates the prototypical valency cases from prototypical non-valency cases is a correct definition of valency.
  4. At least seven selectional relations achieve the differentiation described in 3.
  5. The seven relations are independent of one another: for each pair of them at least one example can be found where they contradict each other (i.e. differ in the classification of the example).
  6. Empirical, possibly universal, implications exist between the relations.
  7. Valency grammarians erroneously blamed correct definitions (in the sense of 3) when their application contradicted the application of other correct definitions.
  8. Valency is not a unified phenomenon. The term should only be used as a cover term for the seven independent relations.

In subsequent publications, Jacobs reduced the selectional relations but, apart from the irrelevant selectional relation "exocentricity", I will list all of them:

  1. Notwendigkeit/Realisierung (Real-Sel)
  2. Beteiligtheit (Rel-Sel)
  3. Argumenthaftigkeit (Arg-Sel)
  4. Formale Spezifität (Form-Sel)
  5. Inhaltliche/Sortale Spezifität (Sort-Sel)
  6. Assoziiertheit/Informationelle Relevanz (Info-Sel)

(As the abbreviation "Bet-Sel" (= Beteiligtheit) will be used more congruously for a different relation later on I have replaced Jacobs’ abbreviation by "Rel" (= Relationalität).)

These "partner binding relations" ("Begleiterbindungsbeziehungen") were arrived at by comparing a prototypical complement with a prototypical adjunct:

(3) (weil) Peter Herrn Meier rasierte

(4) (weil) Peter auf der Wiese schläft

Real-Sel: Herrn Meier is obligatory (apart from elliptic use and particular environments: Peter kann rasieren.), auf der Wiese is optional.

Rel-Sel is based on Tesnière’s distinction between "actants" and "circonstants" which is seen as captured in a division of Fillmore’s original semantic roles (1968:24f.) into +Rel (Agent, Patient, ...) and -Rel roles (Location, Time, ...). The role of instrument remains undecided. Herrn Meier is +Rel, auf der Wiese is -Rel.

Arg-Sel draws on a semantic representation which correctly reflects the intuitive propositional content of the sentence in question and which does not ignore natural language correlates of argumentivity ("natürlichsprachliche Korrelate der Argumenthaftigkeit"; Jacobs 1994:17). As semantic representations using the predicate calculus can have a variety of forms, the decision will in the most crucial cases depend on these "correlates". The best candidate for this turns out to be the "tun/geschehen test": if a phrase cannot be taken out of a sentence and paraphrased in a co-ordinated commenting sentence it is an argument.

(3a) *Peter rasiert, und das geschieht Herrn Meier.

(4a) Peter schläft, und das geschieht auf der Wiese.

Herrn Meier is an argument, auf der Wiese is not.

Form-Sel applies to Herrn Meier in (3) as the accusative is selected by rasieren while no form feature of auf der Wiese is selected by schlafen in (4).

Sort-Sel applies as rasieren assigns the role of patient to Herrn Meier and for literal use requires fillers for that slot either to have the feature [+anim] if not [+higher animal], or to be perceived as such. The role location of auf der Wiese is not dependent on schlafen, as substitution of schlafen by other verbs shows. Also, no particular semantic restrictions are imposed on the slot filled by auf der Wiese. Thus Sort-Sel does not apply.

Both Form-Sel and Sort-Sel have a derived relation which I include in the relation proper (Jacobs later dropped the differentiation (1994:71f) - see §3.4.0).

Info-Sel is based on Heringer’s association experiments (1984:45ff.; 1986:83ff.). Informants were confronted with verbs in the infinitive and asked to formulate questions about the respective actions, processes or states. A quantitative analysis was carried out and candidates for complements were ranked: all the traditional complements outranked adjuncts apart from the genitive complement. By analogy, the question wen? for rasieren would outrank the question wo? for schlafen and is thus +Info, while the latter is -Info.

The six relations are of course as problematic as they were when used in valency definitions. The following points should be considered:

- Different degrees of obligatoriness can be differentiated (as was indicated above).
- It is by no means always obvious whether the form features of an element are determined by another element and, if so, which element determines them (cf. Eisenberg’s attempt to show that the cases in locative complements (Sie wohnt unter dem Dach. vs. Sie zieht unter das Dach.) are verb-selected (1994:295-6; cf. also Zifonun’s justified reservations about this attempt (1995:179-80)).
- Sort-Sel relies on semantic roles which pose well-known difficulties.
- Info-Sel turns graded empirical findings into a two-valued relation. The assignment of values could theoretically change with new test results. For verbs that have not been tested, an extrapolation from existing results is required (presumably using intuitive judgements).

Jacobs suggests that a number of empirical implications between the relations apply (1992:96):

1. +Form / +Real / +Info > +Sort > +Arg
2. +Real > +Info

(If a clause element is +Form or +Real or +Info, then it is also +Sort. If a clause element is +Sort, then it is also +Arg, if it is +Real, then it is +Info.)

Jacobs expects the implications of 1 to be relative universals (i.e. phenomena with a high statistical probability) (1994:64). He also speculates that these implications reflect stages of the grammaticalisation of arguments: historically, argumentivity would have been a precondition for semantic restrictions to apply, which in turn would have been a precondition for formal determination (1994:66f, 71).

With the stipulation of the above implications, Jacobs challenged valency theory to look beyond classificatory questions. This challenge was taken up as will be shown later.

An assessment of the validity of Jacobs’ implications presupposes unambiguous applicability of the individual relations (which is wanting for some). I will not discuss the implications as such but pursue the more modest aim of comparing the individual relations with common valency classifications.

3 Which relations are valency relations?

Jacobs considered all six "partner binding relations" as equally valid attempts at arriving at a precise valency definition. In this section, I discuss how these relations perform when further valency examples are considered. I use Jacobs’ methodology and show that none of his relations, with the possible exception of Info-Sel, separate fairly prototypical complements from prototypical adjuncts. The valency examples below are to varying degree less prototypical than Herrn Meier in (weil) Peter Herrn Meier rasierte (cf. sentence (3) in section 2.) but all are of types classified as complements in applied work such as Helbig & Schenkel (1969), Engel & Schumacher (1976), Schumacher (1986), and also in Heringer (1996).

I also widen the investigation to include two relations that also meet Jacobs’ criteria (specificity and synsemanticity).

3.1 Real-Sel

(5) Sie isst eine Torte. -Real

(6) Sie isst.

eine Torte in (5) is truly optional. This means that Real-Sel cannot separate complements from adjuncts in accordance with applied valency work.

3.2 Rel-Sel

Rel-Sel ("Beteiligtheit") is based on a controversial distinction in Fillmore’s original set of roles. Some of Fillmore’s roles are relational (+Rel), some are not (-Rel): while it only makes sense to talk about an agent or patient, for example, in relation to a particular scenario, places and times are independent of particular scenarios, and can therefore be classed as non-relational. This distinction might well be defensible as it largely coincides with that between synsemantic and autosemantic phrases. However, it is, from the point of view of valency, the wrong kind of distinction because it ignores one of the basic tenets of valency theory (post Tesnière): each clausal element should be assessed for its particular relation to the verb and not be sorted on categorial grounds. Thus clause elements denoting places can have different relations:

(7) Sie erholte sich in Venedig. Locative (-Rel) (supplement)

(8) Sie wohnte in Venedig. Locative (-Rel (complement)

(9) Venedig hat sie krank gemacht. Agent/Locative? ? (complement)

But Rel-Sel does not only contradict common valency analyses, it is also a misnomer: only in (7) is the place phrase "unbeteiligt" (uninvolved) in an informal sense. In (8) and (9) it is clearly involved in constituting the scenario. This is not acknowledged by Rel-Sel for (8) while it is not clear whether the role of agent or locative should be attributed to Venedig in (9): relational and non-relational roles overlap here due to their inconsistent conception. While the non-relational roles ignore any perspective a verb might shed on a scene, the +Rel roles incorporate such a perspective to a degree: they are not just abstractions of relational features such as "acting vs. non-acting" — this would give two "actors" for verbs such as kaufen, verkaufen — but also reflect the syntactic function of the phrase that refers to an "actor". However, in encoding a verb perspective they only go half-way. Subjects and passive von-phrases refer to agents alike while undoubtedly they encode a different perspective on a scene. Whatever use Rel-Sel might have, it is, despite its name, notionally and extensionally not compatible with the valency idea.

3.3 Arg-Sel

Arg-Sel is beset with certain very specific difficulties. These derive from the fact that the argument relation is not a grammatical but a logical relation. Formal logic and grammar, however, have different aims in principle. Formal logic aims at reproducing intuitively valid inferences through a strictly rule-governed procedure. Any formulization that can be used for this purpose is a good formulization. Thus Sie legt das Heft auf den Tisch. could be represented as

(10) legt (sie, das Heft, auf den Tisch) or auf(legt(sie, das Heft), den Tisch)

The fact that Sie legt das Heft is not a grammatical sentence need not worry the logician. There is also no logical reason why most verbs should not be represented by predicates that feature argument slots for place and time, in order to be able to reproduce intuitive inferences such as Wenn Peter schläft, dann schläft er irgendwann irgendwo.

For the appeal to semantic representation to be decisive, one particular representation must be introduced. Arg-Sel would then be as consistent as are the reasons used for determining arguments vs. higher predicates in that representation.


3.3.1 The tun/geschehen test

Jacobs does not introduce a semantic representation but relies on the tun/geschehen test to limit the logical representations that are allowed. It is therefore in effect this test that decides Arg-Sel, and I will discuss it in some detail. I apologise to the reader for the artificiality of the test-sentences that are to follow (which by itself limits the applicability of such tests). The discussion might nevertheless be of interest as the tun/geschehen test was originally used as a valency test: the geschehen test featured in Helbig & Schenkel (1982:38-46). It was chosen by Eroms as the valency test (1981:44-50). An equivalent do so test (see below) was suggested for English by Somers (1987:17) and was also used by Radford (1988:234).

My discussion draws on Storrer’s assessment of the test. She mainly discusses the test as a valency test (Storrer 1992:63, 77, 85-7, 216-25) but most of the criticisms also apply to it as a test for argumentivity. My findings criticise the test more conclusively as a valency test.

The tun/geschehen test, as introduced by Jacobs (1994:17f.), involves the following steps:

  1. Single out a phrase to be tested.
  2. Delete this phrase.
  3. Use geschehen or tun to form a second sentence that supplies the missing information in the form of a comment on the reduced original sentence in such a way that
    a) the deleted phrase occurs verbatim in this sentence
    b) the subject of geschehen or the object of tun is a proform referring to the reduced original sentence.
  4. Add the second sentence to the reduced original sentence. The resulting sentence must have the same descriptive meaning as the original sentence.
  5. If the resulting sentence is ungrammatical, then the phrase in question is +Arg (= refers to an argument).
  6. If the resulting sentence is grammatical, the Arg-Sel status of the phrase in question cannot be decided by the test.

For example:

(11) Peter besteht auf Etikette.
(11a) *Peter besteht, und das geschieht/tut er auf Etikette. ==> auf Etikette is an argument.

The proviso that the tested phrase has to occur verbatim in the second, commenting sentence prevents the test from being inconclusive in many cases, as any entity can, with a little imagination, be perceived as a "victim" of an event (Markus bediente den neuen Gast. ==> Markus bediente, und das geschah dem neuen Gast.).

The tun/geschehen-test has serious flaws:

  1. the arrangements seem somewhat arbitrary;
  2. the outcome of the test is often not clear-cut;
  3. the verbs geschehen and tun do not function in the same way;
  4. the rationale for the test is questionable.

As far as the arbirariness of the arrangements is concerned, the question is a simple one. Why limit the test to certain verbs, and insist on anaphoric reference to the original sentence and co-ordination of the two sentences? (All these restrictive devices make more phrases fail the test.)

Secondly, the outcome of the test is often not clear-cut as the grammaticality of the resulting sentence can be a matter of opinion. For example, if the test on

(12) (weil) Peter Herrn Meier rasierte

was carried out using tun rather than geschehen (cf. (3a) above) the result could be inconclusive for some speakers:

(12a) ?Peter rasiert, und das tut er Herrn Meier.

Thirdly, and this is a substantial point, the verbs geschehen and tun do not function in the same way and have to be considered separately. Geschehen establishes a relationship between a scenario and reality as a whole (Das geschah.) or an entity (Das geschah ihr.) or a circumstantial aspect (Das geschah hier.). Therefore it does not function as a proverb in the sense that it could replace another verb. As Storrer has pointed out, it has its own valency (1992:80-2, 220f.):

- the subject only allows reference to a scenario (in the test: to the whole scenario denoted by the previous utterance);
- no accusative object or directional complement is possible;
- the dative/prepositional object can only refer to the patient (in relation to the whole scenario).
As the geschehen test requires verbatim repetition of the tested phrase in the commenting clause, all the clausal elements prohibited by the valency of geschehen cannot be tested:

(13) Petra aß eine Torte. ==> *Petra aß, und das geschah eine Torte.

(14) Jürgen machte seiner Freundin eine Torte. ==> ?Jürgen machte eine Torte, und das geschah seiner Freundin.

(13) does not prove anything about the argumentivity of eine Torte, just as (3a) does not prove anything about the argumentivity of Herrn Meier. Both test sentences are ungrammatical because of the syntactic valency of geschehen alone.

Regardless of how the grammaticality of the test sentence in (14) is viewed, the test again fails to apply. The semantic valency of geschehen has changed the semantic role of seiner Freundin from beneficiary to patient: the test sentence does not have the same meaning as the input-sentence. (The labelling of the role is debatable. What is important is to realize that the relationship between the dative and the two verbs is not identical.)

A defender of the test might claim that the valency of geschehen simply demonstrates why it was chosen for the test in the first place: the arguments, whether realised or not, (together with the tested verb) are "thematised" by the subject of geschehen and therefore cannot occur as rhematic elements. This, however, makes the test procedure quite superfluous in most cases: it could be substituted by studying the valency of geschehen. My dative examples below show that neither the test nor the valency of geschehen are reliable indicators of argumentivity.

The failure of geschehen to test accusative objects and beneficiary datives for argumentivity might be seen as negligible as there is little or no doubt about their argument status anyway. This is not so with the so-called free datives, and the geschehen test is also not applicable to the majority of them. The test might work for some instances of the dativus incommodi:

(15) Dem Koch ist das Essen verbrannt. ==> Das Essen ist verbrannt, und das geschah dem Koch.
(16) Dem Gast ist das Essen verbrannt. ==> ?Das Essen ist verbrannt, und das geschah dem Gast.

The question is whether the dative denotes the person in whose sphere of responsibility the misadventure happened in both the input and the test-sentence. This is so in the first pair of sentences (15) and the test does not classify dem Koch as an argument. In the second pair (16), however, while the guest is seen as responsible in the input sentence due to its construction, he is arguably seen as the victim that was not involved in the cooking process in the test sentence. The reason is pragmatic: cooks cook, guests do not. But a change in meaning implies that the test is not applicable.

The test is applicable with verbs that have similar meaning and share the construction with geschehen:

(17) Der Nachbarin ist ein Unglück passiert. ==> Ein Unglück ist passiert, und das geschah der Nachbarin.
(18) Der Nachbarin ist ein Unglück zugestoßen. ==> *Ein Unglück ist zugestoßen, und das geschah der Nachbarin.

The test does not classify the dative phrase of passieren as an argument. (The same applies to geschehen itself.) This, however, is rather surprising, particularly in comparison with the dative of zustoßen, which the test shows to be an argument, consistent with our expectations. The reason for this contradiction is obvious: the test result of passieren is not informed by argumentivity but by the true optionality of the dative phrase. Though this explains the contradictory test results, the reliability of the test is rather undermined. (Arguably, the stronger claim that passieren has two closely related readings (one with and one without dative) could be entertained in defence of the test.)

The test is also applicable to autosemantic prepositional phrases and adverbials, and for these it can be used to establish argumentivity (but, strictly speaking, not non-argumentivity!), though doubts remain about grammaticality judgements on artificial sentences for which it is difficult to imagine any use. Also, factors other than argumentivity such as optionality or the number of phrases in the tested sentence might affect the grammaticality of test sentences (cf. Eroms 1981:45-50).

The use of tun in the test can be seen as linked to tun as a colloquial/regional quasi-modal or auxiliary verb:

(19) Sie tut das Zimmer aufräumen.

The verb tun can have a variety of functions in this capacity: it can function in a similar way to do but more often than not it lacks any emphasising effect. It can also indicate the ongoing and/or concurrent nature of an activity. It might also simply be motivated by the desire to postpone the choice of verb. When the dependent infinitive construction is replaced by a pro-form the use becomes standard:

(19a) Sie tut es/das.

Unlike geschehen, the valency of tun, like that of modal verbs, is restricted to a verbal complement (Verbativergänzung), the other complements being dependent on the second verb:

(20) Helfen tut er einem Mann.
(21) Verdächtigen tut er sie der Unterschlagung.
(22) Mich tut es schwindeln.

This allows the tun test a wider application than the geschehen test. However, tun can hardly be used with verbs featuring subjects that denote an inactive entity:

(23) ?Sie tut einen Brief erhalten.

The usefulness of tun is further restricted by often inconclusive test results:

(24) Er isst einen Apfel. ==> ?Er isst, und das tut er einen Apfel.

An indulgent approach to the grammaticality of the resulting sentences might lead to the conclusion that tun tests for obligatoriness rather than argumentivity.

Finally,the rationale for the geschehen/tun test is that clausal elements which can be separated from their governing verb can be construed as predicates, or at least do not have to be seen as arguments. This rationale is compatible with any means of making a clause element part of a commenting sentence, e.g.:

(25) Er ärgerte sich über die Steuererhöhung. => (?) Er ärgerte sich. Das tat er über die Steuererklärung. Er ärgerte sich. Geärgert hat er sich über die Steuererklärung.

Such a "Satz-Test", however, would clearly test for obligatoriness (Welke 1988:34—6).

To summarise: Jacobs offers no satisfactory operational procedure to secure Arg-Sel. The relation runs into similar problems to that of valency. An appeal to the practice of the majority of semanticists or to a particular semantic model would only help to the extent that they might offer criteria for argumentivity beyond appeal to semantic intuition. In defence of Arg-Sel it could be argued that the idea of what is a "natural" argument (not: what is logically possible) is somewhat clearer than the notion of valency and can be supported by semantic considerations. Indeed, I will show below that Arg-Sel provides a route to the intuition on which valency is, in my opinion, based.

Even with a sympathetic assignment of Arg-Sel values, however, the relation fails to identify predicative complements:

(26) Sie nennen ihn den Aussitzer.

Semantically, den Aussitzer is a predicate taking ihn as an argument to form a proposition which is the second argument of nennen. However, valency grammarians are quite adamant that

  1. den Aussitzer is a complement
  2. nennen is trivalent (giving credit to the surface structure of (26))

Thus Arg-Sel does not separate complements from adjuncts satisfactorily — despite being the most promising relation so far. (Therefore the term argument is often used as a substitute for complement, in effect stripping it of its logical-semantic content, and applying it to clause elements rather than semantic units (cf. Engel 1996:226, footnote 22). This is not the usage intended by Jacobs and investigated here.)

To conclude this section, I will briefly discuss the do so test which fares only slightly better than the geschehen/tun test. The test requires co-ordination of the sentence containing the tested clause element with a do so phrase: complements cannot be repeated, although adjuncts can be. E.g.

(27) John goes to school in Manchester and Jim does so in London. vs.
(28) *John lives in Manchester and John does so in London.

Somers (1987:18) points out that the do so test can neither test for adjunct status (as adjuncts can, but do not have to be, pronominalized), nor for the subject, but otherwise it has a wider application than either of the German tests. However, do so fails to pronominalise complements in certain contexts and sometimes must pronominalise adjuncts:

(29) *Pamela quickly downed a whisky and Paula did so a Martini.
(30) Pamela runs fast and Paula does so slowly.

In both cases, the choice of different contexts would lead to the correct result but if the test results have to be checked against one’s intuitive semantic judgement, one might as well rely on the latter in the first place.

Whether they are used to test for argumentivity or valency, both the geschehen/tun test and the do so test are, in suitable contexts, better used simply to illustrate the difference between arguments or complements on the one hand and non-arguments or adjuncts on the other than as diagnostic tools.

3.4 Form-Sel

The next relation to be discussed, Form-Sel, can be used when form features of a phrase are determined by a particular verb lexeme and thus have to be recorded in the lexicon, rather than being derivable through general rules.

Undoubtedly, Form-Sel is the most important relation in the description of complements. This is so because forms are tangible and because form features are particularly important for foreign language learners. Therefore Form-Sel is central to valency dictionaries. This does not necessarily mean that Form-Sel is suitable for sorting complements from adjuncts, however.

Jacobs’ rather restrictive definition of Form-Sel can be paraphrased as follows:

A constituent X of a sentence is +Form in relation to another constituent Y (in this case: the main verb), if and only if at least one form feature of X is determined by Y (= the fact that a constituent with this form feature can accompany Y in this sentence is a specific property of Y). (Cf. Jacobs 1994:22)

For example, in

(31) Die Universität bedarf zusätzlicher Finanzmittel (GEN).

The genitive of zusätzlicher Finanzmittel is determined by the verb bedürfen. It is a specific property of bedürfen that it can (and indeed must) be accompanied by a phrase in the genitive. The genitive cannot be derived from general syntactic rules (most verbs do not accept a genitive constituent) or from semantic rules (the entity that is needed does not have to be expressed through a genitive phrase), as (32) shows:

(32) Die Universität braucht zusätzliche Finanzmittel (AKK).

(The idiosyncracy of the genitive with bedürfen is underlined by the fact that it might have been replaced by the accusative without historical efforts to preserve it; cf. Engelen 1975:66).

Originally, Jacobs wanted to limit Form-Sel to features that cannot be predicted. He defined the notion specific property of Y as encompassing all properties P of Y (or the lexeme of which Y is a form) that cannot be predicted by grammatical rules from properties of the grammatical environment of Y, or from any different properties of Y (or its corresponding lexeme) that are logically independent of P (Jacobs 1994:23).

For example, he considered the fact that passive forms can feature a prepositional agent phrase as -Form because it can be derived from the active by rules. Similar arguments apply to possessive datives (Sie massiert ihm die Schulter). However, as the rules refer back to another verb form or usage of the same verb, a derived relation applied. In the 1993 epilogue to Kontra Valenz, Jacobs combined the two relations by stipulating that the specific property must only be inherent to Y (Jacobs 1994:71f). Nevertheless Form-Sel does not identify all complements:

(33) She is sleeping.

The form feature of the subject (subjective rather than objective) is structurally assigned. The evaluation of Form-Sel in relation to the German subject is more difficult, as subjectless sentences (e.g. Mich schwindelt., Ihr wurde gedroht.) and sentences with a postponed subject in unmarked order (Der Weser drohen ökologische Schäden.) have to be considered, as well as the possibility of expanding such sentences with the subject element es (Es schwindelt mich., Es wurde ihr gedroht., Es drohen der Weser ökologische Schäden.) Be that as it may, Form-Sel fails to differentiate adverbial complements from adjuncts:

(34) Sie wohnt in der Innenstadt.

Jacobs considers the adverbial complement in (34) as -Form.

3.4.1 Rekt-Sel

Eisenberg suggested a different concept of form selection which is based on government (Rektion):

A constituent f1 governs a constituent f2 if the form of f2 is determined by a "paradigm category" (Paradigmenkategorie) of f1 (a category applying to the f1 lexeme = the paradigm to which f1 belongs) (1994:52).

Eisenberg divides the relation (Rekt-Sel) into categorial government (Rektk-Sel), whereby a whole category/word class determines the form of a constituent, for example German nouns can take a genitive phrase, and lexical government (Rektl-Sel). The "paradigm category" corresponds to Jacobs’ "specific property" but without restrictions. In this respect Rektl-Sel can be expected to be more inclusive than the original Form-Sel. On the other hand, Form-Sel can be based on instances of a lexeme which can make Form-Sel both more and less restrictive than Rektl-Sel.

The German subject is +Rektk but -Rektl-Sel as every German verb can take a subject element. The other case complements are +Rektl but -Rektk. Thus neither subrelation can determine the set of complements. The combined relation Rekt seems to fare better. Eisenberg claims that in der Innenstadt in (34) above was +Rekt as the dative was verb-selected:

(34a) *Sie wohnt in die Innenstadt.

His analysis is controversial but, if it is accepted, adverbial adjuncts would also be +Rekt as Zifonun pointed out (1995:179f):

(35) Sie verletzte sich in der Innenstadt.
(35a) * Sie verletzte sich in die Innenstadt.

The dative in (35) would have to be seen as +Rekt. (A parallel case could be based on directional complements if directional adjuncts are accepted: Sie schimpfte aus dem Fenster. Er rief in den Nebel.)

Rekt-Sel can therefore not separate traditional complements from prototypical adjuncts. Eisenberg’s syntactic notion of valency reaches far into traditional adjuncts, contradicting the practice of applied valency work.

While form relations (under whichever interpretation) are central to grammatical description and arguably more important than valency, they cannot supply a criterion that reflects or explains the practice of valency grammarians.

3.5 Sort-Sel

Sort-Sel is the semantic counterpart to Form-Sel: it specifies the semantic determination of phrases by a verb. Jacobs’ technical introduction for Sort-Sel mirrors that of Form-Sel, replacing form features, the choice of which is specific to a verb, by content features (Inhaltsmerkmale). I will not repeat the details here.

Sort-Sel really consists of two relations: the determination of semantic features (Merkmale; SortM-Sel), and the determination of semantic role (SortR-Sel). SortM-Sel does not apply to all prototypical complements:

(36) Die Sonne/Dein Verhalten/Dass sie immer zu spät kommt, stört ihn.

As anything can be disturbing, stören does not impose any selectional features on the subject.

SortR-Sel is more difficult to assess. Jacobs claims that directional complements are -Sort as every verb describing motion (fahren, gehen, schwimmen) can be accompanied by one. Thus SortR-Sel, under his interpretation, does not reflect the complement-adjunct divide. Jacobs’ assessment is not necessarily correct, however. The key question is: what is meant by the term spezifisch (specific) in his definition of Sort-Sel? While the ascription of a semantic role (goal or direction) is not specific to individual verbs, it is specific to a whole subclass of verbs. I cannot see how this differs from the claim that verbs of action feature an agent.

Whatever position one takes, a parallel argument to the one just made concerning Form-Sel can be made for SortR-Sel: either both adverbial complements and adverbial adjuncts are +SortR or -SortR. The same applies to certain predicative complements:

(37) Es geht ihr gut.

(38) Die Uhr geht gut.

Valency dictionaries list the obligatory adjective in (37) as a complement but not the optional one in (38). Both adjectives, however, have the same semantic relation to the verb (characterisation) and should receive the same SortR-Sel value. SortR-Sel thus cannot be used as a criterion for complementhood.

3.6 Info-Sel

Info-Sel is based on Heringer’s association experiments that measured the likelihood of certain questions being associated with individual verbs. This likelihood was interpreted as semantic proximity of the corresponding clause elements to the verbs.

As an empirical relation, Info-Sel has a different status: the measurements could be a reflection of one or several of the other relations rather than constitute a relation sui generis. If Heringer’s results are taken at face value, Info-Sel does not correspond to common valency analyses as the genitive object of beschuldigen attracts -Info: the informants were hesitant about associating wessen-questions with the verb. The explanation that this is due to the unusual question word rather than a lack of semantic proximity to the verb is convincing: wessen? is suppleted by warum? and womit?, both of which show higher values than with other verbs (Heringer 1985:93-5). (It should, however, be noted that warum? has a wider application than wessen? as it can refer to the reasons for an accusation rather than the presumed offence: Warum beschuldigst du ihn dessen? It is therefore unlikely that weil-clauses will replace the genitive in future as Heringer speculates (1985:94-5). A preposition, for example mit because of womit? seems a better candidate.)

If Heringer’s explanation about the low score of wessen? is accepted, Info-Sel reflects the traditional complement-adjunct divide and can be seen as evidence for its validity. I discuss Info-Sel further in 4.3.

3.7 Spez-Sel

Jacobs does not believe that specificity (Spez-Sel) supplies a valid definition of valency because it fails to separate prototypical instances of valency from prototypical instances of adjuncts.

Specificity (Subklassenspezifik, originating from subcategorisation in generative grammar), the main criterion used by Engel, is solely based on distribution. It says that complements are specific to (semantic) subclasses of verbs (i.e. are only combinable with these subclasses) while adjuncts are aspecific (i.e. can be combined with any verb in principle):

Glieder, die von allen Elementen einer Wortklasse abhängen können, sind Angaben.
Glieder, die nur von bestimmten Elementen einer Wortklasse abhängen (können), sind Ergänzungen. Oder: Ergänzungen sind subklassenspezifische Glieder. (Engel 1994:99)
[...] Solche auf Teile von Wortklassen beschränkte Rektion nennen wir fortan Valenz. Damit ist Valenz nichts als subklassenspezifische Rektion. (Engel 1994:96, cf. 1988:24)

While the criterion is not without problems, critics sometimes fail to appreciate that specificity is based on classes rather than individual phrases (e.g. Eisenberg 1994:297). The claim is not that each individual adjunct is combinable with every verb; rather, we should consider general classes. This proviso means that semantic incompatibilities are not allowed to determine Spez-Sel. For example:

(39) *Die Aufführung beginnt sehr um 7 Uhr.

The sentence is ungrammatical as beginnen is not scaleable. However, other "grading particles" can be combined with beginnen:

(40) Die Aufführung beginnt bereits um 7 Uhr.

(The particle bereits grades the start of this particular performance in relation to others.)

So far there is no reason to assume that the class of particles subclassifies verbs, though individual ones can do so, for example sehr. According to Engel’s constructionist view, the complement-adjunct-divide depends on the choice of adjunct classes: if relatively small ones are chosen, the divide disappears (1994:101).

Subklassenspezifik would likewise classify a case complement as non-specific (and demote it to adjunct) if different instances of it were combinable with every verb without any particular instance being combinable with all verbs: the individual phrase does not matter, apart from being a representative of a class, or forming a class of one member.

Jacobs maintains that the underlined phrases in both his example sentences are specific to verb subclasses:

(41) (weil) Peter Herrn Meier rasierte

(42) (weil) Peter auf der Wiese schläft

I agree with his conclusion concerning the accusative phrase in (41). However, I am less happy with the reasoning that led to it: Jacobs’ replacement of rasiert by hofft can be judged as inconclusive, as Herrn Meier in the interpretation of (41) is strictly speaking not semantically compatible with hoffen. A different verb should have been chosen, e.g. lachen.

Jacobs’ assessment of the place adverbial in (42), however, has to be rejected. In order to show the commutation class of auf der Wiese as +Spez, Jacobs replaces schläft with tot sein (1994:26):

(43) ?? (weil) Peter auf der Wiese tot ist

The replacement is rather infelicitous as many valency grammarians would consider sein rather than tot sein as the verb in (43), avoiding the necessity to list as verbs all adjectives that can be used predicatively (cf. Engel 1994:152f). But if tot sein is accepted as a verb for the sake of the argument, the oddness of (43) does not qualify place adverbials as +Spez, as it is semantically motivated (tot sein is not linked to particular places, therefore mentioning a place violates the conversational maxim "Be relevant"). The oddness disappears in contexts where a place adverbial is either relevant (44) or evocative (45):

(44) (weil) Peter in ihrem Traum tot war

(45) (weil) Barschel bereits in der Badewanne tot war

Spez-Sel constitutes an additional valid relation in the sense of Jacobs. Indeed, he himself takes a more conciliatory approach towards Spez-Sel in his 1993 epilogue to Kontra Valenz but for different reasons than the ones I have discussed: he claims that Spez-Sel is valid if the direction of combinational restrictions is taken into account: only the ones imposed by heads are relevant, the ones imposed on heads by governed elements (adjuncts) have to be discarded (1994:72).

Spez-Sel not only differentiates Jacobs’ prototypical complement from his prototypical adjunct, it also provides an appealing image. If complements are verb-specific and adjuncts aspecific, this creates a distributional reflex of Tesnière’s tenet that complements denote participants and adjuncts circumstances, given that scenarios and hence their participants vary, while circumstances are always of the same kind. Furthermore Spez-Sel provides an elegant and precise way to incorporate valency into a dependency grammar: complements are "satellites" that are governed by a subclass of the governing category, adjuncts can be governed by all elements of the governing category (cf. Engel 1988:23, 1994:95f).

However, Spez-Sel fails on several counts. It cannot be used as a discovery procedure to sort valency from non-valency cases and there are also conceptual problems. I investigate first the two areas which the valency approach treats in a different way from traditional grammar, namely subjects (1) and adverbials (2), and then try to identify more general problems of Spez-Sel (3 and 4):

1. Spez-Sel is based on distributional data. This is its strength and its weakness. If subject elements such as it, there (It is raining, There was hope) are not assigned complement status as they do not commute, the English subject is +Spez. Imagine now a language that is identical to English apart from featuring a full (commutable) subject in every sentence. For example, instead of meteorological verbs there would be nouns: Rain is occurring, etc. Subjects would thus be -Spez and would have to be classed as adjuncts if Spez-Sel was the only criterion. Spez-Sel thus does not capture the essence of Tesnière’s valency metaphor.

2. Apart from treating subjects as complements, there is a broad consensus, in particular in applied work, that adverbials with certain verbs qualify as complements:

(46) Sie wohnt auf dem Dach.

(47) Sie geht auf das Dach. (directional complement)

Spez-Sel fails to identify a whole subclass of adverbial complements: situational complements. Situational adverbial phrases (auf dem Dach, in der Stadt, dann) can be combined with every verb in principle. Engel has frequently drawn attention to this (e.g. 1994:99, Engel & Schumacher 1976:96). In earlier work, he mentioned speaker expectation of clausal elements, i.e. Info-Sel, as a criterion (1970:372, Engel & Schumacher 1976:69; cf. Varnhorn 1986:3). Emphasizing arbitrariness in the definition of grammatical categories (1994:101), he does not seem to really consider specificity as a discovery procedure. The value of the criterion for him is rather that it is not restricted to verbs but applicable to all phrase building word classes and fits neatly into his structuralist dependency grammar.

Though situational adverbial phrases are not specific to subclasses of verbs, it should be noted that situational complements have different distributional characteristics to situational adjuncts: they are obligatory:

(48) *Sie wohnt.

or, if optional situational complements are allowed (there is no good intuitive reason why they should not be), there is a high chance that they will accompany the verb in question:

(49) Das Konzert findet (auf dem Dach/um 20 Uhr) statt.

But these distributional facts have to be dealt with by a second relation (Real-Sel or a statistical version of it), they cannot be used by Spez-Sel as a discovery procedure. Once situational complements are identified, one can obviously say that they are specific to semantic verb classes, and in this sense Engel’s definitions quoted at the beginning of this section (3.7) are vindicated.

3. The criticism of 2 shows a conceptual flaw of Spez-Sel as a discovery procedure. Complements are relational, not categorial units. They are complements of a verb, not complements per se. However, in order to be able to function, Spez-Sel has to be blind to the relational nature of complements: different commutation classes are equated on the basis of their categorial features (their form, as shown in their anaphora):

A. Er singt einen Schlager
eine Arie
ein Lied
was ihm in den Sinn kommt
B. Er isst einen Wecken
eine Stulle
ein Brot
was ihm in den Sinn kommt
C. Er träumt den ganzen Tag
eine Weile
ein Weilchen
während der Vorlesung
sobald er die Augen schließt
D. Es dauert den ganzen Tag
eine Weile
ein Weilchen

es, solange
E. Sie wohnt auf dem Dach
unter der Brücke
in Düsseldorf
wo der Pfeffer wächst
F. Sie friert auf dem Dach
unter der Brücke
in Düsseldorf
wo immer sie hingeht

A and B feature communication classes of the same kind because of the shared anaphora, while C is excluded as it has a different set of anaphora. D shows a partial overlap of anaphora with C, demonstrating the practical difficulties of arriving at Spez-Sel values. E and F again share the same anaphora and thus have communication classes of the same kind.

Valency status is now assigned irrespective of the environment of a particular communication class. It is assigned on categorial grounds: types of communication classes that happen to be restricted to a subclass of verbs, such as types A and B, are +Spez while types of communication classes that are combinable with all verbs, such as types of E and F, are —Spez, ignoring the particular relation to the verb of the communication class in E. This contravenes the relational nature of valency (cf. Varnhorn 1986:5f.). Spez-Sel only arrives at the correct results for most complements because no adjunct versions of the respective categorial types of communication classes exist. It fails where one does, as was shown above.

4. Distributional analyses are difficult to handle. Spez-Sel relies on a number of judgements and assumptions: