ChapterPDF Available

Situation types, valency frames and operations


Abstract and Figures

In a functional perspective, situation types are defined on a cognitive basis. These are the bases of argument frames of predicates, and these are the bases of valency frames of verbal lexemes/stems. Argument frames are converted into valency frames by valency operations, which change the meaning of the lexeme/stem. A stem is then subject to diathetic operations, which respond to communicative needs and do not change the meaning beyond compositional grammatical adjustments.
Content may be subject to copyright.
Christiani Lehmanni inedita, publicanda, publicata
Situation types, valency frames and operations
huius textus situs retis mundialis
dies manuscripti postremum modificati
occasio orationis habitae
Conference on Valency Classes in the World's Languages.
Max Planck Institute for Evolutionary Anthropology, Leipzig,
April 14-17th, 2011
volumen publicationem continens
Comrie, Bernard & Malchukov, Andrej (eds.), Valency classes. A
comparative handbook. Berlin & New York: Mouton de Gruyter
annus publicationis
Situation types, valency frames and operations
Christian Lehmann
Universität Erfurt
1 Introduction
1.1 Semiotic constraints vs. cognitive and communicative functions
The language system is a semiotic system. As such, it is the result of the interplay of two essentially
independent forces:
Structure: formal constraints: The constraints on a semiotic system and on the messages
constructed from it are of a different nature. On the one hand, laws of logic, information theory and
physics determine the ways in which signs may be selected, combined and transmitted. These are
complemented by other laws of nature in the case of semiotic systems used by a particular species,
e.g. homo sapiens.
Functions: communication and cognition: The world surrounding us which we conceptualize
is in many respects the same for every speech community; and the same holds for the tasks of
communication in such a community. These two domains provide the total of content and its
conveyance in the widest sense.
Thus, entities of grammar, including valency classes, have a purely formal side determined by
the constraints imposed on any semiotic system. At the same time, this formal side is not empty, but
is laden with cognitive and communicative content. In more concrete terms: grammatical
categories, relations, constructions and operations are necessary for a semiotic system of some
complexity to operate, and they do have some purely formal properties. At the same time, these are
categories like tense, relations like the indirect object relation, constructions like the causative
construction and operations like causativization; and none of these is purely formal, all of them
have their semantic side. Putting it yet another way: in a semiotic system, everything concerning the
sign as a whole is significative (meaning-bearing).
Applied to valency classes, this conception implies:
a. On the one hand, verbs form valency classes because these are the systematic aspect of the
combinatory potential of verbs. More specifically, valency classes are the logical condition for
the semantic compositionality of verbal clauses; and semantic compositionality is a
precondition for an analytic approach to linguistic messages.
b. On the other hand, verbs form valency classes because the situations that human beings
conceptualize have an inherent structure that they react to in their categorization.
The association of form and function in language is not biunique. A classification of semiotic
entities, including grammatical ones, by semantic criteria yields different results from a
classification based on formal criteria. This is true for valency classes just as for any other
Christian Lehmann, Situation types, valency frames and operations 2
grammatical category.
The double-sidedness of valency classes has many methodological
consequences. One is of immediate relevance here: Any analysis of valency classes aiming at
understanding their nature has to take a double approach to them, a formal (alias semasiological)
and a functional (alias onomasiological) approach. In this article, only the functional approach will
be taken. This implies that the approach does not do justice to the functional profile and polysemy
of the valency patterns and operations of the individual language. Instead, it provides a conceptual
framework that an onomasiological description may be based on and that a semasiological
description may refer to.
1.2 Levels of analyzing argument structure
A typology of valency confronts its object at three semantic levels, which are represented in Table 1
(cf. Lehmann 2006, §2):
Table 1 Levels of representation of valency frames
# domain range semantic level components roles examples
3 communication
and cognition extra-
linguistic sense
construction situation: situation
core, participant … participant role moved entity,
instrument …
2 linguistic
typology cross-
linguistic designatum proposition: predicate,
argument, satellite,
relator …
instrumental …
1 language
system language-
specific significatum clause: verb, actant,
adjunct, case … syntactic function
+ significatum of
case relator
direct object, with-
phrase …
Level 2 is an abstraction from level 1, generalizing over the latter’s variation. Level 3 comprises
what is conveyed in a speech act. Although this happens by means of units of level 1, it is partly
extralinguistic, since sense construction involves not only the significata and semantic rules of the
language system, but crucially also inferencing on the basis of an appraisal of the speech situation
and activation of experience and world knowledge. The typology of valency uses concepts of level
2. However, the other two levels are implicated, too. Generalizations at level 2 are operationalized
and, thus, falsified at level 1. And on the other hand, linguistic types differ by the strategies they
employ at level 2 in order to code the sense conveyed at level 3. In this way, level 3 serves as the
tertium comparationis in typological comparison.
There is much terminological variation in the domain here under study, part of which stems
from the fact that the levels of Table 1 are not always distinguished. As the table suggests,
distinguishing the levels entails the use of different terms for the entities of the last three columns
depending on the level being referred to:
Previous research has emphasized either the correlation between form and function in valency (Levin 1993)
or its divergence (Faulhaber 2011).
Entities belonging to level 3 are sometimes considered as “phenomena in the world” (Van Valin & LaPolla
1997:83). However, phenomena in the (physical, “real”) world are of no relevance to linguistic analysis.—
Apart from that, the approach of Van Valin & LaPolla 1997, esp. ch. 3, is an important model for the
approach taken here.
Christian Lehmann, Situation types, valency frames and operations 3
The most generic hyperonym for events, actions, processes, states-of-affairs etc. at level 3 is
situation. Situation cores are relational concepts. At the level of cross-linguistic semantics (#2
of Table 1), a situation core may be represented in the form of an open proposition, i.e. a
combination of a predicate with unbound argument variables. At level 1, it is typically coded by
a verb.
The entities surrounding a situation core are participants. Languages distinguish central
participants from peripheral ones. At level 2, the former are called arguments. An argument is
what a predicate (representing a concept) opens a position for (as in Van Valin & LaPolla
1997:90). It is, thus, not a valency-dependent clause component, which latter is, instead, an
(or complement).
Peripheral participants may be called satellites at level 2; they are
typically coded as adjuncts at level 1.
A semantic role (variously thematic role, as in Van Valin & LaPolla 1997, or theta role) is a
cross-linguistic concept coded in the structure of some languages, but possibly not of others. It
is to be distinguished from a participant role, which is situated at level 3 of Table 1, grounded in
functions of communication and cognition and, therefore, partly independent of linguistic
1.3 The status of semantic roles
The identity of a concept includes its argument structure, i.e. its argument places with their semantic
roles. Therefore, P(x) and P(x,y) are not the same concept. The concept of ‘break’ is the same in #a
and #b of E1, but different in #c.
E1 a. Linda broke the twig.
b. The twig was broken by Linda.
c. The twig broke.
Semantic role operations operate at the level of the predicate, changing its argument structure. This
shapes the meaning of a sentence. For instance, a valency-changing derivation such as the
deagentive (e.g. break (tr.) becomes break (itr.), as in E1.a vs. c) is described like this: The semantic
macro-role of the actor is blocked. Consequently, there remains, at the level of semantic roles, a
single argument, viz. the undergoer. The hearer uses this semantic information, as well as inferences
on the basis of the speech situation and world knowledge, to construct the sense of the utterance. At
this level, a semantic role operation may have different effects. In E1.c, the hearer is not asked to
believe that the twig broke without the intervention of an acting force. Instead, there is just no
particular acting force implied.
Apart from being the traditional term for the concept in question, actant has also been used in typology, e.g.
in Lazard 1998.
What Rappaport et al. 1993 and their followers dub ‘predicate-argument structure’ is actually the
(syntactic) valency frame of a verb. And this is not merely a terminological issue; as argued above, the
concepts of argument structure and valency need to be distinguished.
Again, a peripheral argument is still an argument. Since the centrality of arguments is roughly determined
by their sequential order following the predicate, a peripheral argument is essentially one at a position > 2,
i.e. one that is neither actor nor undergoer.
Christian Lehmann, Situation types, valency frames and operations 4
E2 a. k-u haan-t-ik
he eats it’
b. k-u haan-al
he eats’
In E2.a, haant is transitive, reflecting an argument structure with an actor and an undergoer. E2.b is
intransitive, reflecting an argument structure with an actor, but no undergoer. At the level of the
designatum (#2 of Table 1), the actor is busy eating; no eaten object is being represented. At the
level of sense construction, the difference between #a and #b is another one. In both cases, there is
an eaten object, since eating is inconceivable (in a sense, impossible) without an eaten object. In
other words, excising the eaten object from the concept of eating would result in a totally different
concept, maybe exercising one’s ingestive organs. Again, the hearer receiving E2.b does not
conclude that the actor eats nothing. Instead, he concludes that the actor eats something which is not
represented in what is conveyed to him, but which he might try to infer from other evidence, for
instance on the basis of world knowledge or by just looking.
Similarly, the actor coded in E3.a is absent in #b.
E3 a. t-in ch'am-ah u chuun le che'-o'
YM PRFV-SBJ.1.SG bruise-CMPL POSS.3 base DEF tree-D2
‘I bruised the trunk of the tree’ (EMB&RMC_0033)
b. h ch’áam u chuun le che’-o’
PRFV bruise\DEAG POSS.3 base DEF tree-D2
‘the trunk of the tree got bruised’
Nonetheless, this is so only at the level of semantic structure (#1 and 2 of Table 1). At the level of
sense construction, the addressee of the utterance conveying E3.b is not asked to believe that trees
can get bruised without the intervention of an actor. Quite on the contrary, a complex sentence such
as E4 is fully consistent, although the first clause codes an actor, while the second clause does not.
E4 t-in koh<ah> in coche ka h ch'áam-ih
‘I hit my car so that it got bruised’ (EMB&RMC_0032)
Finally, the same point can be made in a semasiological perspective.
E5 a. Linda peeled the orange with her pocket knife.
b. Linda filled the bucket with beer.
Both #a and #b of E5 feature the semantic role of the instrumental, which in this language is coded
by a prepositional phrase introduced by one of a small set of prepositions like with. However, only
the situation coded by #a involves a participant with the participant role of instrument, while what is
coded as an instrumental in #b is rather a moved object at the level of sense construction (s. §3.3.5).
Thus, semantic roles are schematic; they do not provide direct access to the sense, but are rather a
generic means of structuring a situation in terms of a limited number of concepts and relations.
A situation type is an abstraction over a set of particular situations. This concept is therefore
situated at level #3 of Table 1. Level #2 provides strategies used by languages to convert situation
types into each other and to code a situation type by a type of construction.
Christian Lehmann, Situation types, valency frames and operations 5
2 Situations and participants
Given the onomasiological approach of this article, we will start by characterizing participants and
situations at the language-independent level and then gradually pass on to concepts that have some
linguistic specificity. Situation types are conceived at level 3 of Table 1. They are converted into
types of predicates with their argument frames at level 2. These represent linguistic
conceptualizations of situations, and mostly there are variant conceptualizations of a given type of
situation. Each of the variants may be useful under different conditions having to do with the
particular speech situation. There are therefore, still at the typological level, paradigmatic
relationships between predicate-argument constructions which may manifest themselves in
individual languages in the form of coded or uncoded alternations among such constructions. We
will first consider the problem of representing participants of a situation as arguments of a
2.1 Mapping participants onto arguments
Consider E6 as a simple example to show that a predication represents a selection among the
participants involved in a situation:
E6 a. Erna glaubte mir.
‘Linda believed me.’
b. Erna glaubte diese Geschichte.
‘Linda believed that story.’
c. Erna glaubte mir diese Geschichte.
‘*Linda believed me that story.’
There is a situation type which may be represented as B
(x, y, z), where x is the believer, y
the person believed and z the abstract object believed. In English, one selects either y or z for
linguistic representation (E6.a, b), while in German one may represent all of them in one clause
More generally, there is no biunique mapping between the arguments of a predicate in a
semantic representation and the actants of a verb.
Instead, there are mismatches in both directions:
1. a subset of the actants corresponds to an argument (other actants are semantically empty)
2. a subset of the participants is mapped onto arguments and, thus, actants (the others are
optionally coded by adjuncts or not coded at all).
Van Valin & LaPolla (1997:173) postulate the following “Syntactic template selection principle: The
number of syntactic slots for arguments and argument-adjuncts within the core is equal to the number of
distinct specified argument positions in the semantic representation of the core.” Translating into the
terminology used here: the number of actants (at the structural level) is equal to the number of arguments (in
the semantic representation). However, in their account, the semantic representation of the lexical meaning
of a verb essentially reduces to writing it in bold-face and providing it with some operators and its syntactic
argument variables.
Christian Lehmann, Situation types, valency frames and operations 6
2.1.1 Actants not mapped onto arguments
We will first briefly illustrate the first phenomenon with a few examples of obligatory verb actants
that have no semantic counterpart:
E7 opiodípote domátio tha káni
any:ever room FUT do:3.SG
‘any room whatever will do’
E8 Jedes Zimmer tut’s.
‘Any room will do.’
E9 Dein Rücklicht tut’s nicht.
Your backlight is not working.’
E10 Diese Idee bringt’s auch nicht.
‘That idea is not going to work, either.’
E11 prendersela con qualcuno
dump on / wade into / pick on somebody’
The predication intended in E7 – E9 requires a monovalent predicate. The speaker, however,
chooses a transitive verb, thus being left with a superfluous valency slot. English and Greek (E7)
just leave it unoccupied by introversive lability (s. §3.2.4), so that no mismatch arises. The German
counterpart of the verb in question has an obligatory direct object (E8f). It is represented by a third
person pronoun which would otherwise be anaphoric or deictic, but here refers to nothing. The
same is true for the highlighted pronouns in E10f.
Such semantically empty actants occupy a regular valency position of the verb, i.e. in structural
terms, the construction has nothing special to it. However, the clitic pronoun is neither omissible
nor substitutable, so there is no way to tell its reference. In other words, in coding the predicate with
its arguments, a verb has been chosen which has one valency position too many.
2.1.2 Participants not mapped onto arguments
Such cases as the above are, however, unsystematic, idiomatic and therefore of limited interest to
grammar. The converse case of participants that are not reflected in the argument structure is much
more important. They are present at the level of sense construction; but the predicate chosen has no
argument position for them, and consequently they do not appear as actants in the expression.
Consider, as a first example, intermediate relatives in the semantic representation of kin terms, as
exemplified in E12.
x is y’s uncle: y is child of z
and z
is child of z
and x is child of z
and x is male
Any decomposition of the sense of uncle must mention the intermediate relatives z
(y’s parent) and
(y’s grandparent) in order to account for the relationship of the uncle (x) to his nephew or niece
(y). However, the former two have no chance of being coded in an expression of the kind ‘x is y’s
As has been known since Jespersen (1924:88f), adverbs differ from adpositions in lacking a
governing slot. At the level of sense construction, however, they have a position for a participant
Christian Lehmann, Situation types, valency frames and operations 7
that is occupied deictically. For instance, E13.b is understood as implying a reference object that
Linda is in, just as E13.a does.
E13 a. Linda is inside the capsule.
b. Linda is inside.
The same applies to certain German verbs which are compounded with an adverb. For instance,
German packen ‘grasp’ (E14.a) is transitive, the undergoer being coded as direct object. The
compound verb zupacken, as in E14.b, likewise implies that the actor grasps an object. It is,
however, impossible to code this object, as the verb is intransitive.
E14 a. Erna packte den Dieb
‘Erna seized the thief’
b. Erna packte kräftig zu
‘Erna seized vigorously [anaphoric object] / sailed in’
A given lexeme representing a situation core in a language thus provides an argument frame for a
subset of participants to be accommodated as arguments in the construction. There may then be a
residue of participants which, although implied in the lexical semantics, cannot surface because no
argument position is provided. We may say that they are not exteriorized from the underlying
concept (cf. Lehmann 1991, §3.2 and Van Valin & LaPolla 1997, ch. on verbs of saying).
2.2 Participant properties
Participant roles are defined by heterogeneous criteria, viz. by their function in a situation type, but
also by absolute properties of their bearers. The relevant properties reduce to the position of the
referent in question on the hierarchy of Table 2 (also known as the animacy hierarchy):
Table 2 Empathy hierarchy
1 speech-act participant
2 other human being
3 animal
4 individual object
5 non-individual object
6 place
7 proposition
For many purposes, it suffices to lump certain levels of the empathy hierarchy together: #1 3 are
animate, #1 – 4 are individuals, #1 – 6 are concrete as opposed to #7, which is abstract.
2.3 Articulation of situations
The situation core is conceptualized as the core of the predication coded by a clause. The lexemes
chosen there may belong to any of the major word classes – nouns, verbs, adjectives and adverbs or
Christian Lehmann, Situation types, valency frames and operations 8
language-specific variants thereof. Focusing here on dynamic relational concepts (what typologists
sometimes call a ‘verbal concept’), adjectives may be foregone. Even if the word that fulfills the
syntactic function of predication is a verb, this does not necessarily convey the bulk of the lexical
meaning. Some important types of constructions which go beyond a simple verbal predicate include
the following: complex verb, verb series, light verb construction. A few comments on each of these
must suffice:
(a) The situation core may be coded in a complex verb, a compound like Ket at
(by.pouring-put) or German wegschütten (away:pour:I
) or a derivative like German verschütten
; cf. E37) ‘spill’. These participate in more or less regular alter-
nations to be discussed extensively in §3.2.
(b) The situation core may be articulated as a combination of verbs, i.e. a verb series, as in E15:
E15 Ade ju òkúta kan mi.
ORUBA Ade throw rock IND meet 1.SG.ACC
‘Ade threw a rock at me. / Ade struck me with a rock.’ (ValPal Database, Yoruba, (88))
The ways that verb series alternate to change the argument structure of the underlying predicate
remain to be investigated.
(c) The situation core may be categorized in some lexeme class distinct from the verb, which
will generally be combined with a verb in order to form a predicate. There are some variants of this
strategy: First, categorization of the situation core in terms of a noun may be the primary one. A
salient example is provided by weather phenomena like ‘rain’ and ‘hail’, which are primarily nouns
in quite a few languages. ‘Blink’, ‘scream’ and ‘cook’ are primarily nouns in Japanese, ‘sing’ is a
noun in several languages, and so forth. However, in no language is this the primary categorization
strategy used for dynamic relational concepts in general; cases like the ones mentioned obey at best
some subregularity (like the weather phenomena) and otherwise remain essentially idiosyncratic.
Consequently, the verbs supporting such a noun in a clause predicate are largely determined on a
lexical semantic basis, i.e. they will form phraseologisms with it. Alternations of such constructions
are largely idiosyncratic, too, and will not be treated in what follows. The regular and compositional
variant of the nominal strategy occurs if a predicate is primarily categorized as a verb in the
language, but this is nominalized and made dependent on a light verb, as in German etwas zum
Abschluss bringen (something to:
conclusion bring:
) ‘bring sth. to conclusion’, which
alternates with etwas abschließen ‘terminate something’. Finally, the inner dependent of a light verb
construction need not be a noun, but may belong to some other appropriate category,
as in Persian
xejālat kešid-an (shame pull-INF) ‘to get ashamed’ (s. Lehmann 2012, §3.1). This is a basic strategy
in languages like Jaminjung which have a closed class of verbs. Here again, a verb combines with
an inner dependent of some adverbial class to code a dynamic relational concept. Examples are
below in E24 and E28. The inner dependent carries the bulk of the lexical meaning, while the verb
serves as little more than a valency and aktionsart operator.
As long as the combination follows the rules of syntax in a compositional manner, that category may have a
freer distribution and thus be equatable with the noun, adjective or adverb of the language. To the extent that
the combination coalesces, words which may serve as inner dependent form a class of their own, dubbed
‘verb completor’ in Lehmann 2012, §3.1.
Christian Lehmann, Situation types, valency frames and operations 9
2.4 Basic types of situations
The more strictly relational properties of participants are derivative of the configuration of the
situation in which they participate. More precisely, they are largely determined by the situation core,
which appears as a predicate at the typological level. In this respect, the conception of participant
roles and semantic roles has changed since Fillmore (1968) first proposed case roles: Schank &
Abelson 1977, Van Valin & LaPolla 1997, ch. 3 and Fillmore 2003, §6 suggest that an analytic
approach that composes a proposition of a predicate and a couple of dependents each of which
contributes its semantic role to the complex is insufficient; and instead a holistic approach must be
taken which starts from types of situations (‘types of states of affairs’ in Van Valin & LaPolla 1997,
‘frames’ in Schank & Abelson 1977 and Fillmore 2003) and derives participant roles from these.
We shall see at the end of this section that, as usual in language, neither of the two perspectives is
sufficient in itself, and instead they must be combined. However, in the spirit of the top-down
approach taken here, we will start by defining types of situations. As already said, these definitions
relate to level #1 of Table 1, although their notation necessarily involves predicates and arguments.
Table 3 tabulates a set of basic situation types that underlie many situations and recur in the
specialist literature (Van Valin & LaPolla 1997, ch. 3). Some more will be introduced in subsequent
sections. Given the focus of the present volume on verbal valency, we limit ourselves to the more
dynamic situations; i.e. we exclude class inclusion and properties and start with states. The
examples given in the last column are for illustration. They do not represent English verbs, but
predicates which in many languages are primarily lexicalized in the argument-frame illustrated.
“The role that an entity plays in a state of affairs is always a function of the nature of the state of affairs,
and it is nonsensical to separate participant roles from the states of affairs in which they occur. Thus it is
states of affairs which are fundamental (i.e. basic), not participant roles (which are derived).” (Van Valin &
LaPolla 1997:89; cf. also pp. 86 and 113)
Table 3 Basic types of situation (s)
type dynamicity constellation participant properties control
roles example predicates
phase dynamic P
(s) s: abstract - start, end, happen
ambience stative/
durative A
(1) 1: place - 1: L rain, snow
state stative I
(1) - 1: O
existence stative E
(1, 2) - - 1: O
2: L there is, be located
(posture) stative P
(1, 2) 1: individual
2: place +/- 1: O
2: L stand, lie, sit
possession stative P
(1, 2) 1: concrete
2: animate + 1: O
2: Pr belong, have
physical state stative I
(1) 1: concrete - 1: O dry
mental state stative M
(1, 2) 1: human
2: abstract + 1: Ac
2: know, intend, refuse
emotional state stative E
(1, 2) 1: human
2: - +/- 1: Exp
2: will, want, like, please, fear
process /
event dynamic U
(1, s) 1: concrete - 1: O die, explode
change of state dynamic C
(1) 1: concrete U.aff (monovalent) burn, break, melt
motion dynamic M
(1, 2) 1: concrete
2: place - 1: U.loc
2: L fall, sink, roll
emotional event punctual E
1: human
2: - 1: Exp
2: frighten
action /
act durative /
punctual D
(1, s) 1: animate + 1: Ac work, bark
controlled motion dynamic M
(1, 2) 1: concrete + 1: Ac run, climb, jump
Christian Lehmann, Situation types, valency frames and operations 11
2: place 2: L
L=Source: come from, leave, go out
L=Goal: go to, come to, arrive at,
return, enter
L=Path: pass
2: animate meet
experience ~
sensation dynamic
(1, 2) 1: animate
2: concrete +/- 1: Ac
2: +attentive: look, listen, sniff
1: E
2: -attentive: see, hear, feel, smell, taste
relation - N
(1, 2) 1: abstract
2: abstract - cause, condition, entail, imply, prevent
Ac actor
Exp experiencer
L place
O object (non-specific central role)
Pr possessor
U.aff affected undergoer (= patient) considered undergoer
U.loc: locomoted undergoer (sometimes
called ‘theme’)
The first column of Table 3 labels the situation types. The participant and control properties
of columns 4 and 5 are to be taken as prototypical. The control of column 5 is a relation
between participant #1 and s, which extends to the other participants of s. The definition of a
type of situation is composed of the cells of columns 2 – 5. A participant role may be defined
by the set of properties of columns 3 – 5 of a (small) selected set of rows. The set of these
definitory features is then labelled in column 6. In other words, a concept like Agent (Ac) is
defined as the first argument of a set of predicates which is animate and controls the situations
in question.
At this level of generality, the argument frame of a predicate comprises all those
participants which may be relevant to characterizing the situation in question. Usually only a
subset of these will be used when the situation is conceptualized by a (type of) predicate. As a
tendency, the order of participants of a given situation (type) roughly reflects their relevance
for the predicate: the first participants are the central ones, the further towards the end of the
sequence a participant is positioned, the more peripheral it is. We will come back to this
distinction in a moment.
It is clear that the participant features of columns 4 and 5 do not suffice to distinguish
semantic roles. For instance, recipient, experiencer and addressee are not distinct by their
absolute and their control properties. They can only be distinguished by the situations in
which they function, viz. transfer (Table 6), experience and communication (Table 9), or in
other words, by the basic predicates T
, P
and C
ingredients they are.
On the other hand, a difference in the kind of participant may make a difference in the
kind of situation. This is importantly the case for human vs. non-human participants, which
condition distinct predicates in many cases. For instance, many – though not all languages
distinguish between G
(1, 2, 3) and P
(1, 2, 3) on the sole basis of the feature +/- human
of argument 3 (transfer vs. collocation in Table 6). Similarly, a language may have two verbs
for ‘wash’ depending on the animacy of the object. Moreover, there are situation types, as in
particular process vs. action, which differ exclusively by the control of their first participant.
It therefore appears that semantic role and situation type are interdependent and determine
each other.
We finally come back to the distinction between central and peripheral participant roles.
The central ones are constitutive of their situation, while the peripheral ones may freely be
added or omitted without affecting the nature of the situation. None of the participants
appearing in Table 3 is entirely peripheral. Examples of participants which are peripheral to
most situations (although not to their definitory situation types; see below) include the causer,
the beneficiary and the instrument.
E16 (mare) nunc qua a sole conlucet albescit
sea:NOM.SG now where from sun:ABL.SG shine:PRS:3.SG white:INCH(PRS):3.SG
‘the sea now becomes white where the sun makes it glisten’ (Cic. Luc. 105, 16)
E17 Linda sold books for her cousin.
E18 Linda solved the problem with a calculator.
While the beneficiary in E17 and the instrument in E18 require no explanation, the causer in
E16, viz. sol ‘sun’, is coded by the causer adjunct strategy (Lehmann 2016, §3.4), which is
less familiar. What is important at the moment is that there is nothing in the process of shining
that would imply a causer, nothing in the notion of selling that would entail the presence of a
Christian Lehmann, Situation types, valency frames and operations 13
beneficiary in a selling situation and nothing in the notion of problem-solving that would
require an instrument. The most peripheral semantic roles are independent of the nature of the
situation in which they appear. Therefore, the holistic approach which derives semantic roles
from situation types cannot mean that all semantic roles are an outgrowth of simple situations
only to be grasped holistically. There are composite situations, properly including the agentive
situation (s. Table 6 below), the benefactive situation and the situation involving an
instrument, which an analytic approach reveals as formed in a compositional way from a
basic situation and an additional participant.
More precisely, the participants of such a complex situation include a base situation s, as
agentive situation: C
(1, s), where 1 = causer
benefactive situation: G
(1, 2, s), where 1= benefactor and 2 = beneficiary
instrumental situation: U
(1, 2, s), where 1 = Ac and 2 = I.
Thus, these peripheral semantic roles may, again, be conceived as deriving from the nature of
the respective situation type. This, however, does not change the fact that they are not implicit
in the base situation s. This consideration, thus, leads to the same conclusion as before: The
holistic approach to situations cannot be set as absolute. In particular, central participant roles
are substantiated by certain basic situation types; but peripheral participant roles have their
own properties which they contribute in a like fashion to many different situations.
2.5 Merger of basic situation types: action-processes
Many situations with more than one participant can plausibly be analyzed as combinations of
a base situation with an additional participant. These will be treated in §3.2. There remains
one basic situation type which cannot plausibly be analyzed in such a way, and this is the
action-process. Table 4 displays its formation and a few important subtypes.
Table 4 Action-processes
constellation participant
roles subtypes example
process /
punctual D
(1, 2)
1: individual
2: concrete + 1: Ac
2: U.aff U: -animate
U: +animate
sew, eat /
beat, grasp
action/act durative/
(1, 2) 1: human
2: inanimate
+ 1: Ac
2: U: concrete
U: abstract read, count /
(1, 2) 1: animate
2: inanimate
+ 1: Ac
2: U.eff make, build,
speak, utter
U.eff: effected undergoer
As its name indicates (cf. Chafe 1970, ch. 11), the action-process is the fusion of an action
with a process, as one of its arguments acts, while the other one undergoes the situation as a
process. In the examples adduced in Table 4, the fusion is complete, i.e. the two situations
Christian Lehmann, Situation types, valency frames and operations 14
cannot be disentangled in such a way that, for instance, a situation of eating would be
composed of an intransitive act of eating and an intransitive process of undergoing ingestion.
Instead of an addition of a particular argument to a self-sufficient base situation, such action-
processes are more plausibly conceived as the symmetric and irreducible merger of an action
and a process.
The first argument of an action-process is an actor (Ac), the second is an undergoer (U).
Mental actions and acts are not among the prototypical action-processes because the
undergoer is not affected. This kind of unattained undergoer is categorized as ‘considered
undergoer’ ( Equally non-prototypical are situations of production, since their
undergoer is effected rather than affected (U.eff). The taxonomy is as follows: U is a kind of
O. U.aff, and U.eff are specifications of U which prove relevant in some valency
patterns. U.aff is the same as patient.
Being a basic situation type, the action-process may serve as a model for the productive
formation of complex situations on the basis of simpler situations:
On the basis of a process, a derived action-process may be formed by introducing an
On the basis of an action, a derived action-process may be formed by introducing an
These are two of the operations on semantic roles to be surveyed in §3.2.
3 Argument-structure operations and alternations
3.1 Two types of argument-structure operations
From an onomasiological point of view, we start from a certain situation with its core and its
participants and code it in a syntactic construction of a particular language. This may be
described as a transition in two steps:
1. From among all the participants and features of the situation, a selection is made which is
conceptualized as a predicate or combination of predicates with their central and
peripheral arguments and the semantic roles of the latter. The predicates are mapped onto
a set of lexemes of the language each of which is represented by a stem and which are
combined syntagmatically as indicated in §2.3. The arguments are represented as a set of
referential expressions, which will not occupy us further. Each of the stems involved has a
certain valency, which is an abstraction of the set of constructions (viz. diatheses) that
forms of this stem may be used in. For instance, a stem like eat has the valency of being
monotransitive, which includes the possibility of forming a passive.
Importantly, stems
derived from the same root may differ in their valency.
2. Given a certain verbal lexeme with the stem representing it, the latter is inserted in a
particular syntactic construction by conjugating it in a particular form and combining it
with (a subset of) its dependents in particular syntactic functions. Such a construction is a
diathesis of the verb stem.
Given the tradition of valency grammar, which includes, among other things, the production of
valency dictionaries, it is inadvisable to speak of different valencies with respect to diathetic alternants
of a verb. See Lehmann 1992.
Christian Lehmann, Situation types, valency frames and operations 15
In both steps, a set of alternative representations is available which bear paradigmatic
relations among them. In a dynamic perspective, such paradigmatic relations may be
described as alternations of constructions or as (directed or symmetric) operations that
transform one construction into another.
In step 1, an abstract construction is selected from a set of alternatives each of which
involves stems in certain word classes. These stems differ in their valency (and, possibly, their
aktionsart). The paradigmatic relations among alternate conceptualizations of a situation
involve valency changes. The latter concern, importantly, the semantic roles associated with
the predicates, i.e. they change the conceptualization of a situation by representing some
participants rather than others in the form of arguments of a predicate, by determining the
centrality vs. peripherality of each of the arguments and by changing the first argument’s
control feature. The function of these operations is to create and change particular predicates
with particular constellations of arguments. An operation fulfilling this function may be called
a valency operation (or semantic role operation). Consequently, the variants at that level are
not synonymous (just as sit and set are not synonymous); and the semantic differences among
them may be peculiar to the particular verb or verb class.
In step 2, a particular verb stem is given, and the variants that are in paradigmatic
relationship are its diatheses, i.e. the verb forms in the appropriate voice (if any) with their
respective complements and adjuncts. These paradigmatic relations may be described by
operations that transform one diathesis into another, e.g., an active into a passive construction.
They operate on verb forms and syntactic functions, i.e. they change the relations of nominal
components to the clause core by allowing the speaker to select between a clause that does or
does not comprise a certain syntactic component, and by changing the latter’s syntactic
relation. This is generally done in order to adapt that syntactic component to the thematicity
of its referent. An operation fulfilling this function may be called a diathetic operation (or
syntactic function operation). Such changes leave the semantic roles intact. Consequently,
the variants at this level (like Linda eats the apple and the apple is eaten by Linda) are either
synonymous or, at least, their semantic differences are a compositional consequence of the
application of general grammatical rules.
Diathesis concerns the coding of arguments of a predicate as complements and adjuncts
of a verb, thus, the conversion of semantic roles into syntactic functions.
It comprises both
syntactic operations and inflectional processes. One of the latter is voice, a conjugation
category coding diathesis (s. Kulikov 2011, §1).
The distinction between a valency operation and a diathetic operation is best illustrated by
an example that is well established in the literature (cf. Kulikov 2011:392), viz. the contrast
between deagentive (alias anticausative) and passive:
E19 a. Hwaane’ t-u kach-ah le che’-o’
YM John-TOP PRFV-SBJ.3 break-CMPL DEF wood-D2
‘John broke the stick’
b. le che’-o’ h káach (*tuméen Hwaan)
DEF wood-D2 PRFV break\DEAG by John
‘the stick broke (*by John)’
“Diathesis is determined as a pattern of mapping of semantic arguments onto syntactic functions
(grammatical relations).” (Kulikov 2011:370)
Christian Lehmann, Situation types, valency frames and operations 16
c. le che’-o’ h ka’ch (tuméen Hwaan)
DEF wood-D2 PRFV break\PASS by John
‘the stick was broken (by John)’
The transitive construction in E19.a has two intransitive counterparts, viz. the deagentive
construction of E19.b, whose verb root has a high tone, and the passive construction of #c,
whose verb root bears an infix. The functional difference is that the passive (just like the
active of #a) implies the participation of an actor, which in #c is less thematic than the
undergoer, but may be coded in an agentive prepositional phrase, while the deagentive
excludes the presence of an agent phrase in the clause, thus inviting the inference that the
situation happens spontaneously, i.e. without the intervention of an actor. Deagentivization is
a valency operation or semantic role operation, passivization is a diathetic or syntactic
function operation. By definition, if the members of an argument-structure alternation provide
for the syntactic representation of a different number of arguments, it is a valency alternation.
These two functional types of argument-structure operations are ultimately subordinate to
the cognitive and the communicative functions of language, respectively. They are clearly
distinct in principle. However, since coding strategies are typically polyfunctional, a particular
argument-structure process may combine a semantic function with a discourse function. And
on the basis of some parallelism between the two functions, some of these functional bundles
are relatively common, having often made it difficult to disentangle the two types of
operations. One case in point, viz. lability, is discussed in §3.2.2. Moreover, diathetic
operations typically involve promotion and demotion; and these are not always easily
distinguished from the valency operations of argument introduction and suppression, resp.
Specifically, the applicative is not categorially distinct from extraversion (undergoer
introduction); and passive and antipassive do not differ sharply from the valency operations of
deagentivization and introversion, resp. This problem will be taken up in §3.3.7.
Just as the semantic roles and syntactic functions themselves, linguistic operations on
them are conceived at the cross-linguistic semantic level (Table 1, #2). That is, they may be
instantiated in several languages in like fashion, but they are typically not instantiated in all
languages. It is important to appreciate the trade-off between basic lexicalization of a situation
core and the set of operations generating alternants of it (cf. Lehmann 2012): The base verb
coding a certain situation core in a language may be an intransitive verb, and this may require
an operation of transitivization if more participants of the situation are to be accommodated in
central syntactic positions. The basic categorization may appear as firmly given and the
operation as a flexible way of getting beyond the default. However, the operations need an
operand to operate on; some choice must be made to begin with. What is actually given at a
certain stage in the diachrony of a language is a pair of basic lexical categorization and a set
of operations to adapt it; and that pair is subject to change. For instance, Latin at some stage
antedating the written documentation had an intransitive verb specio ‘look’. It also had the
process of preverbation, which had extraversive side effects, thus producing (among other
compounds based on this root) aspicio (tr.) ‘look at, see’. Latin itself no longer has the
simplex, being left with a set of transitive ‘see’ verbs like aspicio. At this stage, it uses
undergoer lability (s. §3.2.4) in order to get rid of the argument provided in the valency but
occasionally not needed. In other words, what is generated by an alternation in one language
or at one stage of a language appears as a base form in another language. For the functions of
cognition and communication, the choice does not make a big difference. It may, however, be
relevant for the linguistic type.
Christian Lehmann, Situation types, valency frames and operations 17
3.2 Valency operations
3.2.1 Formal relations in alternation
Alternations between valency frames may be systematized by a variety of formal criteria. The
first criterion concerns the paradigmatic relation between the alternants. The alternation may
1. symmetric, or undirected
2. asymmetric, i.e. directed in the sense that one alternant is basic, the other is derived from
The second criterion concerns the coding of the alternation. It may be
1. coded by segmental means
a) on the verb
b) elsewhere in the clause (i.e., generally on the dependents)
2. not coded by segmental means (i.e. the alternation reduces to presence vs. absence or a
different order of constituents).
These distinctions will be illustrated by examples in §3.2.3ff. First, however, a
methodological problem requires some discussion: What is the criterion for directionality of
an alternation; in other words, how do we know which of two alternants, if any, is basic and
which derived? What we require here are criteria intrinsic to the language system, i.e. we
forego both considerations of frequency and customariness and evidence of historical
primacy. The general criterion relevant here is markedness of the derived variant. In the clear
case, this involves an additional morpheme with an additional semantic (or grammatical)
feature as opposed to the base variant. Thus, the German applicative using the be- prefix is a
directed coded alternation, where herrschen ‘reign (intr.)’ is the basic, beherrschen ‘dominate
(tr.)’ the derived variant, even if the text frequency of the latter is higher than that of the
former. Similarly, if an agentive verb displays an alternation between a transitive stem with an
undergoer argument and an intransitive stem without it, then this is a case of extraversion if
the former is marked (as in E30 below), and one of introversion if the latter is marked (as in
No morphological markedness is to be discerned if the alternation is either uncoded or if
both alternants are formally equally complex. The following subdivision applies here (cf.
Haspelmath 1993, §2):
1. In both of the alternants, the same verb stem appears; i.e. the alternation is not coded by
segmental means on the verb. Depending on the theoretical approach, this is conceived as
conversion or category indeterminacy. An important subcase, called lability, is the use of
the same verb stem in transitive and intransitive function, as in English break (tr./intr.).
2. Each alternant shows a different verb stem, which are morphologically unrelated. This is
a lexical alternation, as in Yucatec Maya took (tr.) ‘burn’ – éel (intr.) ‘burn’. This coding
could be called suppletive to the extent that the paradigms in question are productive and
Lability is sometimes called ambitransitivity (e.g. in Mithun 2000). Since lability is the traditional
term for use of the same verb stem in both transitive and intransitive constructions, ambitransitivity
may be used for a slightly wider concept, viz. use of the same lexeme in both constructions, allowing,
thus, for stem alternations (typically appearing as conjugation classes).
Christian Lehmann, Situation types, valency frames and operations 18
regular. In the case at hand, the language has a productive and regular morphological
causativization process that could easily apply to éel. Consequently, tóok would be a
suppletive agentive (or causative) of éel. Likewise, French montrer ‘show’, a lexical
agentive of voir ‘see’, could be called a suppletive agentive since there is a regular
process of causativization, so that montrer could be seen as a lexicalization of faire voir
‘make see’.
3. The alternants contain the same verbal base, with each of them bearing some
morphological mark. This is an equipollent alternation, as in Jap. atum-aru (intr.) – atum-
eru (tr.) ‘gather’.
If none of the alternants bears a morphological mark lacking from the other, it may still be
possible to diagnose a directed alternation. Namely, an alternation is directed if one of the
alternants is subject to special constraints or carries a certain semantic feature absent from the
other; i.e., it is functionally marked. German has some actor-labile verbs (s. §3.2.3). The
criterion just mentioned determines that this alternation is undirected with some verbs, but
directed with others. E20 illustrates actor lability for rollen ‘roll’.
E20 a. Erna rollte den Reifen auf die Straße.
‘Linda rolled the hoop onto the street.’
b. Der Reifen rollte auf die Straße.
‘The hoop rolled onto the street.’
The difference in the distribution and meaning between the transitive version in #a and the
intransitive version in #b reduces to the presence vs. absence of an actor; no constraints on the
distribution or other nuances of meaning of either of the versions are involved.
There does
not appear to be a way of determining the direction of the alternation, i.e. to speak of
agentivization or deagentivization with respect to German rollen.
E21.a and b illustrate the same uncoded alternation. The versions #b and #c share the
absence of an actor. Here, the version #b without the parenthesized reflexive pronoun carries
the semantic nuance of characterizing the subject by a property. The deagentive reflexive
construction illustrated by #c does not have this feature and thus relates semantically to #a in
the same way as E20.b relates to #a.
E21 a. Erna schloss die Tür.
‘Linda shut the door.’
b. Die Tür schließt (
sich) nicht dicht.
‘The door does not close tightly.’
c. Die Tür schloss
(sich), und wir waren gefangen.
‘The door closed, and we were caught.’
To the extent that this distributional and semantic difference between the transitive and the
intransitive versions of E21.a and b represents a subregularity, the conclusion is that the
former is basic, the latter derived. The same criterion would apply in seemingly symmetric
diathetic alternations like the English locative shift; s. §3.3.5.
To be sure, each of the valency alternants of the stem may develop its own polysemy or idiomatic
uses. For instance, the intransitive der Verkehr rollt ‘traffic is rolling’ has no transitive counterpart.
This would not count as a systematic constraint.
Christian Lehmann, Situation types, valency frames and operations 19
We are not entering into the details of the formal techniques of coding an alternation, or
signalling an operation. It suffices to recall that, given that we are talking about the formation
of a predicate with its argument frame at the lexical level, relevant structural processes are, in
principle, lexical (compounding or derivational) in nature. To the extent that such a process is
grammatical (syntactic and/or inflectional) in nature, it comes under diathetic operations
(including voice) rather than valency operations.
3.2.2 Types of valency alternations
The maximum quantitative valency for which there are dedicated general operations is
trivalency. Most plurivalent constructions may be described in terms of three macro-roles,
actor, undergoer and indirectus. While the former two have been presupposed throughout (s.
Foley & Van Valin 1984), the indirectus needs to be defined (s. Lehmann et al. 2004): It is the
macro-role which neutralizes the specific semantic roles of recipient/emitter, addressee,
experiencer, beneficiary and sympatheticus and which is typically coded as an indirect object
and/or by a case resembling the dative. Since the two most central arguments are mostly
coded as actor and undergoer, the indirectus appears typically – although not exclusively – as
the third argument of a predicate. The argument in question is prototypically human. If
entities lower on the empathy hierarchy take this macrorole, the goal may join the set of
semantic roles comprised by it. Typical examples appear below in E38 – E41. While actor and
undergoer are universally applicable descriptive concepts, some languages have an indirectus,
others do not.
Valency operations represent a paradigmatic relationship between two predicate-argument
frames which differ in that one comprises a certain argument which the other lacks. The most
important of these paradigmatic relations are based on the inclusion of the macro-roles: the
frame does or does not involve an actor, an undergoer or an indirectus, resp. Table 5 presents
the alternations ordered by the criteria discussed in §3.2.1. In the first two columns, an
intransitive verb alternates with a transitive verb. In the last column, alternation is
prototypically between a monotransitive and a ditransitive verb, although exceptions are
possible and some will be noted in §3.2.6.
Table 5 Alternations between presence and absence of macro-roles
on verb
actor undergoer indirectus
symmetric none actor lability undergoer lability indirectus lability
lexical lexical agentive lexical extraversive
lexical indirectus
equipollent equipollent
agentive equipollent
extraversive equipollent indirectus
asymmetric suppression deagentivization introversion indirectus suppression
introduction agentivization extraversion indirectus introduction
Christian Lehmann, Situation types, valency frames and operations 20
3.2.3 Actor alternations
Most of the situations in Table 3 may be expanded into an agentive situation by an operation
that adds an actor, as follows: A complex situation is created whose highest predication is
(1, s), where 1 is the actor and s is the base situation. The difference between a basic
action-process and an agentive action-process lies in the conceptual separability of the actor
and the rest of the situation: While it is not naturally possible to extract the actor from the
example predicates given in Table 4, addition of an actor which is separable from the core
concept is a natural interpretation of such agentive action-processes as ‘burn’, ‘break’, ‘melt’
(all taken to represent bivalent predicates), shown in Table 6.
Nevertheless, the concept of the actor is given with the basic action-process, and it may
therefore serve as a model for agentivization. The predicates most amenable to it are probably
those that display widespread actor lability, like the ones just mentioned. From there,
agentivization may apply to further basic situations to turn them into agentive situations. This
generates a large number of additional situation types. Importantly, application of this
operation to bivalent situations yields trivalent situations. Table 6 displays some agentive
situation types, together with their base as it appears in Table 3. Participants 2 and 3
correspond to #1 and 2, resp., of the base situations. By virtue of the agentivization, the O of
the base situation becomes an U, and Pr becomes R/Em. In Table 6, the column ‘control’ is
omitted, as participant 1 always controls s. Likewise, the column ‘subtypes’ is unnecessary
since these appear in the table lines.
Table 6 Types of agentive situation
type base
(Table 3) constellation participant
properties roles example
situation s C
(1, s) 1: individual 1: Ac cause
change of
change of
state C
(1, 2)
1: individual
2: concrete 1: Ac
2: U.aff burn, break,
melt (tr.)
transport uncontrolled
AUSE (1,
3)) 1: animate
2: concrete
3: place
1: Ac
2: U.loc
3: L
bring, carry,
throw, push
collocation position
1: animate
2: individual
3: place
1: Ac
2: U.loc
3: L
put, seat, lay
transfer possession
AUSE (1,
OSS (2,
3)) 1: animate
2: concrete
3: animate
1: Ac
2: U.loc
3: R/Em
give, take
motion &
AUSE (1,
2)) &
SE (1, 3, s)
FFECT (1, 2)
1: human
2: concrete
3: concrete
1: Ac
2: U.aff / L.Goal
3: I / U.loc
fill, load,
sprinkle, stuff;
hit (I against U),
experience experience
1: human
2: animate
3: concrete
1: Ac
2: Exp
show, hide
Christian Lehmann, Situation types, valency frames and operations 21
Em emitter
R recipient
Comparison among the situation types reveals the following paradigmatic relations:
An agentive change of state differs from a basic action-process (of Table 4) essentially by
the separation of the agent from the process. Consequently, the agent may be suppressed from
the former, but not from the latter situation type. (The agentive change of state is not called
causative because it need not be produced by a causative derivation.)
Transport is like collocation in requiring U to move with respect to L. It differs minimally
from collocation in that the latter involves a resulting position of U at L.
Transfer, too, is like collocation in that both require U to move (given that possession
requires contact between possessor and possessed, change of possession by default implies
locomotion for the possessed). The difference between the two stems from the difference
between the respective base situations: the last argument is prototypically a place in position
and collocation, but an animate being in possession and transfer.
In situations of manipulation, the agent affects (manipulates) one, stationary object by
applying another, movable object to it. See §3.3.5 for alternate conceptions of this
The following alternations relate the situations of Table 6 to those of Table 3:
Actor lability (“patientive ambitransitivity” in Mithun 2000):
E22 a. The pot broke.
b. Linda broke the pot.
Lexical agentive:
E23 a. Linda died.
b. Irvin killed Linda.
Equipollent agentive:
E24 a. ngabulgja=biya yirra-gba=ni wangguwarla-nyunga
‘we were washing/bathing because of (i.e. to get rid of) the saltwater
b. ngabulg=gun ba-rra jalig majani hot gan-unggu-m
bathe=CONTR IMP-put child maybe hot 3SG.A:3SG.P-say/do-PRS
‘bathe her (the child), maybe she is hot (child, in river)’ (ValPal Database,
Jaminjung, (162) and (23))
E24 shows the adverb carrying the bulk of the meaning of the predicate combined with an
intransitive verb in #a, but with a transitive verb in #b, rendering intransitive and transitive
‘bathe’, resp. Given that these verbs are semantically empty like light verbs, they function like
valency operators.
This generally involves the anticausative as a morphological operation on a transitive verb,
coded by high tone on the root in E3.b.
Christian Lehmann, Situation types, valency frames and operations 22
E3. a. t-in ch'am-ah u chuun le che'-o'
YM PRFV-SBJ.1.SG bruise-CMPL POSS.3 base DEF tree-D2
‘I bruised the trunk of the tree’ (EMB&RMC_0033)
b. h ch’áam u chuun le che’-o’
PRFV bruise\DEAG POSS.3 base DEF tree-D2
‘the trunk of the tree got bruised’
By far the most important subtype of agentivization is causativization, which involves a
morphological operation on the base verb, like the suffix in E25.b.
E25 a. h he’l-ech
‘you rested’
b. t-in he’-s-ech
‘I put you to rest’
However, a hyperonym like agentivization is needed, as there are also nominal strategies
thereof, already exemplified by E16 in §2.4.
3.2.4 Undergoer alternations
Some actions and acts are compatible with an undergoer that they extend to. There are,
consequently, alternate views of such situations, always with an actor, but with or without an
undergoer. For instance, a situation of thinking may primarily be conceived as being based on
a solipsistic actor and only secondarily be taken as the basis for an operation of undergoer
addition, which in this case may supply the theme that the thinking is devoted to or the
proposition effected by it. The semantic operation of adding an undergoer role to an action is
called extraversion.
Like the agentive action-process, it takes the basic action-process as a
model and creates derived action-processes. Table 7 shows the internal structure of such
Table 7 Type of extraversive situation
type base
(Table 3) constellation participant
properties roles example predicates
action-process action/act D
(s, 2) - 1: Ac
2: U serve (sb.), sweep (a
The following are examples of the kinds of alternation summarized in the undergoer column
of Table 5:
For the sake of simplicity, E25 is presented in a verb status that conditions morphologically ergative
In Lehmann & Verhoeven 2006, the term is used as the lexical counterpart to the (supposedly)
syntactic operation of the applicative.
Christian Lehmann, Situation types, valency frames and operations 23
Undergoer lability (“agentive ambitransitivity” in Mithun 2000):
E26 a. Linda hunts.
b. Linda hunts the bear.
Lexical extraversive:
E27 a. Linda spoke to Irvin.
b. Linda said ‘Hello’ to Irvin.
Equipollent extraversive:
E28 a. mayi gambaja ga-yu
AMINJUNG man laugh 3SG.S-be.PRS
‘the man is laughing’
b. mayi-ni gambaja gani-mangu janyungbari
man-ERG laugh 3SG.A:3SG.P-hit.PST.PFV other
‘the man laughed at the other one’ (ValPal Database, Jaminjung, (59) and (158))
E28 illustrates the same kind of construction as E24, except that the transitive verb of E28.b
does not have a causative, but an extraversive effect.
In introversion, effectuated in Yucatec Maya by low tone on the root vowel (E29.b), the
undergoer slot is blocked so that there remains no way of mentioning the participant in
question in that clause. This distinguishes this alternation from the antipassive, which only
demotes the undergoer; s. §3.3.2.
E29 a. k-in xok-ik (le analte’-a’)
‘I read it / this book’
b. k-in xook
‘I read/study’
In extraversion, effectuated in Yucatec Maya by means of an extraversive transitivizing suffix
(E30f), a participant of the situation is integrated which is absent from the base predication.
This distinguishes extraversion from the applicative, which promotes a clause component to a
more central position; s. §3.3.3.
E30 a. k-in ts’íib
YM IMPF-SBJ.1.SG write
‘I write’
b. k-in ts’íib-t-ik (le analte’-a’)
‘I write it / this book’
E31 a. k-in meyah
‘I work’
b. k-in meyah-t-ik-ech
‘I serve you’
Christian Lehmann, Situation types, valency frames and operations 24
3.2.5 Relations between actor and undergoer alternations
The representation of Table 5 suggests a set of mirror-image relations between the
paradigmatic relations and corresponding operations concerning actor and undergoer. One of
these may be formulated as follows: Deagentivization undoes the effect of agentivization, just
as introversion undoes the effect of extraversion. This symmetry is, in fact, reflected in
linguistic structure to a certain extent. For instance, the causative as illustrated in E25
introduces an additional highest agent. This effect is undone by the reflexive appearing in
E32.b, as this marks coreference of the new argument with the argument already present in
the base. The result is near-synonymy of E32.a and b.
E32 a. k’abéet a mas he’l-el
YM necessary SBJ.2 more rest-INCMPL
‘you must rest more’
b. k’abéet a mas he’-s-ik a báah
necessary SBJ.2 more rest-CAUS-INCMPL POSS.2 self
‘you must get yourself more rest’ (BVS_10.01.09)
However, this symmetry has limits. The following two subsections deal with the asymmetries
between these operations. The agentive – extraversive asymmetry
Agentivization and extraversion were introduced in Table 5 as mirror images. Just as
agentivization introduces an actor that causes the base situation, so extraversion introduces an
undergoer that is concerned by the base situation. This is depicted in Diagram 1:
Diagram 1. Agentivization and extraversion
base s
derived C
(A, s) C
(s, U)
agentive extraversive
This symmetry extends, to some extent, to the internal composition of s in Diagram 1: Since
the causative adds an actor, it applies most easily, and most commonly, to situations which
comprise an undergoer, but lack an actor (Lehmann 2016, §2.3). Conversely, since the
extraversive adds an undergoer, it applies most easily, and most commonly, to situations
which comprise an actor, but lack an undergoer. However, this symmetry concerns only the
prototypes. Apart from these, there are essential asymmetries between the two operations as
they appear in linguistic structure.
What appears, at first sight, as a symmetry, viz. the mirror image relation of the argument
structure of the two base situations just mentioned, proves to be an asymmetry on deeper
inspection: The presence of an actor presupposes a dynamic situation (with the partial
exception of controlled postures), while the presence of an undergoer makes no requirement
on the dynamicity of the situation. As a consequence, extraversion of a stative situation is the
Christian Lehmann, Situation types, valency frames and operations 25
exception, while agentive alternants may easily be formed from stative situations. Such
alternants are, in fact, so basic and widespread that they often differ structurally from
causatives based on dynamic situations, a fact which earned them the term ‘factitive’ in
descriptive linguistics.
While causatives are formed more easily on the basis of situations that lack an actor,
nothing in principle excludes agentivization of an active situation. As a result, none of the
situation types enumerated in Table 3, Table 4 and Table 6 is in principle immune to
agentivization. A productive causativization process may causativize even causative
constructions. This is in sharp contrast with the productivity of extraversion: only active
situations may be extraverted; and the operation is not recursive.
The two operations differ also in their structural manifestations: Most languages have a
periphrastic causative construction based on a verb which means something like ‘do’
(including ‘make’, ‘cause’ and the like), thus coding pretty much the semantic structure
shown on the left-hand side of Diagram 1. Its extraversive mirror image would be a
periphrastic applicative construction based on a transitive light verb which means ‘affect’,
‘extend to’, ‘concern’ or the like, thus coding the predicate appearing on the right-hand side of
Diagram 1. While such a construction is certainly not unheard of, it is not the default
applicative construction; and existent applicative morphology, to the extent that its etymology
may be ascertained, is generally not grammaticalized (or lexicalized) from such verbal bases.
The semantic role born by the actor introduced by agentivization is essentially unitary: It
is the argument that has highest control in the situation; thus, a prototypical agent. This is true
whether the base situation already comprises an actor or not. On the other hand, the semantic
role born by the undergoer introduced by extraversion varies considerably (Peterson 2007)
and depends essentially on the meaning of the predicate and of the undergoer constituent. For
instance, in E33.b from Warembori (Lower Mamberamo, Indonesia), the fact that the river
serves both as a place and as an instrument in the situation follows exclusively from the
meanings of the verb and the undergoer plus world knowledge. The applicative suffix does
nothing but transitivize the verb.
E33 a. make matin-do (nana ipa-yave)
AREM boy wash-IND OBL river-DEF
‘(the) boy is washing (in a/the river)’
b. make matin-na ipa-yave
boy wash-APPL river-DEF
‘(the) boy is washing in the river’ (Donohue 1999:9)
The picture offered by Diagram 1 thus hides a basic asymmetry: While the additional Ac in a
causative construction does bear the prototypical agent role which is appropriately represented
by some such predicate as C
, the role of the additional undergoer in an extraversive
construction is not the prototypical patient role (Kittilä 2011:354) and therefore only
characterized rather vaguely by the predicate C
. Putting it yet another way:
agentivization is semantically specific in a way that is compatible with many base situations
in essentially the same way, while extraversion is semantically non-specific, gets its specific
relational meaning from the context and is yet incompatible with many situation types.
Christian Lehmann, Situation types, valency frames and operations 26 The deagentive – introversive asymmetry
Again, just as the deagentive blocks the actor argument, the introversive blocks the undergoer
argument. And it is true that these two operations are symmetric to a certain extent. For
instance, quite a few languages use one detransitivizing process to achieve both. The Russian
reflexive is a case in point (cf. Kulikov 2011:376, 382): From the transitive base rugat’
‘scold’, the reflexive shows an introversive meaning: rugat’sja ‘grumble, curse’; but on the
transitive base razrušat’ ‘destroy’, the reflexive razrušat’sja ‘get ruined’ has a deagentive
Quite generally, given a construction produced by an operation that introduces a certain
argument, then that argument cannot be omitted in the construction, since its presence is
exactly what that operation conveys. Instead, the obvious way of getting rid of the argument
in question is simply not to apply the operation in question. Consequently, there is generally
no deagentive of a causative;
and likewise there is no introversive of an extraversive.
However, it is not the same transitive verbs that may be deagentivized and introverted.
Almost all of the basic action-processes of Table 4 may easily be introverted, but can hardly
be deagentivized. Likewise, extraversive action-processes need not be introverted; it suffices
to revert to their base; but they cannot easily be deagentivized, either. Conversely, the
agentive changes of state of Table 6 are easily deagentivized by reverting to their base, but
hardly introverted. In the opposition between agentive and extraversive action-processes,
basic action-processes thus side with the extraversive ones. The actor is constitutive for them;
if it is eliminated, a different situation (or none at all) results. This may indicate that basic
action-processes are not as balanced as assumed in §2.5 and that instead they are essentially
actions that extend to an undergoer.
3.2.6 Indirectus alternations
While there are elementary, i.e. undecomposable, monovalent and bivalent predicates,
probably all trivalent predications can be decomposed into combinations of bivalent
predications. If a non-first argument of a trivalent predicate is high on the empathy hierarchy,
it is most probably an indirectus. There are essentially two ways that such a situation may be
composed. One is by an expansion of a bivalent situation which demotes one of the basic
arguments to indirectus function. For instance, upon agentivization of a possessive situation,
we get a transfer situation, whereby the possessor becomes an indirectus. Upon agentivization
of an experiential situation, the experiencer remains or becomes an indirectus. Likewise, the
goal or recipient of a transport has that macrorole, too. Such cases were subsumed in Table 6
and need not be repeated in Table 8, although some relevant examples will be given below.
On the other hand, upon introduction of an undergoer in a bivalent situation that already
Constructions like E32.b constitute an explicable exception to this, as the actor introduced by
causativization is not actually suppressed in the reflexive construction, but rather identified with the
The latter is, incidentally, the reason why the ‘omission test’ by which Germanists seek to
distinguish between complements and adjuncts (hoping that complements are non-omissible) works
well for derived transitive verbs such as bearbeiten ‘process’ and the like, but shows nothing for base
transitive verbs like jagen ‘hunt’.
Christian Lehmann, Situation types, valency frames and operations 27
contains an animate being as second argument, the latter may be demoted to indirectus
function. Since these cases involve demotion, they will be reviewed in §3.3.3 (s. E56). One of
the most important situations involving an indirectus, viz. communication, will be analyzed
separately in §3.2.7.
The other way of expanding a predication by an indirectus is by introducing it without
further change, normally in a bivalent situation. This is schematized in Table 8.
Table 8 Indirective situations
type base
(Table 4) constellation participant
properties roles example
situation s D
(3, s)
1: animate
2: -
3: animate
1: Ac
2: U
3: Ind
tell, excuse
In what follows, the alternations provided for in the last column of Table 5 will be illustrated.
Indirectus lability:
E34 a. Linda brought the packet.
b. Linda brought me the packet.
Lexical indirectus alternation:
E35 a. Erna entschuldigte den Lapsus.
ERMAN Linda excused the:ACC lapse
b. Erna verzieh mir den Lapsus.
Linda forgave me.DAT the:ACC lapse
Equipollent indirectus alternation:
What is sought here is a pair of derivations of a common base one of which produces a
monotransitive stem while the other produces a ditransitive stem, with actor and undergoer
kept constant. If the process were anything like regular, the base would probably have to be
intransitive. No data corresponding to this construct have been found.
Indirectus suppression:
Most of the processes known in this domain demote rather than suppress the indirectus (see
§3.3.2). However, one of the many functions of the German prefix ver- is exactly that
(although E37 is relatively marginal because the role suppressed is the goal).
E36 a. Erna meldete mir den Erfolg.
ERMAN Linda reported me:DAT the :ACC success
b, Erna vermeldete den Erfolg.
ECREASER:reported the:ACC success
E37 a. Erna schüttete die Suppe in die Terrine.
ERMAN Linda poured the soup in the tureen
b. Erna verschüttete die Suppe.
ECREASER:poured the soup
‘Linda spilled the soup.’
Indirectus introduction:
Christian Lehmann, Situation types, valency frames and operations 28
In German, one of the functional variants of what is structurally preverbation with zu ‘to’ has
this function.
In the following series, the #a examples contain the base verb, whose valency
excludes an indirectus, while the #b examples show the derived verb, whose valency includes
an indirectus.
E38 a. dass Erna diese Ausgaben billigte
ERMAN that Linda these expenses approved
b. dass Erna mir diese Ausgaben zubilligte
that Linda me.DAT these expenses conceded
E39 a. dass Erna die Schilder ordnete
ERMAN that Linda the.PL tags ordered
b. dass Erna die Schilder den Gästen zuordnete
that Linda the.PL tags the:DAT.PL guests assigned
E40 a. dass Erna ‘Hallo’ rief
ERMAN that Linda hello shouted
b. dass Erna mir ‘Hallo’ zurief
that Linda me.DAT hello to:shouted
E41 a. dass Erna den Ball spielte
ERMAN that Linda the:ACC ball played
b. dass Erna mir den Ball zuspielte
that Linda me.DAT the.ACC ball to:played
E42 a. dass Erna arbeitete
ERMAN that Linda worked
b. dass Erna mir zuarbeitete
that Linda me.DAT to:worked
‘that Linda did preparatory work for me’
E43 a. dass Erna blinzelte/zwinkerte
ERMAN that Linda blinked/winked
b. dass Erna mir zublinzelte/zuzwinkerte
that Linda me.DAT winked_at
Indirectus introduction is not to be confused with the applicative derivation: the former
creates an actant position for an indirectus, which in several European languages including
German surfaces as an indirect object, while the latter creates an actant position for an
undergoer, which generally amounts to a direct object. The locus of the indirectus is in
trivalent verbs, as in E38.b – E41.b. In languages which have such a syntactic function and
mark it by some dative-like case, an indirectus may even be introduced on an intransitive
verb, as shown by E42 and E43. In languages lacking an indirect object, indirectus
introduction on intransitive verbs might reduce to an applicative derivation.
As implied by its definition (§3.2.2), the macro-role of the indirectus is less central in an
argument frame than the other two macro-roles, the actor and the undergoer. And since, in the
The same derivational process has several other functions, and even some verbs which show the
same valency change as those of the example series may be lexicalized in a completely different
meaning, like gestehen ‘confess’ vs. zugestehen ‘concede’.
Christian Lehmann, Situation types, valency frames and operations 29
prototypical case, it only appears if these two are already there, it generally corresponds to
argument #3 in a frame. A corollary of this is that syntactic functions subsumed under this
macro-role are either complements, but less central ones, or they are adjuncts, but the most
central ones. Since we are here dealing with valency alternations, the possibility of having
adjuncts in one of the indirectus functions (beneficiary or goal) is of little concern here.
Suffice it therefore to say that in many languages, the dative (or allative) used to mark the
indirect object also marks the beneficiary, as in E44f:
E44 Erna trug mir den Koffer.
ERMAN Linda carried me.DAT the:ACC suitcase
E45 Erna reparierte mir das Fahrrad.
ERMAN Linda repaired me.DAT the bike
While there is an indirect object in E35.b and E38.b – E43.b, there is none in E44f, as proved
by the usual tests for actancy: while the adjunct in the latter examples is easily replaced by
whatever construction yields the benefactive sense, the complement in the former examples is
in the form required by its verb. Moreover, the indirect object in most of E38.b E43.b is
3.2.7 The communication situation
The situation of communication has a complicated status in the set of situation types. On the
one hand, it is the one situation whose model is omnipresent in language: the speech situation
is, of course, the model of this situation type. One might therefore expect it to constitute a
basic trivalent situation type. However, as already anticipated, all trivalent situation types can
plausibly be generated by expansion of a bivalent situation. This is true for the
communication situation, too. And similarly to the manipulative situation type (§3.3.5), there
is more than one way of composing a situation of communication.
A situation of communication may be analyzed as shown in Table 9 (cf. Van Valin &
LaPolla 1997, ch.
Table 9 Situation of communication
constellation participant
properties roles example predicates
4) 1: human
2: human
3: [ling. object]
4: -
1: Ac
2: (Ad)
3: U.eff
4: (
say, tell
(1) 1: Ac talk, sing
2) 1: Ac
2: Ad
2: Com
speak to sb.
chat with sb.
3) 1: Ac
3: U.eff utter, say sth.
3) shout sth. at sb.; tell sb. sth., order,
Christian Lehmann, Situation types, valency frames and operations 30
4) talk with sb. about sth., ask sb. for sth.
(1, 2, 3, 4) involves the following participants: 1 is the active communicator, 2
(Ad = addressee, Com = comitative) is the interlocutor, 3 is the message uttered by 1, and 4 is
the topic of communication. 1 and 2 are prototypically human beings, 3 is a linguistic object
which may either be quoted or characterized, and 4 may be anything. Some of the semantic
roles follow from this constellation: 1 = Ac; 2 may be an Ad and, thus, an indirectus; 3 is an
effected undergoer; and 4 may get the undergoer macro-role if that has not yet been assigned.
All four participants are present in E46.
E46 Linda said nothing to Bill about the matter.
Now the various predicates of communication differ by the selection they make from among
this maximum scenario. Some important constellations are enumerated in Table 9 and
illustrated in the last column by English constructions instead of verbs from other languages
whose valency is actually confined to the respective frame. For instance, the verbs meaning
‘say’ in Hoocąk, Indonesian and Ojibwe are just bivalent, illustrating the frame
(1, 3). The basic verb rendering ‘talk’ instantiates the pattern C
(1, 4) in Ainu, Balinese, Chintang, and Sliammon, but the pattern C
(1, 2) in
Bezhta, among others.
Although most predications representing a situation of communication are only partial
renditions in that sense, they are often composed from even simpler predications. A few
examples from the wide cross-linguistic variation may be mentioned. In E47, the predication
(1, 2, 3) is conceptualized as C
(1, P
(2, 3)); thus, like the caused
experience of Table 6.
E47 sinrit oruspe an=e=nu-re na
INU ancestor story IND.A=2SG.O=hear-CAUS FIN
‘I will tell you the story of the ancestors’ (ValPal Database, Ainu, (101))
Similarly in E48, ‘tell sb. sth.’ is conceptualized as ‘give sb. sth. to know’:
E48 Erni kasi-tahu Tom rencana outing
NDONES Erni give-know Tom plan outing
‘Erni told Tom the plan for the outing’ (ValPal Database, Indonesian, (114))
Here the addressee is plainly conceived on the model of the recipient and, thus, an indirectus.
3.3 Diathetic operations
Diathetic operations change the functions of the dependents of the verb much like valency
operations (or semantic role operations) do. The difference is that valency operations affect
the semantic roles carried by these syntactic components, while diathetic operations only
change their information status. Their main purpose is to give a certain syntactic component
that syntactic function that best suits its information status.
3.3.1 The hierarchy of syntactic functions
The paradigmatic relation between two diatheses of a given predication is commonly
conceived in terms of operations of promotion and demotion. These refer to a hierarchy of
Christian Lehmann, Situation types, valency frames and operations 31
syntactic functions which is displayed in Diagram 2. They involve additional argument
functions beyond the actor, undergoer and indirectus which are the object of valency
operations. That is because one of these may be demoted to a lower position on the hierarchy,
or may be the goal of a promotion from a lower position on the hierarchy. This concerns,
specifically, local, beneficiary and similar adjuncts.
Diagram 2. Hierarchy of adverbal syntactic functions
subject absolutive
direct object | primary object ergative
indirect object | secondary object
other complement
The hierarchy of syntactic functions plays an important role in many fields of syntax. In
independent declarative sentences, it mainly reflects the thematicity of the nominal
expressions occupying its levels.
The principle is: the more thematic a verbal dependent is,
the higher the function assigned to it on Diagram 2.
Syntactic functions have little semantic import taken by themselves. The higher up a
syntactic function is in the syntactic function hierarchy, the emptier it is semantically.
particular, the subject function by itself in most languages including English does not code the
actor role, since there is a subject in the passive construction that is transparent to the
undergoer role. The opposition (paradigmatic contrast) among syntactic functions represented
by Diagram 2 pertains more to their discourse function. The little semantic potential that is
associated with syntactic functions stems from the fact that many verb roots pair the same
structural valency frame with the same semantic role frame, so that this set may serve as a
model exerting a certain analogical attraction. The semantic role potential of a syntactic
function may therefore remain latent and become relevant only in syntagmatic contrast.
For instance, although the subject by itself does not code actor function, in the transitivity
schema, its referent is ascribed the highest control in a situation (s. Hopper & Thompson
1980), since there the subject contrasts with the direct object. The latter’s semantic potential is
itself weak enough, but in the transitivity schema, the direct object is the undergoer, so that
the actor role remains for the subject. Similarly the indirect object in languages such as Latin
and German by itself means very little. Its semantic potential remains mostly latent even in
bivalent frames such as E49.a and E50.a.
E49 a. Erna folgte dem Einbrecher.
‘Linda followed the burglar.’
b. Erna verfolgte den Einbrecher.
‘Linda pursued the burglar.’
Roughly, a referent is more thematic the shorter the distance from its last mention and the denser the
frequency of its mentions in comparison with other referents in the preceding discourse (Givón (ed.)
Christian Lehmann, Situation types, valency frames and operations 32
E50 a. Erna folgte Erwins Rat.
‘Linda followed Irvin’s advice.’
b. Erna befolgte Erwins Rat.
‘Linda adhered to Irvin’s advice.’
However, in E49.b and E50.b, the dependent in question has been promoted to direct object
function by a derivational process that marks this promotion (s. §3.3.3). It is chiefly by the
paradigmatic contrast between transitive #b versions and the intransitive #a versions that we
perceive a stronger control cline in the former than in the latter, the agent being more
exclusively focused on the patient.
Just as in the case of valency operations (§3.2.1), the direction of an operation of
promotion or demotion is determined by markedness. Roughly, the variant that involves more
grammatical formatives is the derived one, and it constitutes the target of a promotive or
demotive operation. If no difference in markedness is to be discerned, then no directed
operation can be diagnosed.
E51 a. Linda outwitted Irvin.
b. Irvin was outwitted by Linda.
E52 a. Linda loaded the wagon with hay.
b. Linda loaded hay on the wagon.
Thus, in E51, the passive is clearly marked against the active by an additional auxiliary and an
additional preposition. Therefore the active is basic and the passive is derived; and
consequently we speak of passivization rather than of activization. Conversely in E52, no
difference between the two versions in terms of structural complexity can be discerned, and it
is therefore not possible to know which of them is basic.
In English, the passive promotes a nominal constituent from any position on Diagram 2
up to the highest level, at the same time demoting the nominal constituent that was there.
Since this is a bipartite operation, its function may be either to get the promoted referent into
the position that corresponds to its thematicity or else to move out of the thematic chain the
argument that would occupy the subject position in the active version. Similarly, the
antipassive promotes the actor to the highest position on Diagram 2 while at the same time
demoting the undergoer so that it gets out of the way. Eliminating a referent from the thematic
thread is a negative step with two facets: either that referent is not mentioned at all, or else it
is mentioned, but in the rhematic part of the sentence. In the first case, we have a passive or
antipassive construction with only the subject or absolutive, resp.; in the second case, the
passive actor or antipassive undergoer appear in an adjunct.
Similarly, applicativization promotes an argument to direct object position, thus allowing
it to function as secondary topic. If the base is already transitive, this entails demotion of the
less thematic noun phrase occupying the direct object position. Just as in the former case, this
The question of directionality poses itself in a different way for diathetic operations than for valency
operations as discussed in §3.2.1. Since the function of a diathetic operation is getting a certain
referent into a certain discourse position, semantic effects such as those mentioned in §3.2.1 cannot be
relied on as a criterion. There may still be distributional differences in the sense that one diathesis has
a restricted distribution and can be considered as less basic, as shown in §3.3.5.
Christian Lehmann, Situation types, valency frames and operations 33
is an ambivalent operation, as it may entail either omission of that argument or, on the
contrary, its appearance in an adjunct with rhematic function.
If an operation of demotion frees a position high on the syntactic function hierarchy, the
grammar of the language may require that position to be occupied, so that the demotion
triggers a promotion. That is true of the passive in most languages, where demotion of the
subject is accompanied by promotion of the erstwhile direct object to subject function.
Similarly, if an operation of promotion targets a position on the hierarchy already occupied by
another actant, it will normally oust the latter from that position so that it is demoted. That is,
for instance, true of the applicative, which may be applicable to transitive bases; but then the
erstwhile direct object has to vacate its position and take a lower one. From the point of view
of the result, such operations may appear to be operations of rearrangement, which make two
verbal dependents swap their places. We will assume, instead, that every diathetic operation
has either a promotive or a demotive function, and that any further rearragements are just side
effects conditioned by general constraints of the grammar. In what follows, only a few
diathetic operations will be briefly illustrated, for the sole purpose of delimiting valency
operations against them.
3.3.2 Antipassive
The antipassive is the diathetic counterpart to the valency operation of introversion. If the
direct object affected by the operation is not suppressed, but demoted, it is antipassivization
rather than introversion.
E53 a. ti he-v
ARIS tree chop-PRS
‘chop down a tree’
b. ti-m he-the-v
tree-DAT chop-INTR-PRS
‘chop on a tree’ (Foley 1986:109)
E53.b shows antipassivization, with accompanying demotion of the undergoer from direct
object to indirect object, indicating lack of a complete change of state.
3.3.3 Applicative
The applicative is the diathetic counterpart to the valency operation of extraversion.
Whenever the alternation between two syntactic constructions one of which comprises a
direct object which the other lacks does not affect the semantic roles and may instead be
described by a rule of syntax, it is applicativization rather than extraversion. In that case, any
direct object already present in the base version is not suppressed, but demoted. E54
illustrates the German be- applicative, E55 the Yucatec -t applicative.
Christian Lehmann, Situation types, valency frames and operations 34
E54 a. Erna denkt (an ihre Aufgabe).
ERMAN Linda thinks at her task
b. Erna bedenkt ihre Aufgabe.
Linda APPL:thinks her task
E55 a. h áalkab-nah-en t-u beel-il in kool
‘I ran on the way to my milpa (field)’
b. t-in wáalkab-t-ah u beel-il in kool
PRV-SBJ.1.SG run-TRR-CMPL POSS.3 way-REL POSS.1.SG milpa
‘I ran the way to my milpa’ (AVC_0003/4)
As anticipated in §3.1, a given structural process may have a purely diathetic function in some
cases, but may, in addition, change the semantic roles in other cases. Thus, the Yucatec -t
transitivization is applicativization in E55.b, but extraversion in E30f.
In some cases, the applicative and the causative are alternative means of expanding a
bivalent predicate into a trivalent one. This is shown in E56:
E56 a. anak=e luh ento ngisin-ang yeh ke lumure=e ento
ALINESE person=DEF female that fill-CAUS water to glass=DEF that
lit.: ‘the girl filled water into the glass’
b. anak=e luh ento ngisin-in lumure=e ento aji yeh
person=DEF female that fill-APPL glass=DEF that with water
‘the girl filled the glass with water’ (ValPal Database, Balinese, (75) and (44))
Visibly, the causative derivation of the base has the instrument of the manipulation predicate
(the liquid, in this case) in direct object function, while the applicative promotes the affected
object (the container) to direct object function.
3.3.4 Indirectus demotion
The same structural process that was seen in §3.2.6 to suppress the indirectus with some verbs
demotes it with others. This is shown in the series E57 – E59.
E57 a. dass Erna uns die Preise gab
ERMAN that Linda us the awards gave
b. dass Erna die Preise (an uns) vergab
that Linda the awards at us gave_away
E58 a. dass Erna uns die Briefe sandte
ERMAN that Linda us the.PL letters sent
b. dass Erna die Briefe (an uns) versandte
that Linda the.PL letters at us sent_out
E59 a. dass Erna uns das Buch lieh
ERMAN that Linda us the book lent
b. dass Erna das Buch (an uns) verlieh
that Linda the book at us lent_out
The indirect object is part of the valency of the simplex in the #a versions, but at most
adjoinable by a preposition in the derived #b versions. The prefix might be glossed as a
valency decreaser as before. However, it is not clear that the former indirectus falls out of
Christian Lehmann, Situation types, valency frames and operations 35
verbal valency, since the preposition an is not replaceable by any other one and consequently
appears to be valency-governed.
3.3.5 Locative shift
Manipulation (§3.2.3) is a particularly complex situation type. It involves three participants
whose mutual relationships allow for two alternate conceptions of this situation type:
1. M
(1, 2, 3) is an action-process in which 1 manipulates 2. The kind of
treatment is such that it necessarily involves another object 3 which 1 uses as an
instrument and which, by the manipulation, comes into contact with 2. For instance, ‘1
fills 2 with 3’ may be decomposed into s: C
(1, B
(2))) & U
(1, 3, s).
Consequently, 1 = Ac, 2 = U.aff and 3 = I.
2. M
(1, 2, 3) is a kind of transport in which 1 causes 3 to move to 2. This
conception is in consonance with the fact that 3 typically (though not in the case of ‘hit’
and ‘throw’) remains with 2. For instance, ‘1 fills 3 into 2’ (as in E56.a) may be
decomposed into C
(1, M
(3, 2) & B
(2))). Consequently, 1 = Ac, 2
= L.Goal and 3 = U.loc.
As may be seen, the two conceptions differ in the assignment of argument positions to
participants 2 and 3: In the first conception, 2 is U, whereas in the second conception, 3 is U.
In either case, U becomes direct object in English, whereas the other argument is demoted to
an adjunct function in which its particular role – instrument or goal, resp. – may be coded.
In the first conception, the use of an instrument is intrinsic in the concept of the treatment,
and often a verb of this semantic class involves a specific kind of instrument as a semantic
feature, like sprinkle involves some kind of liquid. On occasions when the nature of the
instrument used is not specified beyond what is implied by the lexical meaning, the
instrument need not be exteriorized (s. §2.1), so that a bivalent predication M
(1, 2)
results, like sprinkle the lawn. This is another example of selection among the participants
involved in a situation for representation in an argument frame. Thus, the double nature of
manipulation is the precondition for variation concerning predicates of manipulation both
across languages and within a language: On the one hand, such situations are converted, in
different languages, into bivalent predicates of different argument structure. On the other
hand, predicates of this class participate in peculiar valency alternations, among them the
English locative shift and the alternation, seen in E56 above, between a causative and an
applicative construction of the same base.
E60 and E61 illustrate locative shift in English and German:
E60 a. Linda stuffed the chicken with onions
b. Linda stuffed onions into the chicken
E61 a. Erna schmierte die Achse mit Fett
Linda smeared the axle with grease
b. Erna schmierte Fett an die Achse
Linda smeared grease at the axle
The argument frame of manipulation that allows this alternation was introduced in §3.2.3:
Apart from the actor, there is an affected undergoer which may alternatively be conceived as a
goal, and there is an instrument that may also be conceived as a locomoted undergoer.
Apparently, these two verbal dependents swap their syntactic functions. However, as said
Christian Lehmann, Situation types, valency frames and operations 36
above, the notion of an object that serves as an instrument in the manipulation is intrinsic in
the lexical meaning of such verbs. Consequently, it may easily be omitted, as in E62.a and
E63.a, while omitting the goal in the #b versions leads to unacceptibility.
E62 a. Linda stuffed the chicken
b. *Linda stuffed onions
E63 a. Erna schmierte die Achse.
‘Linda smeared the axle’
b. *Erna schmierte Fett
‘*Linda smeared grease’
This is, thus, not a symmetric alternation. Instead, it appears that the instrumental role of the
movable participant is basic, so it cannot be rendered as in the #b sentences. By this criterion,
the alternants of E60.a and E61.a, which code this participant as an instrumental argument,
are taken to be basic. The #b versions are derived by applicativization of that instrumental
argument. As a side effect, the affected undergoer is demoted and adjoined by a semantically
appropriate preposition.
3.3.6 Indirect participation
E64f (from ValPal Database, Eastern Armenian, (26), (25), (100), (97)) and E66 display an
alternation which has been dubbed ‘indirect participation’ in Lehmann et al. 2004.
E64 a. aȑǏik-ə makȹɾ-ecȹ seȑan-icȹ pȹoši-n
RMENIAN girl-DEF remove-AOR.3SG table-ABL dirt-DEF
‘the girl wiped the dirt from the table’
b. aȑǏik-ə makȹɾ-ecȹ seȑan-i pȹoši-n
girl-DEF remove-AOR.3SG table-GEN dirt-DEF
‘the girl wiped the dirt from the table’
E65 a. tȑa-n pȹajt-icȹ keȑev-ə klp-ecȹ
RMENIAN boy-DEF stick-ABL crust-DEF peel-AOR.3SG
‘the boy peeled the bark off the stick’
b. tȑa-n klp-ecȹ pȹajt-i keȑev-ə
boy-DEF peel-AOR.3SG stick-GEN crust-DEF
‘the boy peeled the bark off the stick’
E66 a. ts’a’-b nook’ ti’
YM give-CMPL.PASS dress LOC(3.SG)
‘he was given a dress’
b. ts’a’-b u nook’
give-CMPL.PASS POSS.3 dress
‘he was given a dress’ (HK’AN 0040.1)
By such omission tests, English verbs undergoing locative shift are assigned to different subclasses
in Goldberg 1995:176-178.
Christian Lehmann, Situation types, valency frames and operations 37
The alternation concerns situations with three participants, an actor, an undergoer and another
participant which may have any semantic role except undergoer. This third participant appears
in a local role in E64.a and E65.a and in recipient role in E66.a. The alternants of the #b
versions involve possessive constructions, with the participant in question in syntactic
possessor function. This construction is quite natural if the participant in question is, in fact,
the possessor of the undergoer. This semantic condition is not fulfilled in E64 and E65. In
E66, he will be the possessor of the transferred object. E66.b is nevertheless the version that
appears in the corpus. The alternation is peculiar in that a verbal construction alternates with a
nominal construction.
3.3.7 Valency and diathetic operations
Although valency operations and diathetic operations have little in common in functional
terms, they both affect the presence and syntactic function of nominal components of a clause.
The following parallelisms obtain:
Deagentivization suppresses the actor, and passivization demotes it so that it may as well
be absent. In an accusative language, both operations tend to entail promotion of the
undergoer to subject function. Similarly, introversion suppresses the undergoer, and
antipassivization demotes it so that it may as well be absent. In an ergative language, both
operations tend to entail promotion of the actor to absolutive function. Finally, both
extraversion and applicativization imply introducing a direct object, the sole difference being
that in applicativization, that argument is promoted to that position from a lower position on
Diagram 2, while in extraversion it comes out of the blue. Again, the latter criterion does not
establish a categorial distinction.
Because of this parallelism, many languages do not distinguish formally between
deagentivization and passivization, or between introversion and antipassivization, or between
extraversion and applicativization. For instance, a language may have a single operation of
promoting the direct object to subject function while omitting the basic subject, and
depending on contextual factors, the construction may sometimes have a passive reading and
sometimes a deagentive reading. That is, for instance, the case with the Latin passive. The
passive voice appearing in E67 is translated as deagentive. However, if an agent phrase like a
deo ‘by god’ were added, E67 could only be a passive construction.
E67 et verbum caro factum est
and word:NOM.SG flesh:NOM.SG make:PART.PERF:NOM.SG is
‘and the word became flesh’ (Angelus prayer)
By the same token, it is often methodologically not easy to tell valency operations and
diathetic operations apart. In particular, if a diathesis leaves the number of arguments
represented intact, this does not entail that it has a pure discourse function. For instance,
applicativization affords higher thematicity for the newly introduced direct object. However,
given the control cline regularly associated with the transitive construction, it may also be
relevant in applicativization, with the consequence that the argument in question is more
Christian Lehmann, Situation types, valency frames and operations 38
intensely affected by the situation. Then the operation is, at the same time, a semantic and a
diathetic operation.
4 Conclusion
1. The valency of a linguistic sign is the union set of the actant positions that it provides,
including the grammatical constraints associated with them.
2. The structural basis of valency is the necessity to provide structural relations among
components of a verbal construction. However, such relations may also be provided by
adjunction (adverbial modification).
3. Verbal valency has its functional basis in the argument frame rendering a situation core.
However, the argument frame of a predicate is not given a priori, but subject to
conceptual operations.
4. Participants that are part of the conceptual structure may not be assigned a semantic role
and, thus, not be coded.
5. Valency frames have their functional basis in recurrent types of situations. Elementary
situations are conceived holistically, and central semantic roles are based on such
elementary situations.
6. Valency frames are manipulated not only by valency operations, but also by diathetic
Chafe, Wallace L. 1970, Meaning and the structure of language. Chicago & London: Chicago
University Press.
Donohue, Mark 1999, Warembori. München & Newcastle: L
Faulhaber, Susen 2011, Verb valency patterns. A challenge for semantics-based accounts. Berlin:
Mouton de Gruyter (Topics in English Linguistics, 71).
Fillmore, Charles J. 1968, "The case for case." Bach, Emmon & Harms, Robert T. (eds.), Universals in
linguistic theory. New York etc.: Holt, Rinehart & Winston; 1-88.
Fillmore, Charles 2003, "Valency and semantic roles: the concept of deep structure case." Ágel,
Vilmos et al. (eds.), Dependenz und Valenz. Ein internationales Handbuch der zeitgenössischen
Forschung. Berlin: W. de Gruyter (Handbücher zur Sprach- und Kommunikationswissenschaft);
Foley, William A. 1986, The Papuan languages of New Guinea. Cambridge: Cambridge University
Press (Cambridge Language Surveys).
Foley, William A. & Van Valin, Robert D. 1984, Functional syntax and universal grammar.
Cambridge: Cambridge University Press (Cambridge Studies in Linguistics, 38).
Givón, Talmy (ed.) 1983, Topic continuity in discourse. A quantitative cross-language study.
Amsterdam & Philadelphia: J. Benjamins (Typological Studies in Language, 3).
Goldberg, Adele E. 1995, Constructions: A construction grammar approach to argument structure.
Chicago: University of Chicago Press.
In Yup’ik (Mithun 2000), a benefactive adjunct may be promoted to subject by causativization just
in order to guarantee subject continuity, with no causation being involved semantically.
Christian Lehmann, Situation types, valency frames and operations 39
Haspelmath, Martin 1993, "More on the typology of inchoative/causative verb alternations." Comrie,
Bernard & Polinsky, Maria (eds.), Causatives and transitivity. Amsterdam & Philadelphia: J.
Benjamins (Studies in Language Companion Series, 23); 87-120.
Hopper, Paul J. & Thompson, Sandra A. 1980, "Transitivity in grammar and discourse." Language
Jespersen, Otto 1924, The philosophy of grammar. London: G. Allen & Unwin; New York: H. Holt &
Kittilä, Seppo 2011, "Transitivity typology." Song, Jae Jung (ed.), The Oxford handbook of linguistic
typology. Oxford etc.: Oxford University Press; 346-367.
Kulikov, Leonid 2011, "Voice typology." Song, Jae Jung (ed.), The Oxford handbook of linguistic
typology. Oxford etc.: Oxford University Press; 368-398.
Lazard, Gilbert 1998, Actancy. Berlin & New York: Mouton de Gruyter (Empirical Approaches to
Language Typology, 19).
Lehmann, Christian 1991, "Predicate classes and P
". Seiler, H. & Premper, W. (eds.),
Partizipation. Das sprachliche Erfassen von Sachverhalten. Tübingen: G. Narr (LUS, 6); 183-239.
Lehmann, Christian 1992, "Valenz." Anschütz, Susanne R. (ed.), Texte, Sätze, Wörter und Moneme.
Festschrift für Klaus Heger zum 65. Geburtstag. Heidelberg: Heidelberger Orientverlag; 435-454.
Lehmann, Christian 2006, “Participant roles, thematic roles and syntactic relations.” Tsunoda, Tasaku
& Kageyama, Taro (eds.), Voice and grammatical relations. In honor of Masayoshi Shibatani.
Amsterdam & Philadelphia: J. Benjamins (Typological Studies in Language, 65); 167-190.
Lehmann, Christian 2012, "Converse categorization strategies." Koptjevskaja-Tamm, Maria &
Vanhove, Martine, New directions in lexical typology. Berlin: Mouton de Gruyter; 467-494.
Lehmann, Christian 2016, “Latin causativization in typological perspective”. Poccetti, Paolo (ed.),
Latinitatis rationes. Proceedings of the 17th International Colloquium on Latin Linguistics,
Rome, 2013. Berlin: W. de Gruyter; 917-941.
Lehmann, Christian & Shin, Yong-Min & Verhoeven, Elisabeth) 2004, Direkte und indirekte
Partizipation. Zur Typologie der sprachlichen Repräsentation konzeptueller Relationen. Erfurt:
Seminar für Sprachwissenschaft der Universität (ASSidUE, 13) (Second revised edition).
Lehmann, Christian & Verhoeven, Elisabeth 2006, "Extraversive transitivization in Yucatec Maya and
the nature of the applicative." Kulikov, Leonid & Malchukov, Andrej & de Swart, Peter (eds.),
Case, valency and transitivity. Amsterdam & Philadelphia: J. Benjamins (Studies in Language
Companion Series, 77); 465-493.
Levin, Beth 1993, English verb classes and alternations. Chicago: Chicago University Press.
Mithun, Marianne 2000, "Valency-changing derivation in Central Alaskan Yup'ik." Dixon, Robert
M.W. & Aikhenvald, Alexandra Y. (eds.) 2000, Changing valency. Case studies in transitivity.
Cambridge: Cambridge University Press; 84-114.
Peterson, David A. 2007, Applicative constructions. Oxford etc.: Oxford University Press (Oxford
Studies in Typology and Linguistic Theory).
Rappaport, Malka & Laughren, Mary & Levin, Beth 1993, "Levels of lexical representation."
Pustejovsky, James (ed.), Semantics and the lexicon. Dordrecht: Kluwer; 37-54.
Schank, Roger C. & Abelson, Robert P. 1977, Scripts, plans, goals and understanding. An inquiry into
human knowledge structures. Hillsdale, NJ: L. Erlbaum.
Van Valin, Jr., Robert D. & LaPolla, Randy 1997, Syntax. Structure, meaning, and function.
Cambridge: Cambridge University Press.
... It could be argued that causative meaning of the verb, i.e. the presence of [CAUSE] as a semantic feature of verb semantics, might be associated with a higher likelihood of a second argument genitive with the verbal noun (and with a high transitivity index as proposed below in Section 6, Table 8), since this condition applies to many of the verbs in Table 3 ( do·indnaig 'bestow', do·beir 'give', gaibid 'take', beirid 'bring', do·adbat 'show', con·oscaigi 'move'); see Lehmann (2015Lehmann ( : 1569Lehmann ( -1671 Table 6) on so-called agentive situation types. This semantic feature would in turn be related on the one hand to prevalent trivalent coding frames, and on the other hand to the possibility of anticausative uncoded alternations with non-finite forms, in the absence of voice ('the giving of something' = 'the fact that (someone) gives something' or 'that something is given'; 'someone's movement' = 'the fact that someone is moved' or 'moves'; see Lehmann 2015: 1571-1577 about the asymmetry of valency alternations with different types of transitive verbs). ...
... It could be argued that causative meaning of the verb, i.e. the presence of [CAUSE] as a semantic feature of verb semantics, might be associated with a higher likelihood of a second argument genitive with the verbal noun (and with a high transitivity index as proposed below in Section 6, Table 8), since this condition applies to many of the verbs in Table 3 ( do·indnaig 'bestow', do·beir 'give', gaibid 'take', beirid 'bring', do·adbat 'show', con·oscaigi 'move'); see Lehmann (2015Lehmann ( : 1569Lehmann ( -1671 Table 6) on so-called agentive situation types. This semantic feature would in turn be related on the one hand to prevalent trivalent coding frames, and on the other hand to the possibility of anticausative uncoded alternations with non-finite forms, in the absence of voice ('the giving of something' = 'the fact that (someone) gives something' or 'that something is given'; 'someone's movement' = 'the fact that someone is moved' or 'moves'; see Lehmann 2015: 1571-1577 about the asymmetry of valency alternations with different types of transitive verbs). However, the relationship with causativity is not straightforward, given that for·cain 'teach' should group with causatives while ad·cí 'see' should not. ...
... Examples for each type are given in (23) 39 What is meant here is the animacy hierarchy proper (Comrie 1989, ch. 9; see Dahl and Fraurud 1996), but, regarding the upper part of the hierarchy, it is further assumed that referents including the speaker and the addressee are high in animacy (as commonly held in the literature, see Lehmann 2015Lehmann : 1555, that día ('God'), not usually included in the hierarchy, ranks higher than humans, that referents of expressions such as mo chnamai 'my bones' in (24) are high in animacy. 40 Glossing Latin qui uos commonefaciat (1 Cor. ...
Full-text available
This paper compares argument marking of finite and non-finite forms (verbal nouns) of 26 Old Irish verbs, focusing on the relationship of transitive coding patterns of finite forms with the mapping of the argument in the genitive case with non-finite forms. The collection of argument structures is cast in the framework of the Leipzig Valency Patterns Project (Hartmann, Haspelmath, and Taylor 2013, Malchukov and Comrie 2015). The paper argues that, although most genitive arguments with transitive verbs express the microrole which corresponds to the second argument (the P-argument), this is not a strict rule, and some verbal nouns of transitive verbs clearly allow the first argument (the A-argument) to surface in the genitive. It is claimed that there is a correlation between the likelihood for finite forms to occur with an accusative argument and the likelihood for the genitive argument of non-finite forms to correspond to the P-argument. This likelihood is measured through a transitivity index that tries to supply a criterion that substitutes acceptability judgments, which are unavailable for past varieties. Each verb’s index can be calculated and each verb can consequently be accommodated in the resulting transitivity scale. Finally, the scale ranking Old Irish verbs according to the transitivity index is compared with crosslinguistic hierarchies of transitivity which have been put forward in the literature.
This paper surveys the valency-increasing constructions that apply to motion verbs in 49 genetically diverse languages. These include causative constructions, a variety of applicative constructions and the portative construction which is a valency-increasing construction that is lexically restricted to motion verbs. In the portative construction, the verb’s valency is increased by adding theme as a P-argument, but conceptually there is also an element of (co-motional) causation. However, the paper argues that the construction is neither causative nor applicative. Instead, the portative construction, which is frequent in the sample languages, should be considered a distinct type of valency-increasing construction.
This paper introduces the topic and the contributions of this special issue. While lexical restrictions are well-studied for grammatical relations defining argument coding (case marking and indexation), they are also common with voice and valency constructions, be they morphologically coded or not. The paper defines relevant terms and sketches the development of current usage-based approaches to lexical restrictions, in reaction to earlier lexicalist and constructional approaches. It then reviews existing studies of lexical restrictions on valency-preserving and valency-changing constructions, drawing connections with the other papers in this issue. In closing, it recommends further corpus-based cross-linguistic research of lexical restrictions.
Conference Paper
Full-text available
With the concept of iconicity and markedness, the paper try to explain the diversity of patterns of marking and alignment in some African languages with special reference to the grammatical relations. For typological comparison the important concepts like S, A, P and head/dependent marking, asymmetry of structure are discussed with some already known explanations for it, for example multi-layered structure, economy, frequency of use.
This paper presents a cross-linguistic investigation of the antipassive within the framework of Radical Construction Grammar. Based on function, this study identifies constructions in 70 languages from 25 language families and four geographical macro areas. Iconically motivated correlations were found between functions and the morphosyntactic strategies they employ. The results of this study suggest that constructions indicating the lower individuation of patients and constructions indicating the lower affectedness of patients, previously grouped together as ‘antipassive’, should be considered two separate construction types. This is based on their separate functions, the distinct morphosyntactic strategies used to encode them across languages, and differences in productivity with regard to semantic classes of verbs.
Full-text available
Predicate classes are semantic classes of lexemes that can function as the predicate of a clause. The are formed by the criteria relating to situation cores like dynamicity and telicity, relating to participant roles like degree of involvement, controll, affectedness and by participant properties like animacy. These structuring parameters are defined with examples from many languages. In the final part, numerous predicates are classified by these criteria, to serve as tertia comparationis in onomasiological descriptions of predicate classes of any language.
Full-text available
Semantic relations are analyzed at the levels of participant roles, thematic roles and syntactic relations. The first level serves as a tertium comparationis for linguistic strategies recognized at the two lower levels. A couple of functional subdomains of participation are analyzed, in-cluding causation, benefaction, external possession and concomitance. Lan-guages differ typologically by their choice as to which participant roles they express and which ones they leave to inference. In some participant constellations, the choice of a particular strategy is semantically motivated; in other constellations it is not so motivated, but may still be pursued by a language. A language may be characterized at the typo-logical level by the extension of a strategy over and above its proper functional domain.
The lexicon has typically been viewed as a mere list of lexical entries containing idiosyncratic information associated with individual words. Therefore, linguists have paid little attention to the internal structure of lexical entries. More recently, linguists have recognized a need to develop a more structured and elaborated lexical representation. Central to all current theories of the lexicon is a representation of the argument-taking properties of predicators, often referred to as a predicate-argument structure.1 Motivation for the development of theories of predicate-argument representation has come from two sources.
For both people and machines, each in their own way, there is a serious problem in common of making sense out of what they hear, see, or are told about the world. The conceptual apparatus necessary to perform even a partial feat of understanding is formidable and fascinating. Our analysis of this apparatus is what this book is about. —Roger C. Schank and Robert P. Abelson from the Introduction (