1
Editor's Introduction: Pragmatics in Optimality Theory
Reinhard Blutner and Henk Zeevat
Based on the tenets of the so-called `radical pragmatics' school (see, for instance,
Cole, 1981), this book takes a particular view with regard to the relationship between
content and linguistically encoded meaning. The traditional view embodied in the
work of Montague and Kaplan (e.g., Kaplan, 1979; Montague, 1970) sees content
being fully determined by linguistic meaning relative to a contextual index. In
contrast, the radical view takes it that, although linguistic meaning is clearly
important to content, it does not determine it, as pragmatic principles also play a role.
The central issue of this book is how to give a principled account of the
determination of content. Seeing linguistic meanings as underdetermining the content
(proposition) expressed, there must be a pragmatic mechanism of completion which
can be best represented as an optimization procedure. It is demonstrated that the
general framework of Optimality Theory (OT) makes it possible to formulate the
desired explanatory principles.
The first section of this general introduction outlines the basic framework of
OT as applied to phonology, syntax and morphology. The second section takes a
historical perspective and shows that the idea of optimization was present in the
pragmatic enterprise right from the beginning. Further, it explains the main
advantages of the general framework of OT when applied to the field of pragmatics,
and it puts the whole idea into concrete terms by demonstrating how Horn's (1984)
theory of conversational implicature can be implemented within a bidirectional
optimality theory. In section 3, we rise several basic questions underlying the whole
volume and discuss them from a theoretical and empirical perspective. This part
2
gives a overview of the different topics treated in the book, and it explains in which
respects the single contributions aim to satisfy our cooperative goal: to give a new
impulse to the tradition of radical pragmatics. Section 4, finally, outlines basic open
question of future research.
1 Optimality Theory
OT was initiated by Prince & Smolensky (1993) as a new phonological framework
that deals with the interaction of violable constraints. In recent years, OT was the
subject of lively interest also outside phonology. Students of morphology, syntax and
natural language interpretation became sensitive to the opportunities and challenges
of the new framework. The reasons for this growing interest in OT are empirical and
conceptual. First, it turned out that a series of empirical generalizations and observed
phenomena can be expressed very naturally within this framework; this holds
especially for phonology where in-depth analyses of many languages have provided
a much better insight into cross-linguistic tendencies than we had before the
invention of OT. Second, and perhaps much more important in linking scientists into
a new research paradigm, there are the conceptual reasons, which are many in the
present case: (i) the aim to decrease the gap between competence and performance,
(ii) interest in an architecture that is closer to neural networks than to the standard
symbolist architecture, (iii) the aim to overcome the gap between probabilistic
models of language and speech and the standard symbolic models, (iv) the problem
of learning hidden structure and the logical problem of language acquisition, (v) the
aim to integrate the synchronic with the diachronic view of language.
OT respects the generative legacy in two important methodological aspects:
the strong emphasis on formal precision in grammatical analysis and the goal of
3
restricting the descriptive power of linguistic theory. Seeing themselves within the
Generative tradition, many representatives of OT adopt the fundamental distinction
between Universal Grammar (UG) and a language-specific part of Grammar. UG
describes the innate knowledge of language that is shared by all normal humans, and
aims both to describe the universal properties of language and the range of variation
possible among languages. The language-specific part of grammar typically consists
of the lexicon and a system reflecting the specific structural properties of the
particular language. Within the generative tradition, the concrete theoretical
realization of this distinction has changed over the years. In the principles and
parameters model, for example, UG is conceptualized as a system of (inviolable)
principles, which are parameterized to demarcate the space of possible forms (see,
for instance, Chomsky, 1981). The fixation of these parameters (triggered by
language specific data) determines the grammar of the particular language. OT
realizes an essentially different view of this distinction.
At this point we must emphasize that optimality theory is rooted, at least in
part, in connectionism, a paradigm that makes use of neurobiological assumptions in
an extremely simplified way. As a consequence, OT does not assume a strict
distinction between representation and processing. More than ten years ago, there
was a lively debate in cognitive linguistics concerning the true architecture of
cognition ­ the debate between connectionists and symbolists. The proponents of a
symbolic architecture, among them Fodor and Polyshyn (e.g., Fodor & Pylyshyn,
1988), had the clever idea to take the arguments for connectionism as showing that
symbolic architecture is implemented in a certain kind of connectionist network. This
idea corresponds to the strategy of maintaining classical architecture and reducing
connectionism to an implementation issue. The development of OT demonstrates that
4
the opposite strategy is more exciting: augmenting and modifying symbolist
architecture by integrating insights from connectionism.
Leťs take a closer look now at the background and the nature of OT. Like
other models of grammars, OT sees a grammar as specifying a function that assigns
to each input (underlying representation of some kind) a structural description or
output. For example, in Grimshaw's and Samek-Lodovici's theory of the distribution
of clausal subjects (e.g., Grimshaw & Samek-Lodovici, 1995), an input is a lexical
head with a mapping of its argument structure into other lexical heads, plus a tense
specification. The input also specifies which arguments are foci and which
arguments are coreferent with the topic. An example is
(1) <sing(x), x=topic, x=he; T=pres perf>
It represents the predicate sing, with a pronominal argument that is the current
discourse topic. A possible output is an X-bar structure realizing an extended
projection of the lexical head. Examples are
(2) a.
b.
c.
[IP has [ sung]]
[IP hei has [ti sung]]
[IP has [ti sung] hei ]]
a clause with no subject
a clause with subject he, co-indexed with
a trace in SpecVP
he right-adjoined to VP, co-indexed with
a trace in SpecVP
The general idea of standard versions of generative syntax is to define the acceptable
(grammatical) input-output pairs via a system of rules and transformations. In order
to restrict the descriptive power of linguistic theory, constraints are added. All of
5
these constraints have been viewed as inviolable within the relevant domain. The
idea of inviolable constraints has itself proved to be problematic and this has led to
the "parameterization" of certain constraints, with one parametric setting for one
language and another parametric setting for another language.
In OT the "generative part" of the Grammar is reduced to a universal
function Gen that, given any input I, generates the set Gen(I) of candidate structural
descriptions for I. The central idea of OT is to give up the inviolability of constraints
and to consider a set Con of violable constraints. Furthermore, a strict ranking
relation >> is defined on Con. This relation makes it possible to evaluate the
candidate structural descriptions in terms of the totality of the violations they
commit, as determined by the ranking of the constraints. If one constraint C1
outranks certain constraints C2, ..., Ci , written C1 >> {C2, ..., Ci}, then one violation
of C1 counts more than as arbitrarily many violations of C2, ..., Ci. The evaluation
component selects the optimal (least offending, most harmonic) candidate(s) from
the set Gen(I). The grammar favors the competitor that best satisfies the constraints.
Only an optimal output is taken as an appropriate (grammatical) output; all
suboptimal outputs are taken as ungrammatical. This idea makes the grammaticality
of a linguistic object dependent on the existence of a competitor that better satisfies
the constraints.
Constraints are of two different kinds: markedness constraints that affect
outputs only and faithfulness constraints that relate to the similarity between input
and output. The main representatives of the faithfulness family are (i) PARSE
prohibiting underparsing ("underlying input material is parsed into output structure")
and (ii) FILL prohibiting overparsing ("the elements of the output must be linked
with correspondents in the underlying input"). In OT-syntax the latter constraint is
also called FULL-INT(ERPRETATION): the elements of the output must be interpreted.
6
Markedness constraints are inherently connected with the domain under discussion.
By way of example, we consider the following two constraints in the case of our OT
syntax (distribution of clausal subjects):
(3) a. SUBJ "the highest A-specifier in an extended projection must be
filled"1
b. DROP-TOPIC "arguments coreferent with the topic are structurally
unrealized"
To complete this short introduction to OT, leťs consider a typical OT tableau
relating to the input-output pairings (1) & (2). In the present example, the following
constraint hierarchy is assumed:
(4) FULL-INT >> DROP-TOPIC >> PARSE >> SUBJ
As can be seen from tableau (5), this ranking yields an "Italian" behavior in which
topicalized subjects are suppressed; this is exemplified by the optimal parse (5a).
(5)
<sing(x), x=topic, x=he; T=pres perf> FULL-INT DROP-TOPIC PARSE SUBJ
(a) [IP has [ sung]] * *
(b) [IP hei has [ti sung]] *
(c) [IP has [ti sung] hei ]] * *
This behavior would change to "English" if we would chose the following hierarchy:
(6) PARSE >> SUBJ >> FULL-INT >> DROP-TOPIC
7
where PARSE and SUBJ outrank DROP-TOPIC. In this case, (5b) would arise as the
optimal candidate.
The architecture of OT suggests a simple realization of the fundamental
distinction between UG on the one hand and the language specific part of Grammar
on the other hand: UG consists of Gen (the generator) and Con (the set of
constraints); the language-particular aspect of Grammar is determined by the
particular ranking of the constraints. This proposal bolsters the way for defining a
factorial typology:
Typology by reranking: Systematic crosslinguistic variation is due entirely to
variation in language-specific total rankings of the universal constraints in
Con. Analysis of the optimal forms arising from all possible total rankings of
Con gives the typology of possible human languages. UG may impose
restrictions on the possible rankings of Con. (Tesar & Smolensky, 2000, p.
27)
As already shown in (Prince & Smolensky, 1993), analysis of all rankings of the
constraints considered in the basic CV syllable theory reveals a typology that
explains Jacobson's (1962) typological generalizations. In the case of OT-syntax,
Grimshaw & Samek-Lodovici (1995) were the first who performed an analysis
involving all rankings of the above constraints and derived a typology of subject
distribution on this way.
Typology by reranking is the most famous but not the only pleasant
consequence from the general architecture of OT. Another consequence is the idea of
robust interpretive parsing, which is substantial for many purposes, such as
psycholinguistic applications of OT in describing online language production,
comprehension, and natural language acquisition.
8
Although the term parsing is used more commonly in the context of language
comprehension, in the OT literature it is treated as the general issue of assigning
structure to input, an issue relevant to both comprehension and production. To be
sure, the canonical perspective of an OT grammar is related to production ­ taking
the input as an underlying form, and the output structural description as including the
surface form. This type of parsing is called "productive parsing", and it is
schematically represented in diagram (7) ­ the term "overt structure" is used instead
of "surface structure":
(7) semantic form overt structure
productive parsing interpretive parsing
In the context of language comprehension, another mapping comes into play. It maps
a given overt form to an optimal structural description SD whose overt portion
matches the given form. The process of computing the optimal SD for an overt form
is called interpretive parsing.
It is a common observation that competent speakers can often construct an
interpretation for utterances they simultaneously judge to be ungrammatical.
Whereas it is notoriously difficult to account for this kind of "robustness" of natural
language interpretation within rule- or principle-based models of language, the
interpretation of ungrammatical sentences is much simpler when using an OT
architecture. Robust interpretive parsing is the idea of parsing an overt structure with
a grammar even when that structure is not grammatical according to that grammar. It
is important to recognize that the presence of interpretable but ungrammatical
sentences immediately corresponds to mismatches between productive and
structural
description
9
interpretive parsing. Consider an interpretive parse that starts with some overt
structure OS and assigns an optimal structural description SD. Paired with SD is a
certain semantic form SF. The grammaticality of SD (and its overt structure, OS)
depends on whether the outcome of productive parsing leads us back to SD, when
starting with SF. In case it does, then SD is grammatical; otherwise it is
ungrammatical.
As a simple illustration we reconsider the earlier example from OT syntax.
Leťs take the constraint hierarchy (4) that accounts for "Italian" syntactic behavior.
In Italian, sentences such as he has sung are unacceptable if the pronoun refers to a
discourse topic. Using the hierarchy (4), this is demonstrated in tableau (5), where
the sentence he has sung comes out as suboptimal. Despite its unacceptability, the
sentence is parsed into a structural description, namely [IP hei has [ti sung]]. An
important point in all examples of this kind is that both in productive parsing and in
interpretive parsing the same constraint hierarchies are used. The difference arises
solely from the different candidate sets that are relevant for the different perspectives
of optimization.
The idea of robust interpretive parsing is crucial for the mechanism of
language learning in OT when it is combined with another idea ­ the idea of
constraint demotion (cf. Tesar & Smolensky, 2000). The latter idea conforms to a
mechanism that reranks the constraints in a particular way, such that one prearranged
candidate becomes the winner over the rest of the candidates (cf. Vogel2
). The
combination of both ideas gives the following picture of children's language
acquisitions. Becoming confronted with some overt datum, the child tries to
understand this datum (on the basis of her current grammar). She performs
interpretive parsing, resulting in a structural description that includes an underlying
structure. Next, the child turns to the production perspective: she starts with the
10
underlying form and performs productive parsing. If the results of productive and
interpretive parsing are different, then this information is used to correct the
grammar. The child applies constraint demotion taking the interpretive parse to be
the winner (correct analysis) and the productive parse to be the loser. The child has
succeeded in learning the target grammar if interpretive and productive parsing
always give the same structural descriptions. Note that an overt form will allow the
learner to improve his grammar just in case the current grammar (incorrectly)
declares it to be ungrammatical.
There is an important consequence of this view of learning. The OT learning
algorithm establishes an interesting type of equilibrium: what we produce we are
able to understand adequately and what we understand we are able to produce
adequately. This equilibrium corresponds to strong conception of bidirectional
optimization: a logical combination of optimal comprehension and optimal
generation (cf. Blutner, 2000; Zeevat, 2000; Beaver & Lee). Hence, bidirectional
optimality can be seen as a kind of synchronic law describing the results of language
learning. It should be mentioned that Tesar & Smolensky's (2000) mechanism for
learning hidden structures is one aspect of language learning only. Acquiring
conventions that link structured forms and conceptual contents (via lexical entries
and idiom chunks) is another aspect. Most interestingly, empirical investigations
have shown that also in this case the general pattern of bidirectionality or symmetry
seems to apply.3
In the present volume, Jäger explores this possibility of
bidirectional learning within an evolutionary setting.
Before we apply OT to the domain of pragmatics we must clarify what the
general conditions are that every OT system has to satisfy. The following three
conditions are the core of OT. They are a necessary basis for the family of
procedures that performs grammar learning in OT (Tesar & Smolensky, 2000).
11
(A) Universal Grammar is assumed to be determined by a generative part Gen and a
system of violable constraints Con (UG = Gen + Con). The language-specific
part of Grammar relates to a particular ranking of the constraints in Con. Only
this part of the Grammar is learnable. Language learning simply reduces to
inferring the ranking of the constraints in Con. This excludes both the possibility
that the constraints themselves are learned (in part at least) or that aspects of the
generator are learnable. On the other side it excludes the possibility that the set
of the possible rankings is constrained on a universal basis.
(B) The force of strict domination >>: A relation of the form C >> C' does not
merely mean that the cost of violating C is higher than that of violating C';
rather, it means that no number of C' violations is worth a single C violation.
The force of strict domination excludes cumulative effects where many
violations of lower ranked constraints may overpower higher ranked constraints.
(C) The OT grammar of the language that has to be learned is based on a total
ranking of all the constraints: C1 >> C2 >> ... >> Cn . This condition is crucial
for the convergence of the proposed learning mechanism (Tesar & Smolensky,
2000). It can be shown that the iterative procedure of constraint demotion
converges to a set of totally ranked constraint hierarchies in this case, each of
them accounting for the learning data.
What is the status of these conditions? One way to look at these conditions is to see
them as oversimplifications that are made mainly for didactic and practical reasons.
Oversimplifications may be needed to allow one to concentrate on a central problem
12
and to sweep aside many problems that are less critical for understanding the central
one (i.e., the problem of learning `hidden' structure.) Moreover, oversimplifications
may be necessary to achieve interesting mathematical results that simply are not
possible without them. But it is not necessary to see them as simplifications, we can
also see them as conditions reflecting the true nature of the domain under discussion
and are taken to be empirically justified.
It must be admitted that it is not always simple to find out which position
really is taken by the representatives of OT. For example, concerning the condition
(C), we find the following statement in Tesar & Smolensky (2000):
From the learnability perspective, the formal results given for Constraint
Demotion depend critically on the assumption that the target language is
given by a totally ranked hierarchy. This is a consequence of a principle
implicit in Constraint Demotion. This principle states that the learner should
assume that the description is optimal for the corresponding input, and that it
is the only optimal description. This principle resembles other proposed
learning principles, such as Clark's Principle of Contrast and Wexler's
Uniqueness Principle. (p. 47 ff)
It appears likely to us that learning languages that do not derive from a totally
ranked hierarchy is in general much more difficult than the totally ranked
case. If this is indeed true, demands of learnability could ultimately explain a
fundamental principle of OT: UG admits only (adult) grammars defined by
totally ranked hierarchies. (p. 50)
Taking condition (C) as a kind of principle that indicates when language learning is
simple, however, is a different idea than taking it as a strict demand on theories of
13
learning. In our opinion, the first idea is right and the second wrong. There are
many examples where the target language produces synonymies (scrambling data in
German an Korean may provide a case in point). We agree that this can delay
learning in one case or the other. In this vein, the suggestion is to take (C) as a kind
of oversimplification, the acceptance of which is justified only for doing the first
significant research steps. Notably, Anttila & Fong (2000) take a similar view (cf.
also Beaver & Lee). As a consequence, the condition (C) should be given up in an
advanced stage and a more general theory should be developed, a theory that
explains (C) as a principle about the complexity of language learning. In our opinion,
Paul Boersma's learning theory (Boersma, 1998; Boersma & Hayes, 2001) is on the
right track for doing this job.
With regard to the condition (B), Smolensky himself sees it as a
"regimentation and pushing to extremes of the basic notion of Harmonic Grammar"
(Prince & Smolensky, 1993, p.200). And Gibson & Broihier (1998) argue that this
restriction does not appropriately characterize the manner in which parsing
preferences interact.
What about condition (A)? Many representatives of OT seem to consider it
as a conditio sine qua non. Boersma's work on functional phonology (Boersma,
1998), however, puts forward convincing arguments exposing principle (A) likewise
as a kind of oversimplification.
These questions about the status of the conditions (A)-(C) becomes highly
relevant when we try to extend the domain of applications for OT, especially when
we try to apply the OT framework to the domain of pragmatics. Hence, for
pragmatics in OT debating and clarifying the status of the condition (A)-(C) is an
opportunity and challenge. Most papers in this volume are directly or indirectly
concerned with this task.
14
2 Pragmatics in OT
The idea of optimization was present in the pragmatic enterprise from the very
beginning. Much more than in other linguistic fields optimality scenarios are present
in most lines of thinking: Zipf's (1949) balancing between effect and effort, the
Gricean conversational maxims (Grice, 1975, 1989), Ducroťs argumentative view of
language use (e.g., Ducrot, 1980), the principle of optimal relevance in Relevance
theory (Sperber & Wilson, 1986). However, in the course of the development of OT,
the area of OT semantics and pragmatics was developed after everything else. This
appears rather puzzling, and the reasons for that are not very clear. There may be
stylistic aspects that may frighten a serious semanticist or logician: the curious tables
with shadows, and the famous little hands. A more serious reason may have to do
with an unfortunate `dynamic turn' which was directed against Kamp's (1981)
programmatic outline of a cognitively oriented approach to language. In contrast to
Kamp's original paper, which is based on the tenets of `radical pragmatics', much
research which falls under the rubric of the `dynamic turn' is in the spirit of the
conservative view of language which radical pragmatics set itself against.
While the compositionality assumption underlying the `dynamic turn' has
strengthened the methodology of semantics, it has also led to a mechanistic approach
at points where pragmatics and semantics are difficult to keep apart. The habit of
interpreting trees with fully resolved pronouns fails to make the distinction between
rule-based grammar and the complicated salience weighting of different antecedents
required for pronoun resolution, a process that leads to preferences at best. In
treatments of presupposition, sometimes presupposition boils down to a single
logical operator and so obscures the distinction between the semantic role of
15
presuppositions for their triggers and their function as a sign that the speaker is
making an assumption, a distinction that also shows up as the distinction between
treating a presupposition by resolving it to the context or by accommodating it. We
would submit that an OT theory has an advantage over logical or grammatical
treatments in that the ideal of rational cooperative communication can be almost
directly captured by constraints that directly derive from Grice's analysis of this
cooperative behavior (cf. van Rooy, Zeevat)
What can be called with more justice `radical pragmatics' (cf. Cole, 1981) is
to hypothesize a division of labor between (i) a linguistic system determining the
semantic representation of a sentence (Grammar including the lexicon) and (ii) a
pragmatic system constituting the interpretation of the corresponding utterance in a
given setting (contextual information, encyclopedia). The pragmatic system is taken
as realizing Grice's (1975) idea of conversational implicature, and it is modeled with
the instruments of OT. As a consequence, many linguistic phenomena which had
previously been viewed as belonging to the semantic subsystem, in fact can be
explained within the pragmatic subsystem of OT.
Before we enter the discussion in which way optimality theory may help to
close the gap between formal (linguistic) meaning and interpretation, we have to
consider this distinction more closely. For Grice (1975) the theoretical distinction
between what the speaker explicitly said and what he has merely implicated is of
particular importance. What has been said is supposed to be based purely on the
conventional meaning of a sentence, and is the subject of semantics. What is
implicitly conveyed (scalar and conversational implicatures) belongs to the realm of
pragmatics. It is assumed to be calculable on the basis of the setting ­ a notion
already introduced by Katz & Fodor (1963), and referring to previous discourse,
socio-physical factors and any other use of "non-linguistic" knowledge. Fruitful as
16
this theoretical division of labor may have been ­ especially as a demarcation of the
task of logical semantics ­, it has inherent problems. More often than not, what is
said by a speaker's use of a sentence already depends on the context. Even for
Griceans, propositional content is not fully fleshed out until reference, tense, and
other indexical elements are fixed. However, propositional content must be inferred
in many cases ­ going beyond the simple mechanism of fixing indexical elements.
Proponents of relevance theory (see, for example, Carston, 2002; Carston,
2003a, 2003b; Sperber & Wilson, 1986) have pointed out that the pragmatic
reasoning used to compute implicated meaning must also be invoked to fill out
underspecified propositions where the formal meaning contributed by the linguistic
expression itself is insufficient to give a proper account of truth-conditional content.
A similar point was made in lexical pragmatics (e.g., Blutner, 1998, 2002). Both
relevance theory and lexical pragmatics agree in assuming a Gricean mechanism of
pragmatic strengthening in order to fill the gap between formal, linguistic meaning
and the propositional content (i.e., the explicit assumptions communicated by an
utterance ­ called explicature in relevance theory; cf. Sperber & Wilson, 1981, p.
182).
In a similar vein, de Hoop & de Swart (2000) and Hendriks & de Hoop
(2001) argue with regard to the theory of interpretation, what compositional
semantics gives us is a radically underspecied notion of meaning represented by a
possibly infinite set of interpretations of a well-formed syntactic structure. In
addition, these authors were the first who propose to use the framework of optimality
theory in order to select the optimal interpretation associated with a particular
syntactic structure. For that purpose, they propose a particular set of constraints and a
rankings between those constraints, based on general principles of rational
communication. The interpretive perspective on optimization provides insights into
17
different phenomena of interpretation, such as the determination of quantificational
structure and domain restriction (Hendriks & de Hoop, 2001), nominal and temporal
anaphora (de Hoop & de Swart, 2000), and the interpretational effects of scrambling
(de Hoop, 2000).
Stimulated by Horn's (1984) theory of conversational implicature and related
ideas in relevance theory, Blutner (2000) argued that this design of OT is
inappropriate and too weak in a number of cases. This is due to the fact that the
abstract generative mechanism (Gen) can pair different forms with one and the same
interpretation. The existence of such alternative forms may lead to blocking effects
which strongly affect what is selected as the preferred interpretation. The
phenomenon of blocking has been demonstrated in a number of examples where the
appropriate use of a given expression formed by a relatively productive process is
restricted by the existence of a more "lexicalized" alternative to this expression. One
case in point was provided by Householder (1971). The adjective pale can be
combined with a great many color words: pale green, pale blue, pale yellow.
However, the combination pale red is limited in a way that the other combinations
are not. For some speakers pale red is simply anomalous, and for others it picks up
whatever part of the pale domain of red pink has not preempted. This suggests that
the combinability of pale is fully or partially blocked by the lexical alternative pink.
The phenomenon of blocking requires us to take into consideration what else
the speaker could have said. As a consequence, we have to go from a onedimensional,
to a two-dimensional (bidirectional) search for optimality.4
As
mentioned in section 1, bidirectional optimality can be seen as describing the
equilibrium that results from language learning in the limit.
In the domain of pragmatics, the bidirectional view was independently
motivated by a reduction of Grice's maxims of conversation to two principles: the Q-
18
principle and the I-principle (Atlas & Levinson, 1981; Horn, 1984, who writes R
instead of I). The I/R-principle can be seen as the force of unification minimizing the
Speaker's effort, and the Q-principle can be seen as the force of diversification
minimizing the Hearer`s effort (cf. Horn 1984). The Q-principle corresponds to the
first part of Grice's quantity maxim (make your contribution as informative as
required), while it can be argued that the countervailing I/R-principle collects the
second part of the quantity maxim (do not make your contribution more informative
than is required), the maxim of relation and possibly all the manner maxims.
Conversational implicatures which are derivable essentially by appeal to the Qprinciple
are called Q-based implicatures. Standard examples are scalar implicatures
and clausal implicatures. I-based implicatures, derivable essentially by appeal to the
I-principle, can be generally characterized as enriching what is said via inference to a
rich, stereotypical interpretation (cf. Atlas & Levinson, 1981; Gazdar, 1979; Horn,
1984; Levinson, 2000)
In a slightly different formulation, the I/R-principle seeks to select the most
coherent interpretation, and the Q-principle acts as a blocking mechanism which
blocks all the outputs which can be grasped more economically by an alternative
linguistic input (Blutner 1998). This formulation makes it quite clear that the
Gricean framework can be conceived of as a bidirectional optimality framework
which integrates expressive and interpretive optimality. Whereas the I/R-principle
compares different possible interpretations for the same syntactic expression, the Qprinciple
compares different possible syntactic expressions that the speaker could
have used to communicate the same meaning. The important feature of this
formulation within bidirectional OT is that although it compares alternative syntactic
inputs with one another, it still helps to select the optimal meaning among the various
19
possible interpretational outputs of the single actual syntactic input given, by acting
as a blocking mechanism.
The so-called strong version of bidirectional OT ­ it conforms to the
equilibrium established during OT learning ­ can be formulated as given in (8). Here,
pairs (f, m) of possible (syntactic) forms f and utterance meanings (= interpretations)
m are related by means of an ordering relation <, being less costly (more harmonic).
At the moment, the precise metric underlying this ordering relation is still open, and
the sign < is not much more than a place holder for such a metric. In OT, the
ordering relation < can be constituted by a system of ranked constraints, as discussed
in many contributions to this volume. Another option would be to work with a single,
graded markedness constraint such as RELEVANCE (see Van Rooy).
(8) Bidirectional OT (Strong Version)
A form-meaning pair (f, m) is optimal iff it is realized by Gen and it satisfies
both the I- and the Q-principle, where:
(a) (f, m) satisfies the I-principle iff there is no other pair (f, m' ) realized
by Gen such that (f, m') < (f, m)
(b) (f, m) satisfies the Q-principle iff there is no other pair (f', m) realized
by Gen such that (f', m) < (f, m)
It should be mentioned that the I-principle is very much in line with the monodirectional
view on optimality theoretic interpretation as proposed by de Hoop & de
Swart (2000) and Hendriks & de Hoop (2001), which exclusively, adopts the hearer's
20
perspective on disambiguation. What is interesting in (8) is that it also implements
the Q-principle, thereby also taking the speaker's perspective into account. Hence, a
proper treatment of interpretation in OT has to take into account both the perspective
of the hearer and the perspective of the speaker. Because this framework of
bidirectional OT can be characterized in game-theoretical terms (Dekker & van
Rooy, 2000), optimality theoretic pragmatics can be given a proper formal
interpretation.
One of the main advantages of the optimality theoretic framework is that it
allows to isolate three substantial components of the overall mechanism: (i) the
generator, which provides the potential form interpretation pairs, (ii) the underlying
metric, possibly constituted by a system of ranked constraints, and (iii ) the two
perspectives of optimization. In relevance theory is it relevance that constitutes the
underlying metric; in other frameworks notions of information, efficiency, and
salience are more important (cf. van Rooy).
There are however several old problems with assuming full symmetric
bidirectionality to phonological and syntactic processing in both directions. In
phonology, the problem is mostly discussed as the Rad/Rat problem (cf. Hale &
Reiss, 1998). The German word Rat (council) is pronounced as [rat] without any
change from the underlying form to the surface form. The word Rad (wheel) is
pronounced in the same way but here two constraints come into play: the DEVOICING
constraint that prefers the pronunciation [rat] to [rad] and FAITHFULNESS that would
prefer the pronunciation [rad] and that is outranked by DEVOICING in German. If we
want to apply the same constraints in the direction from pronunciation to optimal
underlying form, Rat is always preferred because of FAITHFULNESS in interpretation.
The same problem can arise in syntactic ambiguities. Again in German, the sentence
Welches Mädchen mag Reinhard? is ambiguous between Which girl likes Reinhard?
21
and Which girl does Reinhard like? The wh-object has a longer road to go from its
canonical position to its sentence initial position than the corresponding wh-subject.
The constraint STAY (= Do not move) adopted by most OT syntacticians then prefers
the reading with the wh-subject. Since there is general agreement that there is a
proper ambiguity in these cases, full bidirectionality needs to be restricted by some
principle which makes the system less symmetric than the Tesar & Smolensky
learning algorithm assumes. In this volume, Jäger uses an asymmetric bidirectional
system for his learning algorithm, Vogel restricts his OT-syntax by powerful
pragmatic principles and Beaver & Lee consider different ways to avoid the Rat/Rad
problem in their survey of bidirectionality.
Another problem has to do with the specific features of blocking we find in
natural languages. The scenario of strong bidirection describes the case of total
blocking where some forms (e.g. *furiosity, *fallacity) do not exist because others do
(fury, fallacy). However, blocking is not always total but may be partial, in that
only those interpretations of a form are ruled out that are pre-empted by a "cheaper"
competing form. McCawley (1978) collects a number of examples demonstrating the
phenomenon of partial blocking. For example, he observes that the distribution of
productive causatives (in English, Japanese, German, and other languages) is
restricted by the existence of a corresponding lexical causative.
(9) (a) Black Bart killed the sheriff
(b) Black Bart caused the sheriff to die
Whereas lexical causatives ­ e.g. (9a) ­ tend to be restricted in their distribution to
the stereotypic causative situation (direct, unmediated causation through physical
action), productive (periphrastic) causatives tend to pick up more marked situations
of mediated, indirect causation. For example, (9b) could be used appropriately when
22
Black Bart caused the sheriff's gun to backfire by stuffing it with cotton. The general
tendency of partial blocking seems to be that "unmarked forms tend to be used for
unmarked situations and marked forms for marked situations" (Horn 1984, p. 26) ­ a
tendency that Horn terms the division of pragmatic labor.
There are two principal possibilities for avoiding the fatal consequences of
total blocking that are described by strong bidirection. The first possibility is to make
some stipulations concerning Gen in order to exclude equivalent semantic forms.
The second possibility is to weaken the notion of (strong) optimality in a way that
allows us to derive Horn's division of pragmatic labor in a principled way by means
of a sophisticated optimization procedure.
In Blutner (1998; 2000) it is argued that the second option is much more
practicable and theoretically interesting. A recursive variant of bidirectional
optimization was proposed (called weak bidirection) which was subsequently
simplified by Jäger (2002):
(10) Bidirectional OT (Weak Version)
A form-meaning pair (f, m) is called super-optimal iff (f, m)  Gen and
(a) there is no other super-optimal pair (f, m') : (f, m') < (f, m)
(b) there is no other super-optimal pair (f', m) : (f', m) < (f, m)
Under the assumption that < is transitive and well-founded, Jäger (2002) proved that
(10) is a sound recursive definition and is equivalent with the formulation in Blutner
(1998; 2000). In addition, he proved that each pair which is optimal (strong
23
bidirection) is super-optimal (weak bidirection) as well, but not vice versa. Hence,
weak bidirection gives us a chance to find additional super-optimal solutions. For
example, weak bidirection allows marked expressions to have an optimal
interpretation, although both the expression and the situations they describe have a
more efficient counterpart. Hence, this formulation is able to describe Horn's
division of pragmatic labor. The notion of weak bidirection is discussed more
detailed by Mattausch (this volume, section 3.2).
The existence of two notions of bidirectionality raises a conceptual problem:
Which conception of bidirectionality is valid, the strong or the weak one? Obviously,
this question relates to the foundation of bidirection in an overall framework of
cognitive theory. As we have seen already, the strong mode of optimisation in (8) what
we produce we are able to understand adequately and what we understand we
are able to produce adequately ­ corresponds to the equilibrium established by the
OT learning algorithm. Hence, the strong conception of bidirectionality can be seen
as a kind of synchronic law describing the results of language learning.
Weak bidirection gives a chance to find additional solutions. Is it possible to give
a natural interpretation for these additional solutions? We want to propose the idea
that these additional solutions are due to the ability and flexibility of selforganization
in language change which the weak formulation alluded to. In other
words, we propose to take these additional solutions as describing the possible
outcomes of self-organization before the learning mechanism has fully realized the
equilibrium between productive and interpretive optimisation.
Jäger (2002) and Dekker & Van Rooy (2000) have proposed algorithms that
update the ordering (preference) relation < such that (i) optimal pairs are preserved
and (ii) a new optimal pair is produced if and only if the same pair was super-optimal
at earlier stages. Consequently, we can take the solutions of weak bidirection to be
24
identical with the solutions of strong bidirection considering all the systems that
result from updating the ordering relation. Recently, Van Rooy (forthcoming) and
Jäger (this volume) have reconsidered this problem and have proposed algorithms
within an evolutionary setting ­ realizing a mechanism of self-organization in
language change. This point may be clarified when we (re)consider Horn's division
of pragmatic labor and relate it to the principle of constructional iconicity in the
school of "natural morphology" (for references cf. Wurzel, 1998)
Constructional iconicity: A semantically more complex, derived morphological
form is unmarked regarding constructional iconicity, if it is symbolized formally
more costly than its semantically less complex base form; it is the more marked,
the stronger its symbolization deviates from this. (Wurzel, 1998, p. 68).
In this school the principle plays an important role in describing the direction of
language change. In fact, constructional iconicity and Horn's division of pragmatic
labor can be proven to be a consequence of weak bidirection. This observation gives
substance to the claim that weak bidirection can be considered as a principle
describing (in part) the direction of language change: super-optimal pairs are
tentatively realized in language change. This relates to the view of Horn (1984) who
considers the Q principle and the I principle as diametrically opposed forces in
inference strategies of language change. Of course, the idea goes back to (Zipf,
1949), and was reconsidered in van Rooy (forthcoming).5
Arguing that Horn's
division of pragmatic labor is a conventional fact about language, this convention can
be explained in terms of equilibriums of signaling games introduced by Lewis (1969)
­ making use of an evolutionary setting (cf. van Rooy, forthcoming).
But is it really the case that weak bidirectionality does not play a role in
synchrony? The Horn example is the pair Black Bart shot the sheriff/ Black Bart
25
caused the sheriff to die. A similar example is Grice's Mrs. T produced a series of
sounds closely resembling the score of "Home Sweet Home", which contrasts with:
Mrs. T. sang "Home Sweet Home". Horn's and Grice's point is that the long and
unusual form are used to convey that there was something special with the killing
and the singing and that this is not accidental. The process by which this special
interpretation is arrived at cannot be diachronic language change: the long and
unusual forms are so unusual that it is not possible to assume a special
conventionalisation process that associates the special meaning with the special form.
Grice's explanation from his maxim Be Brief can almost directly be translated
in OT pragmatics. The relevant constraint is ECONOMY, which we can reinterpret as
the requirement that there is no correct form interpretation pair that is more
economical (or more standard?) in either dimension. This immediately leads us to
reject the association of the complex (unusual) forms with the standard meaning: for
that we have a simpler and more usual form. It likewise rules out the association of
the simple form with the non-standard meaning. The result is that we obtain an
underspecified special meaning for the special forms which must be interpreted
further with respect to the context and the situation to give us the concrete
interpretations (kill in a bizarre way, sing rather badly) that we seem to obtain.
Notice however that the speaker has not said any of this, she has merely suggested
that there is something special going on. There is no convention that fixes the
meaning. The vagueness and cancellability of the extra interpretation suggests that
we are dealing with an implicature and not with part of the truth-conditional content.
There are three points to be made about this reinterpretation of Grice's
stylistic maxim. In the first place, it is a very low constraint which can be overridden
by any grammatical or semantical constraint that one needs to assume. It is the
lowest of the low. Second, it is obviously weakly bidirectional for it to work. If the
26
standard-form/marked-meaning or the marked-form/standard-meaning were in
competition with marked-form/marked-meaning, that last pair would not survive.
And third, it seems that ­ with some charity ­ all other pragmatic principles can be
related to it. As Blutner & Jaeger show, the constraint DO NOT ACCOMMODATE can be
seen as a special case of semantic economy, minimizing the number of new
discourse referents. The constraint RELEVANCE can also be seen as a kind of
semantic economy: irrelevant information is information that the interlocutor is not
seeking for and requires the accommodation of new questions or interests of the
interlocutor. Information that is consistent or consistent with the context is
pragmatically less complex than information that is inconsistent in itself or
inconsistent with the context. The whole of pragmatics would be weakly
bidirectional under this interpretation.
If this would be the case, it would also give us an indication of why weak
bidirectionality is such as a powerful explanatory principle in diachronic linguistics.
Pragmatic weak bidirectionality creates special interpretations that can become
conventionalized. Assume that a marked form is used with some frequency to
indicate the same marked meaning. It will then become a conventional device to
indicate the marked meaning and the marked meaning will no longer be derived by
weak bidirectionality but by a lexical or grammatical convention. Think about
Hebrew optional object case marking conventionally meaning that the referent is
definite. Or about the Dutch wijf ­ originally the standard word for woman, but
pushed away by vrouw (originally mistress) ­ that can now only be used for the
purpose of expressing contempt for the referent in question.
Summarizing, we suggest to take the strong conception of bidirectionality as
a synchronic law and the weak one as conforming to diachrony (with the reservation
and clarification just sketched). In addition, the present conception conforms to the
27
idea that synchronic structure is significantly informed by diachronic forces. Further,
it respects Zeevaťs (2000) acute criticism against super-optimality as describing an
online mechanism (see also Beaver & Lee). From the perspective of
grammaticalization, we are very close to Hyman's (1984) dictum of seeing
grammaticalization as the harnessing of pragmatics by a grammar. And there are
connections to a recent proposals by Haspelmath (1999) for an OT-based theory of
language change.
3 Overview
The aim of this book is to demonstrate that OT also finds fruitful applications in the
domain of pragmatics and can contribute in overcoming the gap between linguistic
meaning and utterance meaning. This section contains an overview of the different
topics treated in the book and it explains in which respects the single contributions
aim to satisfy our cooperative goal: giving the tradition of radical pragmatics a new
impulse.
The promise of OT pragmatics is that by using the OT architecture some
order can be brought to the seemingly unrelated approaches that constitute
pragmatics. There have been a series of studies that try to reformulate treatments of
pragmatic phenomena to optimality theory. De Hoop & De Swart (2000) study the
determination of quantifier restrictions, a classical challenge to compositional
semantics, since that determination is only partially determined by the syntactic tree,
and can involve interactions with the context, the information structure and the linear
order of the quantifier. One of the factors in the solution is relating the interpretation
to given material, either in the topic or in the context. This problem area comes back
in studying pronoun syntax and resolution (Beaver, to appear; Bresnan, 2001),
28
presupposition (Jäger & Blutner, 2000; Zeevat, 2000), the binding theory (Burzio,
1991, 1998; Levinson, 1987, 2000). Other areas of pragmatics where OT has been
attempted are intonation and information structure (Beaver & Clark, 2002;
Schwarzschild, 1999), scalar implicatures (Blutner, 2000; Van Rooy, 2001).
In the present volume, Helen de Hoop provides an in depth discussion based
on real data of the Complementary Preference Hypothesis as an account of stressed
pronouns in English and formulates an alternative account in terms of two
interpretive constraints: Contrastive Stress and Continuing Topic to overcome the
problems with the earlier account.
Petra Hendriks combines a semantic analysis of only (only(A)(B) =
all(B)(A)) with an OT account of how intonation and syntax conspire in determining
the scope and restrictor of determines and and focus-sensitive particles. The account
builds on earlier work of De Hoop & De Swart (2000) and Hendriks & De Hoop
(2001) using optimality theoretic semantics.
Jason Mattausch introduces the influential work of Levinson on the origin
and typology of binding theory and reformulates the different historical stages
assumed by Levinson in bidirectional optimality theory. The reformulation is able to
avoid and solve a number of problems in Levinson's proposal and can avoid the Mprinciple
altogether which comes out as a theorem in bidirectional optimality theory.
Henk Zeevat reviews an earlier attempt to treat discourse particles within an
extended OT reconstruction of presupposition theory and concludes that more
particles can be treated and the analysis becomes simpler if one starts from the fact
that discourse particles are obligatory if the context of utterance and the current
utterance stand in one of a number of special relations, like adversativity, additivity,
contrast, etc.
29
A proper framework of OT is also the correct platform for asking
foundational questions. Given that we have violable principles and reliability ranking
between them (leťs assume this can be decided on empirical grounds), what follows
about the representations on which the constraints have to work, can a rational
foundation be found for each of the constraints and can the order between the
constraints be founded in some rational principle? The notions of relevance and
economy have particularly been in focus here. Another foundational issue concerns
the nature of bidirection and the symmetry assumption (e.g., Zeevat, 2000). Further
questions concern the division of labor between semantics and pragmatics in
particular, and the modularity stipulation in general. And what is the proper
architecture of an overall system integrating elements from syntax, prosody,
semantics and pragmatics?
In the present volume, David Beaver and Hanjung Lee give an overview of
various proposals in bidirectional optimality theory where crucial tests are total and
partial blocking, the Rat/Rad problem and some other problems. They show
conclusively that weak superoptimality cannot be combined with standard proposals
for optimality theoretic syntax with a larger number of constraints.
Gärtner's analysis of Icelandic object-shift and differential marking of (in)definites
in Tagalog addresses the issue of disambiguation in natural languages. In
the first part he suggests a family of OT-constraints called "Unambiguous
Encoding", which can be understood as a correlate of Gricean "Avoid Ambiguity". In
the second part he points out some shortcomings of this approach, and he suggests
that the OT-status of "Unambiguous Encoding" is epiphenomenal. Two ways of
reduction are explored which bolster the way for a functionalist understanding of the
phenomenon ­ viewing grammars as "harnessed" or "frozen" pragmatics (cf. Hyman,
30
1984). In addition, and not unrelated to the contribution of Beaver & Lee, he points
out some serious problems for Blutner's version of bidirectional OT.
Robert van Rooy is arguing that the general framework of optimality
theoretic pragmatics is able to include basic insights from relevance theory. Starting
from the bidirectionality of Blutner (1998; 2000) in terms of the Q- and I-principle,
he develops a decision-theoretic notion of relevance to take ­ in the first instance the
place of the Q-principle in this scheme for pragmatics. Though this leads to
improvements, further problems then force the tentative adoption of a relevance
based exhaustivity operator as a basis for reconstructing the Q-principle, the Iprinciple
and Blutner's bidirectionality. Horn's M-principle is then derived by
minimisation of effort.
Ralf Vogel is addressing the problem of OT architecture. Following
Jackendoff (1997) he is assuming three levels of representation: a semantic (=
conceptual), a syntactic and a phonological level. The correspondence between these
levels is modeled by a (bi-directional) OT grammar. Arguing that syntax is much less
encapsulated and `autonomous' than generative grammar usually assumes, Vogeľs
model is able to restrict OT-syntax by powerful pragmatic principles. In addition,
there is a methodological point that deserves particular attention. The proposed
architecture is not only motivated by its ability to account for certain intriguing
linguistic phenomena. It is also justified by its compatibility with current OT learning
theory.
OT pragmatics is a theory of pragmatic competence that invites to cross the
boundaries of traditional pragmatics and to relate it to psycholinguistic theories of
natural language performance (both production and comprehension) one the one
hand, and to theories of language learning and language evolution on the other hand.
This volume contains two contributions that explicitly conform to this challenge.
31
Jennifer Spenader's psycholinguistic investigation concerns the choice
between two demonstrative forms in Swedish (one simple, the other compound). A
multitude of factors influence the choice of one referential form over another, such as
abstractness, animacy, and the level of activation of the referent. The general finding
is that the simple form typically is used with more accessible and salient referents
while the compound form is used for referents with a lower level of activation.
Spenader argues that stochastic optimality theory is capable to model the subtle, yet
statistically significant differences between the two demonstrative forms ­ making
use of constraints that are independently motivated.
The contribution of Gerhard Jäger can be seen as the first step in a long
research agenda which derives from the view that many syntactic and semantic facts
are frozen pragmatics. It should be possible to show how particular languages
emerge from pragmatics assuming the fairly standard account of the evolution of
phonological forms. Even the advantages involved in moving from a purely
pragmatic language to a language with partial conventionalisation can be studied
from this perspective. The potential contribution of OT here is twofold. OT can
inspire learning algorithms and it can provide the framework for the representation
and evolution of grammatical knowledge. The diachronic perspective here offers a
far more sophisticated picture of the mode of existence of a language. It is not just a
conventional association between form and meaning, happening on some rather
poorly understood hardware and offering a window on the nature of that hardware, it
is one of the possible conventional associations that has a certain degree of stability
due to the conditions under which language is transferred to ever new speakers, their
ways of organizing these data, and the frequencies with which the various elements
making up the association are used.
32
In particular, Jäger applies a bidirectional generalisation of Boersma &
Hayes' (2001) learning algorithm to the formalisation and simulation of the
grammaticalisation processes underlying case systems. He is able to show that
structural case is the natural outcome of pragmatic case marking, that split ergative
systems naturally evolve into nominative accusative systems and that some other
systems are stable whereas others are either unlearnable or very instable. The account
also explains and underpins Aissen's (2000) treatment of differential case marking.
4 Problems and perspectives
The OT approaches to pragmatic phenomena seem to gain empirical advantages with
respect to their non-OT predecessors, but that is not the only advantage. Important is
the fact that we gain a different way of talking about these things in which
uniformities can appear across the description of the different phenomena and that
we have the prospect of a single theory of pragmatics where all the phenomena come
together. This unification is still a prospect but there a number of issues that can
already be distinguished.
The first issue is the existence of a pragmatic factorial typology. If there is a
factorial typology, then it would fly in the face of the pragmatic tradition that has
always maintained that pragmatics is universal and consists of a few principles that
can be founded in the conceptual analysis of linguistic communication, as in
Stalnaker (1999), Grice (1989), Sperber & Wilson (1986), Levinson (1983; 2000),
Horn (1984; Horn, 2003) and others.
Is it really possible that a constraint CONSISTENT (sometimes treated as part of
GEN) could be outranked by a constraint like ECONOMY OF EXPRESSION or
RELEVANCE? This would mean that there could be communities where it is more
33
important to be economical than to be consistent with the context, or more important
to be relevant than to be consistent with the context. In the first case, it would be not
be possible to mark corrections, in the second, the interpretation process would
maximize relevance without bothering about what we know already. It seems though
that there is a general functional case for keeping corrections apart from consistent
updates since the changes that have to be made to the knowledge of the interpreter
are of quite different. Rerankings of this kind have to our knowledge not been found
in the language communities of the world or only marginally (e.g., politeness can
override sincerity).
If we succeed in agreeing on a universal system of ranked pragmatic
constraints, there arises a second issue ­ a foundational one. It concerns the need not
just for explaining why there are these constraints and no others and why they are
ranked in this way. Because of the lack of variation, the factorial typology does not
help to support an empirical argument that our system is correct. Possible strategies
are the classical one of deriving the pragmatic system from pure reason, other
strategies might try to use an evolutionary argument, which establishes that the
pragmatic system is an evolutionary stable state by showing that any mutations
(rerankings, small changes to the individual constraints) are eliminated and moreover
that it is the only evolutionary stable state among a range of competitors. Of our
contributors, Van Rooy and Jäger are following these different strategies, and it is
one of the important questions of future research how to relate these different
approaches (see van Rooy (forthcoming) for a first step in answering this direction)
The third issue is how to reconcile a universal pragmatics with the obvious
fact that there is a great deal of variation in the syntactic, lexical and phonological
expression of pragmatic properties in the languages of the world. It is an important
insight that even if we have a pragmatic system, this does not mean that pragmatics
34
is purely universal. Languages exhibit enormous differences in their inventory of
pragmatically relevant items, like in pronouns (for a basic typology, see Bresnan,
2001), tense and aspect, definite and indefinite markers, presupposition triggers,
elliptical constructions, discourse particles. They also differ widely in their marking
strategies for information structure. The richness of the data here is still largely
unexplored especially in their interaction with the pragmatic treatments that have
been in the focus of OT pragmatics. It is unclear to what extent these typological
variations have a reflection on the abstract semantics. In Bresnan (2001), we see that
Chichewa free pronouns (i.e. the closest analogon to English pronouns) do not allow
antecedents that are topical, unlike English, where the pronoun predominantly refers
to topical elements. The difference is that, in Chichewa, there is a class of bound
pronouns realized in verbal agreement morphology that are used whenever the
antecedent is considered to be a topic. Chichewa is not so different from French: he
French clitic pronouns are used for topic, the free pronouns are used for the other
cases. (These cases are not so easy to delineate.) The morphological distinction
between zero, bound, clitic and free pronouns is not realized in all languages but
seems to align in different ways with a prominence hierarchy on the antecedents.
Whether this hierarchy is universal cannot be decided on the current state of
research. The hierarchy itself may be universal, but it is clear from data in Gundel,
Hedberg & Zacharski (1993) that zero pronouns do not align with the same property
in the different languages that have them in their inventory.
For example, it cannot be decided yet whether pronoun resolution can be split
up in a part to be treated in OT syntax and general pragmatic constraints on pronoun
resolution. If one follows Van Rooy, the general principle is relevance. It would
seem that for a particular treatment of e.g. Chichewa free pronouns, resolution would
35
need additional facts about the Chichewa inventory and the preference of bound
pronouns for topical discourse referents.
A fourth issue is the nature of pragmatic constraints. One of the special
features of the constraints that seem useful in pragmatics is that they seem like small
OT competitions on their own. Consider a neutral, as far as we know original
example, that is reasonably well understood, the resolution of ellipsis.
(11) Jan heeft een rode wollen trui gekocht en Piet drie blauwe.
"Jan has a red woolen sweater bought and Piet three blue"
The resolution process maximizes the similarity between the antecedent sentence and
the ellipsed sentence. In a syntactic copying perspective, it copies the verb, the
auxiliary, the object noun and one of the object adjectives. It does not copy the color
adjective, the subject and the object determiner. It is clear that higher order
unification, a tree assimilation algorithm, computation of the most specific common
denominator, and source reconstruction ­ to mention only some of the techniques
that have been applied to ellipsis ­ all attempt to make the ellipsed sentence as
similar as possible to its antecedent. This can be naturally described as an OT
competition6
. The point is that constraint violations to a constraint MAXIMISE
SIMILARITY must be scored by the existence of more similar candidates and that there
is no alternative to that, since correctness of the resulting sentence misses out on the
presence of optional material in the antecedent sentence, predicting e.g. that (12) is a
correct interpretation even though the adjective wollen ("woolen") is not taken along.
(12) Piet heeft drie blauwe truien gekocht.
"Piet bought three blue sweaters"
36
This seems the correct way to score DO NOT ACCOMMODATE, ECONOMY and
RELEVANCE and STRENGTH, the main pragmatic constraints that people have come
up with. We see whether there are otherwise correct interpretations with less
discourse referents, otherwise correct sentences with less nodes and words that have
the same interpretation in the context, or interpretations that deal with more questions
that the interlocutor can be assumed to entertain.
In concluding these introductory remarks we want to stress once more that
OT gives us a powerful instrument for implementing basic pragmatic mechanisms.
However, one should not forget that having a hammer in one's hands may seduce one
into seeing everything as a nail. For that reason methodological considerations for
restricting the proper domain of OT applications in the area of pragmatics are
important, and the significance of the three general conditions (A)-(C) of Section 1
deserves special attention in the area of pragmatics.7
On the other hand, we are at the
beginning of deeper understanding of our instrument, which ­ unlike a real hammer
­ has proven to be helpful in quite different respects. It possibly will facilitate the
integration of syntax, prosody, and pragmatics. It may allow to develop an
evolutionary perspective showing that particular language traits emerge from
pragmatics. And it may well provide a new framework research in psycholinguistics.
With any luck, the present volume helps to give a start.
37
Notes
1
Roughly, this condition states that the subject position must not be empty.
2
In this Introduction, names in bold type without a date refer to contributions to
this volume.
3
See, for instance, Hayes & Hayes (1989) and Green (1990). Studies with
chimpanzees have shown that they typically fail the symmetry test, but children
older than two years pass it (Dugdale & Lowe, 2000).
It should be noticed that the first half of the equilibrium's condition ­ what
we produce we are able to understand adequately ­ follows from the assumed
initial state of the OT Grammar (the markedness constraints outrank the
faithfulness constraint) plus the assumed mechanism of constraint demotion. In
contrast, the second half of the condition ­ what we understand we are able to
produce adequately ­ is independent of the initial state and an immediate
consequence of the learning mechanism. In the more general case of learning
arbitrary codes it needs extra requisites to ensure the symmetry condition. For
example, it requires a particular asymmetry between expressive and productive
optimization (see Zeevat, 2000; Jäger).
4
The origin of these ideas goes back to Blutner, Leßmöllmann, & van der Sandt
(1996) and Blutner (1998).
5
A very similar point was made in functionalist phonology (e.g., Boersma, 1998).
Most 'phonetically-driven' or functionalist theories of phonology propose that
two of the fundamental forces shaping phonology are the need to minimise effort
on the part of the speaker and the need to minimise the likelihood of confusion on
the part of the listener. The need to avoid confusion is hypothesised to derive
38
from the communicative function of language. Successful communication
depends on listeners being able to recover what a speaker is saying. Therefore it
is important to avoid perceptually confusable realisations of distinct categories;
in particular distinct words should not be perceptually confusable. The phonology
of a language regulates the differences that can minimally distinguish words, so
one of the desiderata for a phonology is that it should not allow these minimal
differences, or contrasts, to be too subtle perceptually. There is nothing new
about the broad outlines of this theory and it very closely relates to Zipf's (1949)
two opposing economies (see also Lindblom, 1986, 1990; Martinet, 1955).
6
with different flavours of resolution arising by different data structures that have
to be made as similar as possible, and the possibility of having different maxima
to account for ambiguities. This is not the place to take a stance on the empirical
and computational issues involved here.
7
Concerning the condition B, for instance, an interesting and new hypothesis is
that the hierarchical encoding of constraint strengths is correlated with the effect
of automaticity in psychological processes. Perhaps it is the area of pragmatics
where this hypothesis can be tested in the most effective way.
References
Aissen, J. (2000). Differential object marking: Iconicity vs. economy.Unpublished
manuscript, Santa Cruz.
Anttila, A., & Fong, V. (2000). The partitive constraint in optimality theory. Journal
of Semantics, 17, 281-314.
39
Atlas, J. D., & Levinson, S. C. (1981). It-clefts, informativeness and logical form. In
P. Cole (Ed.), Radical Pragmatics (pp. 1-61). New York: Academic Press.
Beaver, D. (to appear). The optimization of discourse anaphora. Linguistics and
Philosophy.
Beaver, D., & Clark, B. (2002). The proper treatment of focus sensitivity. In C. Potts
& L. Mikkelson (Eds.), Proceedings of WCCFL 21 (pp. 15-28): Cascadilla
Press.
Blutner, R. (1998). Lexical pragmatics. Journal of Semantics, 15, 115-162.
Blutner, R. (2000). Some aspects of optimality in natural language interpretation.
Journal of Semantics, 17, 189-216.
Blutner, R. (2002). Lexical semantics and pragmatics. Linguistische Berichte, 10, 27-
58.
Blutner, R., Leßmöllmann, A., & van der Sandt, R. (1996). Conversational
implicature and lexical pragmatics. Paper presented at the AAAI Spring
Symposium on Conversational Implicature, Stanford.
Boersma, P. (1998). Functional phonology. The Hague: Holland Academic Graphics.
Boersma, P., & Hayes, B. (2001). Empirical tests of the gradual learning algorithm.
Linguistic Inquiry, 32, 45-86.
Bresnan, J. (2001). The emergence of the unmarked pronoun. In G. Legendre & J.
Grimshaw & S. Vikner (Eds.), Optimality-Theoretic Syntax. Cambridge MA:
MIT Press.
Burzio, L. (1991). The morphological basis of anaphora. Journal of Linguistics, 27,
81-105.
40
Burzio, L. (1998). Anaphora and soft constraints. In P. Barbosa & D. Fox & P.
Hagstrom & M. McGinnis & D. Pesetsky (Eds.), Is the best good enough?
Cambridge, Mass.: The MIT Press.
Carston, R. (2002). Linguistics meaning, communicated meaning and cognitive
pragmatics. Mind and Language, 17(1/2), 127-148.
Carston, R. (2003a). Explicature and semantics. In S. David & B. Gillon (Eds.),
Semantics: A Reader. Oxford: Oxford University Press.
Carston, R. (2003b). Relevance theory and the saying/implicating distinction. In L.
Horn & G. Ward (Eds.), Handbook of Pragmatics. Oxford: Blackwell.
Chomsky, N. (1981). Lectures on government and binding. Dordrecht: Foris.
Cole, P. (Ed.). (1981). Radical pragmatics. New York: Academic Press.
de Hoop, H. (2000). Optimal scrambling and interpretation. In H. Bennis & M.
Everaert & E. Reuland (Eds.), Interface Strategies (pp. 153-168).
Amsterdam: KNAW.
de Hoop, H., & de Swart, H. (2000). Temporal adjunct clauses in optimality theory.
Rivista di Linguistica, 12(1), 107-127.
Dekker, P., & van Rooy, R. (2000). Bi-directional optimality theory: An application
of game theory. Journal of Semantics, 17, 217-242.
Ducrot, O. (1980). Les Echelles argumentatives. Paris: Minuit.
Dugdale, N., & Lowe, C. F. (2000). Testing for symmetry in the conditional
discriminations of language-trained chimpanzees. Journal of the
Experimental Analysis of Behavior, 73(1), 5-22.
Fodor, J. A., & Pylyshyn, Z. W. (1988). Connectionism and cognitive architecture: a
critical analysis. Cognition, 28, 3-71.
41
Gibson, E., & Broihier, K. (1998). Optimality theory and human sentence
processing. In P. Barbossa & D. Fox & P. Hagstrom & M. McGinnis & D.
Pesetsky (Eds.), Is the Best Good Enough. Optimality and Competition in
Syntax (pp. 157-191). Cambridge, Mass.: The MIT Press.
Green, G. (1990). Differences in development of visual and auditory-visual
equivalence relations. Journal of the Experimental Analysis of Behavior, 51,
385­392.
Grice, P. (1975). Logic and conversation. In P. Cole & J. L. Morgan (Eds.), Syntax
and Semantics, 3: Speech Acts (pp. 41-58). New York: Academic Press.
Grice, P. (1989). Studies in the way of words. Cambridge Mass.: Harvard University
Press.
Grimshaw, J., & Samek-Lodovici, V. (1995). Optional subjects and subject
universals. University of Massachusetts Occasional Papers in Linguistics, 18,
589-605.
Gundel, J., Hedberg, N., & Zacharski, R. (1993). Cognitive status and the form of
referring expressions in discourse. Language, 69(2), 274-307.
Hale, M., & Reiss, C. (1998). Formal and empirical arguments concerning
phonological acquisition. Linguistic Inquiry, 29, 656-683.
Haspelmath, M. (1999). Optimality and diachronic adaptation. Zeitschrift für
Sprachwissenschaft, 18, 180-205.
Hayes, S. C., & Hayes, L. J. (1989). The verbal action of the listener as the basis of
rule governance. In S. C. Hayes (Ed.), Rule governed behavior: Cognition,
contingencies, and instructional control (pp. 153-190). New York: Plenum
Press.
42
Hendriks, P., & de Hoop, H. (2001). Optimality theoretic semantics. Linguistics and
Philosophy, 24, 1-32.
Horn, L. (1984). Towards a new taxonomy of pragmatic inference: Q-based and Rbased
implicature. In D. Schiffrin (Ed.), Meaning, form, and use in context:
Linguistic applications (pp. 11-42). Washington: Georgetown University
Press.
Horn, L. (2003). Implicature. In L. Horn & G. Ward (Eds.), Handbook of
Pragmatics. Oxford: Blackwell.
Hyman, L. (1984). Form and substance in language universals. In B. Butterworth &
B. Comrie & Ö. Dahl (Eds.), Explanations for Language Universals (pp. 67-
85). Berlin: Mouton.
Jackendoff, R. (1997). The architecture of the language faculty. Cambridge,
Massachusetts: MIT Press.
Jacobson, R. (1962). Selected writings 1: Phonological studies. The Hague: Mouton.
Jäger, G. (2002). Some notes on the formal properties of bidirectional optimality
theory. Journal of Logic, Language and Information, 11, 427-451.
Jäger, G., & Blutner, R. (2000). Against lexical decomposition in syntax. In A. Z.
Wyner (Ed.), Proceedings of the Fifteenth Annual Conference, IATL 7 (pp.
113-137). Haifa: University of Haifa.
Kamp, H. (1981). A theory of truth and semantic representation. In J. Groenendijk et
al. (Ed.), Formal methods in the study of language. Amsterdam:
Mathematisch Centrum.
Kaplan, D. (1979). On the logic of demonstratives. Journal of Philosophical Logic,
8, 81-89.
43
Katz, J., & Fodor, J. A. (1963). The structure of semantic theory. Language, 39, 170-
210.
Levinson, S. (1983). Pragmatics. Cambridge: CUP.
Levinson, S. (1987). Pragmatics and the grammar of anaphora. Journal of
Linguistics, 23, 379-434.
Levinson, S. (2000). Presumptive meaning: The theory of generalized conversational
implicature. Cambridge, Mass.: MIT Press.
Lindblom, B. (1986). On the origin and purpose of discreteness and invariance in
sound patterns. In P. J. S & D. H. Klatt (Eds.), Invariance and Variability in
Speech Processes (pp. 493-510). Hillsdale, N.J.: L. Erlbaum.
Lindblom, B. (1990). Explaining phonetic variation: A sketch of the H & H theory.
In W. J. Hardcastle & A. Marchal (Eds.), Speech production and speech
modelling (pp. 403-439). Dordrecht: Kluwer.
Martinet, A. (1955). Économie des Changements Phonétiques: Traité de Phonologie
Diachronique. Berne: Éditions A. Franke.
Montague, R. (1970). Universal grammar. Theoria, 36, 373-398.
Prince, A., & Smolensky, P. (1993). Optimality theory. Rutgers Center for Cognitive
Science: Technical Report RuCCSTR-2.
Schwarzschild, R. (1999). GIVENness, AvoidF and other constraints on the
placement of accent. Natural Language Semantics, 7, 141-177.
Sperber, D., & Wilson, D. (1986). Relevance. Oxford: Basil Blackwell.
Stalnaker, R. (1999). Context and content: Essays on intentionality in speech and
thought. Oxford: Oxford University Press.
Tesar, B., & Smolensky, P. (2000). Learnability in optimality theory. Cambridge
Mass.: MIT Press.
44
Van Rooy, R. (2001). Conversational implicatures. In J. v. Kuppevelt & R. Smith
(Eds.), Proceedings of the 2nd SIGdial Workshop on Discourse and
Dialogue. Aalborg.
van Rooy, R. (forthcoming). Signalling games select Horn strategies. Linguistics and
Philosophy.
Wurzel, W. U. (1998). On markedness. Theoretical Linguistics, 24, 53-71.
Zeevat, H. (2000). The asymmetry of optimality theoretic syntax and semantics.
Journal of Semantics, 17, 243-262.
Zipf, G. K. (1949). Human behavior and the principle of least effort. Cambridge:
Addison-Wesley.