Default Semantics, Pragmatics, and Intentions

K.M. Jaszczolt
University of Cambridge

Address for correspondence:

Department of Linguistics
University of Cambridge
Sidgwick Avenue
Cambridge CB3 9DA
United Kingdom

Email: kmj21@cam.ac.uk

ABSTRACT

The paper concerns the division of labour between semantics and pragmatics, discussed mainly in the example of definite descriptions. The status of what is said and what is communicated is subject to debate in the current literature. Three major standpoints can be distinguished in the semantics/pragmatics boundary dispute: (i) traditional semantic ambiguity; (ii) Grice's unitary semantics complemented with conversational implicatures, and (iii) underdetermined semantics in which pragmatic aspects of meaning contribute to what is said. In this paper, I suggest the fourth solution and defend its superiority over the other views. My approach advocates a default semantics in which semantic representation is established with the help of intentions in communication. Intentions come in various types and strengths, including their default values. The default value triggers the default semantics for an utterance, whereas the departures from the default are signalled by the context. The approach is compatible with the pragmatic intrusionism of dynamic semantic theories such as Discourse Representation Theory where no distinction is made between semantic and pragmatic domains and semantics is unitary. My approach supplements dynamic semantics by an ordering of possible interpretations on the scale of salience, indicated by a principled account of intentions in communication.

"The theory of utterance-type meaning
should be a theory of default interpretation."
Levinson (1995: 109-110).

1. Outline
1.1. What is Said

This paper is a voice in the discussion concerning the distinction between semantics and pragmatics. It concerns mainly, although not exclusively, sentences with definite descriptions and demonstrates how an adequate interpretation can be provided without evoking the idea of underdetermined semantic representation.

The status of what is said is still subject to debate. Grice suggested that in the case of utterances which exhibit an ambiguity of interpretation that cannot be attributed to the lexicon or grammar, what is said should be regarded as the same in the two readings, whereas the conversational implicatures are different (Grice 1978). It has been argued since then that what is said is better seen as semantically underdetermined (Récanati 1989b). So, for instance, the difference between the referential and the attributive reading of definite descriptions is placed on the level of what is said because what is said is underdetermined (Récanati 1989b, p.300). The degree to which semantics is reduced and pragmatics enlarged remains in dispute (cf. Cole 1981: xiv). Three possible standpoints have been distinguished concerning the semantics-pragmatics interface. In this paper, I would like to suggest a fourth one and defend its superiority over its rivals. First, there is (i) the traditional position of semantic ambiguity advocated, among others, by Russell. Next, there is (ii) Grice's postulate of unitary semantics complemented with conversational implicatures. There is also (iii) a view that advocates underdetermined semantics according to which a pragmatic aspect of meaning contributes to what is said. The distinction is made here between the semantics of natural language and the semantics of the conceptual representation system, with pragmatic factors bridging the gap. Pragmatic factors are understood as contextual information plus the pragmatic principle of cooperation, including an account of speaker's intentions. The processes of supplementing the semantic representation is called completion (or saturation) and expansion (or strengthening; Bach 1994a; Récanati 1989b).

The fourth view which I am proposing advocates non-ambiguous semantics achieved through the interaction of the speaker's intention with the logical form of the expression. The problem of ambiguity does not arise because intentions 'intrude' into the semantic representation. The view is thus compatible with the pragmatic intrusionism of dynamic-semantic approaches such as Discourse Representation Theory (henceforth: DRT; Kamp and Reyle 1993), where pragmatic factors 'intrude' into a unitary semantic domain. There is one, discourse-level representation, to which syntax, semantics and pragmatics contribute. However, whereas DRT treats all possible resulting representations equally, the account proposed here orders them on the scale of salience and predictability from the default one to the furthest departure from the default.

The problem of semantic representation which is present in one way or another in the heterogeneous set of examples involving quantifiers, negation and other logical connectives was brought to the fore and sustained by intricate argumentation by Wilson 1975; Kempson 1975, 1977, 1979, 1986; Zwicky and Sadock 1975; Atlas 1979, 1989; Kempson and Cormack 1981; Sperber and Wilson 1986; Carston 1988, 1996; Récanati 1989b, 1993, 1994; van Deemter and Peters, eds 1996; Turner 1997, to name only a few out of a vast number of publications advocating in various ways and to various degrees the sense generality thesis or at least trying to exorcize unwanted ambiguities (here: Grice 1978; Horn 1972, 1985, 1988, 1989; Levinson 1988; Atlas and Levinson 1981). I suggest that the same methodological effect can be gained by using a device which, unlike the idea of underdetermined semantics, enjoys an indelible psychological plausibility. This role is performed by the technical notion of intentions in communication which direct the utterance interpretation at the correct semantic representation. I suggest an unambiguous and unitary semantics, obtained essentially along the lines of Kamp's (1984) dynamic semantics. However, I also attempt to resolve the problem of the various possible interpretations of such 'interpretively ambiguous' sentences which are assigned to them by Kamp's DRT. I suggest using the idea of intentions and the principle of economy with respect to the postulated levels of interpretation (see 2.1).

The paper focuses on definite descriptions but it also signals the possibility of accounting for other interpretatively ambiguous utterances, although, unlike for definites, no firm proposal will be offered for the rest.

Definite descriptions allow for two interpretations: the referential one, which makes them akin to proper names, and the attributive one, which makes them about whoever or whatever undergoes the description (Donnellan 1966). So sentences of the type (1) have two readings.

(1) The best architect designed this church.

The ambiguity is best seen as pragmatic (see e.g. Kripke 1977), although the exact placement of the pragmatic component in this analysis differs between current approaches (cf. Schiffer 1995). According to tradition (iii), definite descriptions have a unique, underspecified semantic representation which is further pragmatically enriched into a propositional form before being assigned a truth value. In this paper, I shall use the term 'underspecified' for the semantic representation which is not yet truth-evaluable, and 'underdetermined' in a wider sense of semantic representation in need of either further completion before it becomes truth-evaluable or expansion when it is truth-evaluable but intuitively in need of further elaboration before it renders correctly what is said.1

Another preliminary terminological clarification is due at this point. Semantic representation and logical form are not distinguished for the purpose of utterance interpretation. It is argued that intentions in communication help establish the semantic representation of the utterance and hence the semantics and pragmatic components are interwoven. Both terms are needed, though. Whereas in utterance interpretation we talk about semantic representation or, in DRT terms, discourse representation, in analyzing sentences we talk about logical form as the output of grammar. Hence, there can be more than one logical form corresponding to one sentence but only one logical form/semantic representation corresponding to an utterance. As a result, the notorious expression semantic ambiguity will be used (if at all) in the context of an epistemological commitment to the effect that a sentence has more than one semantic representation. This commitment will be rejected in agreement with the underspecified-semantics theorists, although no commitment to underspecified semantics will be made here. Logical ambiguity is used with respect to sentences which allow for more than one representation in the language of first-order logic. This term does not involve any epistemological commitment: as it will be acknowledged and argued further on, the fact that a sentence can be ascribed a logical form does not yet mean that it has this logical form. Kempson (1986: 79) says that "...a sentence is ambiguous if with respect to a single set of circumstances it can be both true and false". The unifying property of sentences analyzed in this paper is what I call an interpretative ambiguity. 'Interpretative ambiguity' is used as a loose, non-theoretic term (as far as it is possible), meaning a difficulty with assigning a unique interpretation to an utterance. So, for the present purpose, the category of departure for assigning an ambiguity is an utterance, not a sentence and it is the interpretative ambiguity that will interest us.

My main overall conclusion is that the analysis of definite descriptions in terms of intentions suggests that the semantic ambiguity/underdetermination dilemma disappears in a dynamic model of semantics when the latter is enriched with a principled account of intentions in communication.

1.2. Semantic Ambiguity?

Let us first have a look at some standard examples of alleged semantic ambiguity from the literature, beginning with negation.

(2) The present king of France is not bald.

In (2), the picture is this. The sentence has two readings corresponding to the external and the internal negation. According to the first, there is no king of France and the proposition is true, and according to the other merely the property of being bald is negated and the sentence is sometimes seen as lacking truth value (Strawson 1950). In this paper, the contentious issue of presupposition will not be discussed as it is not directly relevant for the concept of interpretative ambiguity which preoccupies us here. According to the standard Russellian interpretation, the sentence is ambiguous due to the fact that negation can take a wide or narrow scope. But the outcome of the discussion mentioned above has been a general consensus that since the two types of negation are not logically independent (Kempson 1979), semantic ambiguity should be exorcized and replaced with pragmatic processes that lead from the semantic form, containing general, unspecified sense (or one, more general reading, here opinions vary, cf. ibid.) to the establishment of a propositional form. This representation is an 'enriched' (in the informal sense) variant of a semantic representation to which we can assign truth values. Applied to (2), this theory renders a semantic representation which is unspecified for the scope of the negation operator and which is further filled in with necessary information according to Atlas's witty pastiche of Kant: "Pragmatic inference without sense-generality is blind, but sense-generality without pragmatic inference is empty." (Atlas 1989: 124, see also Atlas 1979 and for discussion Turner 1991). In other words, in interpreting an utterance we do not select among the available readings but rather construct one interpretation by means of pragmatic inference in the given context.

On the underdetermined semantics account, sentence (2) does not have a single set of truth conditions that could be assigned to it prior to assigning an interpretation. The sentence can be used to express two different senses, and thus it is often regarded as semantically indeterminate (cf. Bach 1987b: 202): it is said to have one, general, underdetermined meaning. The two readings are then obtained on the level of use, i.e. the level of pragmatics.

The same solution is often claimed to hold with respect to propositional attitude constructions, i.e. sentences of the type in (3):

(3) Mary believes that the present king of Sweden is bald.

This sentence is interpretatively ambiguous between the two readings: the transparent one, paraphrasable as in (3a), and the opaque one, paraphrasable as (3b):

(3a) Mary believes of the present king of Sweden that he is bald.
(3b) Mary believes that there is a king of Sweden and that he is bald.

Bach (ibid.: 206-207) argues that the distinction holds only for sentence use rather than for the semantic interpretation. He argues that although sentences exemplified here by (3) can be assigned two representations in first-order logic that differ in the scope of the existential quantifier applied to 'the king of Sweden' (ranging over the belief predicate and being within the scope of the belief predicate respectively), this fact does not suffice to infer that this sentence has these two logical forms and, a fortiori, that it is ambiguous. I am inclined to agree with the latter claim but, by the same non sequitur, I would argue that this consent does not yet mean that semantic representation has to be underspecified as to the particular meaning and as to truth conditions. Before reaching the verdict on the semantic representation of such interpretatively ambiguous sentences, it is necessary to investigate whether there are any general operational principles that would render one interpretation as primary, unmarked, and thus that would under normal conditions function as the standard semantic representation of the sentence. This paper aims at presenting a convincing positive answer to this question, an answer which is, moreover, compatible with the principle of Modified Occam's Razor advocated by Grice (1978), to be introduced in what follows.

First, some more terminological clarification. Underdetermined sense (and underdetermined semantic representation) can be easily explained with reference to (3). When an expression has two readings, such as a transparent and an opaque one in the case of (3), caused by two different occurrences of a definite description, the explanation of these two readings is said not to be provided by the logical form, i.e. by the scope of the existential quantifier: the narrow scope reading can still be either transparent or opaque, depending on the particular occasion of its use. And, if there is no such logical explanation of the ambiguity, there is no semantic ambiguity either. Now, those who exorcize the ambiguity view take a big leap from there to the claim that the semantic representation of (3) has to be indeterminate, underspecified, unfit to be assigned truth conditions (see Bach 1987b: 206-212). They say that only the uses of an underdetermined sentence have truth conditions, and thus the study of reference belongs to pragmatics (ibid.: 61). My standpoint represents a middle ground between the ambiguity thesis and the indeterminacy of sense: in my view, a sentence of the type (3) has a unique semantic representation that corresponds to the standard, default reading of the sentence and is achieved with the help of intentions that 'intrude' in the semantic representation. This representation can be assigned truth conditions and is fully operational on the level of sentence use in some (default) cases. As I shall demonstrate, other cases can be accounted for by the non-standard workings of intentions in communication.

The controversial idea of semantics based on default interpretations will be shown not to be so controversial when derived from the idea of default intentions and default reasoning, attained by a straightforward borrowing from, among others, Grice (1978) and Bach (1984). This clandestine agreement with both sides on the issue of indeterminacy versus unity will be further pursued in Section 2.

In short, I shall agree with the rejection of semantic ambiguity of negation, expressions of propositional attitude and definite descriptions but take a further step to demonstrate that even the level of underdetermined semantics can be disposed of by means of the influence intentions exert on the semantic representation. The main message is that the ambiguity/underdetermination dilemma is only the result of a wrongly posed problem.

The current picture is this: whereas lexical and syntactic ambiguities result in sentences which can be unrelated and cannot be traced back to one meaningful representation, in the case of other interpretative ambiguities there are said to be rules that lead from a general meaningful representation to one or the other fully truth-conditionally specified proposition. This is the idea of parsimony best spelled out by Grice (1978) as Modified Occam's Razor (henceforth: MOR):

MOR: Senses (linguistic meanings) are not to be multiplied beyond necessity.

(see also Récanati 1989a&b, 1994). Here the question arises as to what role this methodological principle plays in utterance interpretation. Is it merely a technical device which ensures that theorizing does not get out of hand and uses unifying rules rather than postulating differences wherever possible, or is it also a cognitive principle of utterance interpretation as performed by the hearer in the act of conversation? Either way, it is problematic as a fundamental device. In order to alembicate its real status, it is necessary to look first at the principle itself.

2. The Parsimony of Levels of Semantic Representation
2.1. Levels of Senses

The first difficulty stems out of the scope of MOR. It can be inferred from Grice's writings, such as his analysis of the sentential connective 'and' (Grice 1975: 24; 1978: 46) that if a sentence is syntactically complete and can express different propositions in different contexts, then it is semantically ambiguous. On the other hand, whenever possible, the differences in interpretation should be attributed to the context of utterance rather than to the sentence. In this way the proposition can be preserved as an unchanged entity across contexts of interpretation and what is actually communicated is retrieved or constructed thanks to the notion of implicature (cf. Récanati 1989b, 1994). However, as is clear from the dispute on ambiguity sketched in 1.1, one can claim that context aids in establishing the proposition (or: propositional form) to a much larger extent than Grice originally suggested (see Levinson 1988; Carston 1988; Bach 1994a, b).

Roughly, the idea is this. Sentences such as (4)-(7) have to be given some 'finishing off' so that the intended meaning could be recovered.

(4) You are not going to die.
(5) I haven't eaten.
(6) Everyone must wear a costume.
(7) Steel isn't strong enough.

(from Bach 1994b: 268). (4) requires 'expansion' to include the specification of the situation, like 'dying of the injury', (5) requires a temporal specification, (6) a restriction of the quantified expression 'everyone', and (7) requires 'completion' concerning the purpose for which steel is under scrutiny. All this requires a pragmatic process that gives rise to a full, intended proposition. This propositional form is called by Bach (1994a, b) 'impliciture' (a term preferred to Sperber & Wilson's 1986 'explicature' since, although it relates to what is said, it does not relate to what is said explicitly). MOR is satisfied as what is said can be resolved from the semantic representation with the aid of the context. My problem with this picture concerns only the place for semantic representation. In other words, must it be underdetermined and postulated in separation from the pragmatic processes that expand and complete it? I hope to dispel this generosity about the levels of sense with the help of intentions. The above sentences are incomplete and thus they are meaningful only in context. But this fact does not yet guarantee the viability of the leap from the (4)-(7) case to the epistemological commitment that semantics is underdetermined.

As we can see from examples (4)-(7), there are many processes in addition to reference assignment filtered into the slot between the semantic and the propositional representations (cf. also Levinson 1988 and Récanati 1994). From a methodological point of view, the question arises as to what advantage is gained from maintaining an underdetermined semantic representation, further filled in to produce a form with a truth value. There is a reading of interpretatively ambiguous sentences in which the distinction between the semantic form and the truth-evaluable propositional form is redundant. This reading will be demonstrated to be the default one. Hence, there would have to be a strong reason to postulate the distinction. There do not seem to be any such reasons. It is possible to give an equally adequate account of a sentence meaning in terms of a sole representation. This strengthening of MOR can be spelled out as follows:

POL: Levels of sense are not to be multiplied beyond necessity.

which I shall call Parsimony of Levels (POL). This principle boils down to a restriction on postulating the two levels of sentence interpretation: semantic representation (here equated with logical form) and propositional representation as separate aspects of meaning (sense), when they are not discernible with cognitive plausibility. If what is said is rendered as a sentence that is logically unambiguous, the truth conditions (or, generally, conditions of success) can be read off the semantic representation. If it is ambiguous, then the standard language usage may provide an interpretation that is the sole candidate for the interpretation in the relevant circumstances. As was pointed out in 1.2, Bach (1987b: 207) correctly observes that "Logicians and linguists commonly assume that because a sentence can be represented by a certain logical form, it has that logical form. This is a non sequitur." The sentence may either have one of these forms on the unmarked interpretation, or, in theory, it might have none and progress from a general sense to a fully specified one in either of the ways delineated by its ambiguity. I shall argue for the first alternative. If what is said was interpreted as a form that is transitional between the semantic form and a truth-conditionally identifiable content (see Carston 1988), then POL would be violated and, in the most radical case, i.e. the one involving underspecification requiring both completion and expansion, we would have to distinguish between (1) semantic form, (2) 'propositional' content of what is said, and (3) the 'true' propositional form which contains all the necessary information for the assignment of truth value. 'Necessary' rather than 'all' because otherwise the propositional form would contain too much information. Here we require a separation between truth-conditional and non-truth-conditional meaning by shifting some implicatures below the required line. In other words, we require only that part of pragmatic information which contributes to truth conditions. This much is widely acknowledged and we need not flog the dead horse.2 After all, if any meaning that is not derived by linguistic decoding was implicated (as Grice claims, see Carston 1988), then in many cases we would not be able to arrive at a single proposition since decoding need not necessarily always precede inferential processes. Or even, as Escandell-Vidal (1996 and personal communication) claims, decoding is always preceded by inferential processes which result in building what she calls a 'schema' or 'frame' ready to admit decoded information.

The idea behind POL is that full propositional representation can be equated with the semantic representation. This proposal is tested here on definite descriptions. But this formulation of the idea is only a tip of an iceberg and thus requires detailed justification. The core lies in intentions in communication as discussed by Grice (1969) on the one hand, and the tradition in the speech act theory on the other (cf. e.g. Searle 1983, 1992, Bach & Harnish 1979). The resolution is presented in the following section, followed by a presentation of its application to some chosen examples.

2.2. The Types of Intentions

This account makes use of three types of intentions: primary (reference-securing), communicative, and informative. The simplest case of referring is when the syntactically and semantically simplest expression is used with the intentions to communicate that the referent is to be recovered and to inform about this very referent (adapted from Sperber & Wilson's 1986 distinction). But there is an important point to be raised with regards to Sperber & Wilson's account of intentions. Grice (1957, 1969) says that the hearer understands the speaker's utterance by recognizing the speaker's intention (cf. Bach & Harnish 1979: xi: "...a communicative intention has the peculiar feature that its fulfilment consists in its recognition." See also discussion in Bach 1987a). Sperber & Wilson (1986) add here the distinction between the communicative and the informative intention, the second embedded in the first. If communicative intention is envisaged in the light of the theory of relevance, then what is communicated is entirely hearer-dependent. However, it is possible to conceive of conversation as a process of assumption creation performed in-between the speaker and the hearer, assigning to the hearer a more responsible role than just recovering the already existing assumptions. In fact, Bach & Harnish treat communication as a cooperative process, based on Grice's idea of speaker's intentions addressed at the hearer. I am not sure whether, if the above is true, one has to dispose of the default role of communicative intention in constructing meaning or make it intersect with the informative one (see Bach 1987a: fulfilment of communicative intention is its recognition. N.B., Bach & Harnish 1979: 7 do not distinguish an informative intention; they talk about the illocutionary-communicative intention being guaranteed by the 'communicative presumption': that whenever the speaker says something to the hearer, he/she is doing so with some (illocutionary) intention.). Communicative intention is defined as making it "mutually manifest to audience and communicator that the communicator has this informative intention." (Sperber & Wilson 1986: 61). It is certain, however, that the hearer is also invited to construct meaning rather than merely recover it (Levinas 1961, Jaszczolt 1996b), and thus that communicative intention may matter to a lesser degree in communication: making it manifest that 'this is what I want to say' can perhaps be overridden by a principle of how exactly the speaker and the hearer contribute to the meaning that is being established between them in the process of conversation. But this topic has been extensively discussed elsewhere (Jaszczolt, ibid.; Bird 1994).

2.3. The Degrees of Intentions for Definite Descriptions

On the referential use of definite descriptions, the intention to refer is present and is derived from the intentionality of the underlying mental act. Intentionality is a property of mental states that makes them about, or of objects and states of affairs (cf. also Searle 1983, 1990a). Only some mental states are intentional: beliefs, hopes, fears are, as opposed to nervousness or elation that are not. Now, some beliefs (thoughts) are externalized by means of language. This is how intention becomes a property of linguistic acts (by being, so to speak, 'inherited' from mental acts). Hence we can say that some intentional states are linguistic. On uttering a sentence, the speaker intends a particular object or person to be salient to the hearer and this intention constitutes the internal, undetachable property of the constructed representation.

Intentionality of securing a referent ensures that semantics is atomistic, about the real world, and that acts of meaning are acts of experience of the real world, about the objects of the world, and primarily aimed at the objects of the real world (in agreement with Husserl 1913). Intentionality so perceived incorporates the workings of intentions in communication, since language is one of the vehicles of thought (pace Lyons 1995; see Jaszczolt 1992, 1996a). The speaker's intention to communicate something and to inform the hearer about something also rely on this primordial intention to secure the referent. So, the claim is this:

PI: The primary role of intention in communication is to secure the referent of the speaker's utterance.

I distinguish a special kind of intention that performs the reference-securing task and call it the primary intention (PI), allowing, naturally, for the instances where this intention is absent either by force of the type of the expression used or by force of particular circumstances. The claim goes back to the nineteenth-century idea of intentionality, put forward by Brentano (1874) and developed by Husserl (1900-1901), although the connection between contemporary studies of intentions in communication and the phenomenological tradition is not always acknowledged and not always appreciated. It was Husserl (1913) who claimed that acts of consciousness are directed toward a real object. These objects trigger a semantic interpretation.3 Meaning is seen here as social, established in the process of social, meaning-giving acts. Knowing the object depends on fulfilling intentions either by thought, by perception, or by imagination. As language is one of the possible vehicles of thought (Jaszczolt 1996a), intentions also govern its use.

Naturally, not every sentence concerns an individual. But whenever a possibility of a referential interpretation is viable, it is secured by the referential intention which is assumed by the hearer to be present. This is the general idea behind PI. It has to be pointed out that the claim made in PI is considerably stronger than the current speech act/intentionality tradition allows for. Bach (1987b: 52), when discussing propositional attitude contexts, admits that when the speaker expresses an attitude towards an individual, the referential intention must be present. However, this claim does not aid the indiscernibility problem of referential and attributive readings of definite descriptions. Other works in the intentionality tradition follow suit. The general idea is this: as long as communication is said to incorporate only the need to inform or perform any other illocution (using illocutionary intention) on the part of the speaker, assumed by the hearer to be present, then the fact that the hearer normally comes up with one interpretation remains unexplained. After all, whether one subscribes to the underspecified semantics view or not, one has no explanation of the fact that although there is more than one interpretation of an utterance of the type (1), the hearer does not hold all of them in the mind, pondering which one to choose. Referential intention, when present, seems to explain this fact of instantaneous interpretation.

Now, having suggested this device for the cognitive level of interpretation (intentions), it seems plausible to maintain it for the semantics. Some linguists would not see a problem with having a technical device in their semantics that does not serve a cognitive explanation, i.e. does not have a psychologically plausible equivalent.4 Hence, they see no danger in multiplying levels of explanation and adding underdetermined representations. On the other hand, the price to pay for semantic parsimony may be high as we may have to revindicate some long abandoned ideas of speech act theory such as the default correlation between an utterance and a speech act. However, referential intention makes only a small demand in this contentious revival: there is no direct correlation between sentence types and speech acts. After all, PI allows for nonreferential interpretations which perform the same speech act as the referential reading of the same utterance (cf. example 1). Moreover, the problem under discussion concerns only the class of representatives and the issue of indirect speech acts remains free to be interpreted at the reader誷 will: to be either rejected outright, or left untouched. All in all, if intentions contribute to the semantic interpretation, they render the reading that depends on the presence or absence of the referential intention and this fluctuation has an indirectness flavour itself. More on this topic in 2.4 below.5

In order to see how this idea works for definite descriptions, it is necessary to establish what the objects of intentions are. In (1), when the speaker is taken to be talking about a particular, known individual, e.g. Christopher Wren, the referent is easily secured. But when the speaker talks about whoever happened to have designed the church and who is otherwise unknown to him/her, then intentions involved seem to be aimed at a person but fail to provide an identifiable referent. And, to take the third interesting case: one has to establish what is the object of the utterance in which the speaker is obviously referentially mistaken. This would be the case if in (1) the speaker was taken by the hearer to be talking about John Smith thinking mistakenly that Smith designed the church. Since the social act (speech act) secures reference, Smith should be the referent. However, one may also say that since intentionality is social, it reflects the use of language in a linguistic community and thus that intentions proceed towards the bearer of the name or description because this is what they normally do for a hearer who is not referentially mistaken. But let us suppose that the hearer is not sure whether the speaker meant Smith or Wren, i.e. the true architect of the church. This confusion may be engendered by the prior assertion of the speaker to the effect that Wren is known to the speaker as an eminent architect.

The explanation of these difficulties proposed here is this:

DI: Intentions come in various sizes, i.e. they allow for degrees.

I call this claim a principle of the degrees of intentions (DI). The principle accounts for the ordering of the three readings of definite descriptions in the following way. In the case of (i) a referential use of (1), intentions are the strongest and the referent is secured outright, as if by default. In the case of (ii) a referential mistake, intentions proceed towards two different objects, one along the path of 'social intentionality' and one along the 'individual path'. By 'social intentionality' I mean here the route which would be representative of an average native speaker who refers correctly, in agreement with the common wisdom. An 'individual path' represents the intentionality as it really occurs in the particular case of a particular speaker who can be referentially mistaken. The gap between them is reduced by utilizing contextual clues (what Husserl calls horizon) and one intended referent is established. In the case of (iii) an attributive use of (1), referential intention sets off as normal, but does not reach any particular referent. We can say that here we deal with the weakest type of intention. The intention is still there: on the attributive reading, the speaker still intends a referent to exist, whoever it might be.

I developed this proposal in detail in Jaszczolt 1997 but the particulars are irrelevant for this study of the semantic and the propositional forms of an utterance. The basic idea is that the semantics allows for the intervention of pragmatics to various degrees. It is essentially a dynamic semantics, exemplified in the 'intrusionistic' models of Kamp (1984); Kamp & Reyle (1993); and Heim (1988) but differs from them in supplementing POL with an ordering of resulting interpretations. In other words, intentions in their strong form secure the referent by their intrusion into semantic representation. The weaker the intentions, the more information has to be read off the rest of the semantic representation, the ultimate case being the attributive use of (1) where it is the semantic representation alone that secures the referent, whoever he or she might be.

2.4. The Primary Intention and Default Semantics

It is widely acknowledged that sentence meaning ultimately has to be analyzed in terms of intentions and other psychological states (Grice 1989; Sperber & Wilson 1986; Récanati 1993). My account subscribes to this claim. Bach (1987b:195) says that the distinction between referential and nonreferential occurrences of referring expressions (the category which includes for him definite descriptions) is drawn pragmatically and so is the distinction between transparent and opaque interpretations of propositional attitude constructions. In other words, the reading is specified on the level of speaker intention. In this section I further develop the claim that intentions come in degrees, that they help to establish the semantic representation, and that the basic intention is the one to secure a referent in communication. This intention guarantees the default semantics for definite descriptions and propositional attitude sentences. In order to fully grasp the role of intentions in communication, it is necessary to refer to the history of the idea that precedes Searle's (1983) idea of intentionality on one hand, and Grice's (1969) definition of meaning as complex intending on the other. I shall demonstrate that intentions in communication cannot be separated from intentionality of cognitive processes and that the accounts just mentioned are partial and insufficient for the purpose of dealing with representations in utterance interpretation. Searle's account conforms to the so called 'double intentionality': that of speech acts and that of physical actions. I am going to claim that this division is unfounded and it is more advantageous to be fully committed to intentions as psychological states per se, i.e. intentions that allow the hearer to 'read information off the world' and avoid the anfractuosities of argumentation for intentionality of speech acts.

Intending means essentially acting out of one's beliefs and desires. Grice observes that for a speaker to mean something by uttering a sentence he/she has to intend the hearer to [1] produce a particular response (cognitive or physical, as Strawson 1964 emphasizes), [2] think that the speaker intends that the hearer produces a response and [3] produce the response on the basis of thinking that this is what the speaker intends (see Grice 1969: 92). Except for some sophisticated situations where there is a need to step down to the nth level sub-intentions (ibid.), the schema seems to work, although in the case of conventionalized procedures, such as bidding in a game of bridge, conventions are said to do the job and secure the uptake of the intention automatically (see Strawson 1964). Whether one agrees with the role of conventions so assigned or not, it is conspicuous that one essential element is missing in these accounts and that is the explanation of what exactly it means to secure a response, according to what procedure and by means of what information. An account of speaker meaning that espouses intentions is wasted if stopped before letting the idea of intentions perform its task. And the task, as I see it, can extend to resolving the interpretative ambiguity problem. In order to do so, we have to say more about intention as securing the referent in conversation.

What is communicated by the utterance of (1) may differ depending on the context of interpretation. In order to account for this difference, one could follow either of the two procedures: either follow the radical pragmatics path and approve of the generality of sense, or construct an account of intentions securing reference. In order to avoid being between Scylla and Charybdis, a convincing decisive argument is required. Both traditions are acclaimed to enjoy a certain psychological plausibility, one derived from Chomsky's account of Universal Grammar and the other from the idea of phenomenology including the intentionality of acts of consciousness. Both pertain to innate abilities of the mind, with the different scope attributed to linguistic phenomena.6 As we have seen in 2.1, if we begin the account of utterance interpretation with the specification of intentions, we obtain, in most cases, a truth-evaluable level of propositional form as the first-level sense because the referent is secured. In other words, POL is satisfied. The level of an underdetermined semantic representation is then excluded by definition: securing reference is the most natural and expected process in utterance interpretation performed at an early stage, after some initial pre-decoding inferences, and necessarily simultaneously with decoding.7 Necessarily, because securing the referent is automatic on hearing the referring expression, just as the quantifier analysis is automatic on hearing an indefinite description. And this tendency to secure a default, unambiguous interpretation has to be captured by any cognitively plausible theory of communication. Kamp & Reyle (1993: 288-304) call it a general tendency of a type of expression that has to be captured by construction rules for discourse representations, i.e. rules for building their intrusionistic semantic representations of utterances. The conjecture to be made at this point is that the underspecified semantic representation may lack psychological plausibility: normally, intentions secure the object of the discourse and the unique reading of an utterance. It remains to be demonstrated that it is also theoretically redundant -- a task that can be only signalled rather than concluded in the present proposal.

In other words, the fact that a speaker means something by uttering a sentence is equivalent to saying that he/she utters it with "...the intention of inducing a belief by means of a recognition of this intention" (Grice 1957: 384). First of all, the speaker intends the object or person spoken about to be easily individuated by the hearer. This is the main preference in communication and the main point of reference for evaluating the role of referring and quantified expressions in utterance interpretation (Levinson 1987). It has been remarked about Donnellan's view that "...in some cases, we use the description to get to the referent and then throw away the description as it were" (Levinson 1988: 47). The hearer does so in the case of a referential interpretation of a mistaken description. In a sense, he/she does so in all cases of referentially used definite descriptions. He/she gets to the object as if by default: normally, people talk about known, existing entities. Thanks to the presence of such preferences in conversation, we can say that this process belongs to dynamic, context-sensitive semantics. Levinson gives here a convincing example to illustrate the thesis that enriched semantics is preferable to post-semantically, pragmatically resolved reference:

(8) The ham sandwich is getting restless.

The definite noun phrase refers, obviously, to a client who ordered a ham sandwich. Levinson claims (after Sag, cf. ibid.: 62) that denotations should be allowed to change within semantics, for instance along the lines suggested by Kaplan's (1989) distinction between character, which is a function from contextual indices to contents, and content itself (cf. also Jaszczolt 1997, fn 4). I discuss this issue in more detail in Section 3.

As is well known, instead of talking about reference, Donnellan suggests talking about reference attribution by the speaker. He adds that the referent should be "historically or causally connected to the speech act", although the theoretical explanation of this link is not provided (see Donnellan 1970: 356). A theory of the recovery of the intended referent along the lines of a theory of intentionality can fulfil this role. After all, who (if anybody) the speaker referred to is a matter of the speaker's intentions, or rather their recognition by the hearer (Donnellan 1966: 297). And so is the use of a definite description, referentially or attributively. So, the ambiguity of sentences such as (1) is neither properly semantic nor pragmatic.8 It is a semantic ambiguity only when we accept a 'static' semantics that lists the truth-conditional impact a description can have in all sorts of contexts, including intensional ones such as (3). The seeming interpretative ambiguity is accounted for by the dynamic semantic representation that allows for default interpretations of the utterance. The utterance is normally unambiguous and its interpretation proceeds towards the default reference, unless some contextual impediments stop it from doing so.

Degrees of intentions, default interpretations and the balance of costs and effects are all signalled in Bach's (1984) account of default reasoning. Bach says that default reasoning means "inference to the first unchallenged alternative" (1984: 38), i.e. to the first sense of the utterance that springs to the hearer's mind and is not immediately refuted in another thought. Default reasoning is automatic, it relies on generalizations and stereotypes of everyday life. People normally take it for granted that the interpretation that springs to mind is correct; no further reasoning process is involved. They do so in accordance with the rule that makes them calculate costs and effects of making judgements and decisions. Jumping to conclusions is such a cost-avoiding and effective default. (As Bach says, it is effective because the hearer knows when to jump to conclusions and when to think twice.) Equipped with PI and DI, we can also say that defaults correspond to the strongest intentions, unclipped by any conditions governing their fulfilment. The interpretation proceeds as referential because people normally talk about other individuals: we can jump to conclusions, there is no need to think twice unless the situation makes us do so.

Now, Bach observes that in order to work out this sketch into a full theory of default reasoning one needs an account of beliefs and intentions because beliefs and intentions influence reasoning in an important way. He also briefly suggests accounting for the "degree or strength of belief and intention" (1984: 49). In effect, the present investigation is such a proposal of an account of intentions in which the degrees of intentions are delineated and combined with the three types of intentions postulated here in accordance with PI and DI. All this has to be juxtaposed with a correct account of calculating costs and effects in the process of utterance interpretation.

It is also worth observing that the idea of the default referential interpretation is compatible with Levinson's (1987) proposal of minimization and with proceeding in utterance interpretation from his 'I-principle', i.e. saying the necessary minimum, towards the 'Q-principle', providing sufficient information for intention recovery. Shorter, more general reference forms are used as a default and only if more information is required, Q-principle is applied and the referring device is elaborated. The idea is also compatible with Sacks and Schegloff's (1979) principle of minimization and the principle of recipient design. The latter concerns communicative intentions and says that the speaker uses a reference form that allows the hearer to recognize the referent.

The intention-based account of meaning illustrates one general point remarked upon by Thomason: if intending to assert p does not differ from intending to assert q, then the meaning of p and q is the same. Intensionality of meaning and intensionality of intention go hand in hand (see Thomason 1990: 349). As Thomason observes, Grice's maxims are too loose to account for intentions in conversation. He suggests a theory of shared plans and goals. Thomason hopes for a theory of default reasoning that would be on a par with conversational implicatures. It is my belief that such a theory has to be based on intentions in communication, default reference, MOR, POL, and above all parsimony of processing, controlling and controlled by gained information.

The act of uttering a statement and the content of the statement are the two entities that were confused in philosophical texts for a long time, with a strong emphasis placed on the latter aspect (see Smith 1990: 31-32). The aftermath of this confusion is still present. It is only in Bolzano's, Brentano's and Husserl's writings where an act of saying and a concept became distinguished on psychological grounds. As Husserl says in his early work (1900-1901), meaning is created in intentional acts. So, in our terminology, the propositional form depends on the physical and psychological acts of utterance production and comprehension. Intentions and other psychological states are, so to speak, read off the world, as Reinach says (Smith 1990: 54) and they provide vital information for the construction of propositional form. If we followed Searle's theory of speech acts and his arbitrary, conventional constitutive rules, we would have to rely on social and methodological rules for utterance interpretation. This standpoint would not comply with the POL principle: a superfluous level of explanation would be created between the reality and the act of saying. Neither would it provide a tool for dealing with what Searle calls indirect speech acts, as was discussed above. By analogy, Searle's idea of intentionality is not consistent with POL either: it takes a long route by avoiding metaphysical commitments. Instead of merely comparing intentional states and linguistic expressions and postulating arbitrary sincerity conditions, one should specify what role linguistic expressions play in intentional states. Otherwise, one would have to have a theory of transfer of the conditions of satisfaction from the mind to the utterance, which is not only superfluous, but plainly ontologically, methodologically and epistemologically false. Searle talks here about the double level of intentionality: the utterance is intentional because conditions of satisfaction of a speech act (dependent on its essential condition) make it express a psychological state and the psychological state is itself intentional. So, intentionality of linguistic expressions is inherited from the intentionality of corresponding psychological states by means of the imposition of conditions of satisfaction (see Searle 1983: 27-28; Harnish 1990: 188). However, since intentional states can be regarded as 'psychological equivalents' of utterances as they express the 'directedness', the relation towards the world (what Searle calls the 'direction of fit'), utterances are speech acts, i.e. units of propositional content plus illocutionary force, matched on the psychological level by representative content and the psychological mode of its presentation (see Baumgartner and Klawitter 1990: 213). Instead of double intentionality, we easily arrive at a notion of a speech act as an example of an intentional act. Intentions are read off the situation, not off any artificially imposed conditions. Finding defaults in reasoning is another, albeit expected, bonus achieved by pursuing intentions in communication. This is, naturally, only the essence of the criticism, utilizing the POL to exorcize constitutive rules. Further elaboration would require a separate paper.

2.5. Application of POL and DI to Negation

I do not have a separate account of negation but I do have an account of definite descriptions which makes the solution for (1) also available as a possible solution for the scope of negation in (9). Russellian readings are presented in (9a) and (9b) and they reflect the difference in the scope of negation.

(9) The king of France is not bald.
(9a) Ex (King of France (x) & Ay (King of France (y) --> y =x) & notBald (x))
(9b) notEx (King of France (x) & Ay (King of France (y) --> y = x) & Bald (x))

where 'x' is an individual variable (cf. e.g. Neale 1990), and 'E' and 'A' stand for the existential and universal quantifier respectively. However, these representations are redundant in utterance interpretation. The speaker uses the sentence with the referential or the attributive reading and these are not represented in the above logical forms. The referential reading assimilates the description to the status of a proper name and, for our purposes, can be represented as (9c) where 'k' is an individual constant standing for the person who is the king of France (see Donnellan 1966).

(9c) notBald (k)

Russell's account does not provide for this reading and is thus standardly rejected.

What interests us here is (9c). This referential reading is the presumed default interpretation as it corresponds to the strongest primary intention. The suggestion that this interpretation is the default one conforms to the intuition that the sentence is problematic: either outright false, or neither true nor false due to not having a referent -- the choice is at the discretion of the informed reader. In a sense, the scope of negation comes later: the attributive reading with the narrow scope of negation is the middle case as far as the strength of intentions is concerned, followed by the weakest intentions in the case of the attributive reading with the wide scope of negation. The intrusion of intentions in the semantic representation guarantees nonambiguity and a determinate, complete semantics.

2.6. Application to Other Sentential Connectives

At present, there is no convincing intention-based account of sentences of the type (10) where the temporal 'and' is meant.

(10) They got married and had a baby.

We can conjecture that the strengthening of the meaning of a logical connective corresponds naturally to the strongest communicative and informative intentions. Therefore, as a default, 'and' will mean 'and then' or 'and as a result'; 'if' will render 'iff' and 'or' will stand for the exclusive disjunction. In such an interpretation, the contextual effect is strongest, the intention to inform of the situation is strongest, and if it really is the default, the processing effort is smallest. If we took the 'weaker' sense to be the default, ambiguity would flourish and the corresponding intentions would be present in their inhibited form. The third option, that of no defaults, is eliminated by force of MOR, POL and because the interpretation tends to be secured by the hearer, as confirmed by common-sense everyday observation. But these are only tentative proposals. After all, in (11) we would be in need of postulating a rule for non-sequentiality and non-consequentiality of 'and' -- perhaps coming from the fact that it is immediately diaphanous to the hearer that no sense-connection is meant between the conjuncts.

(11) He is a novelist and has three children.

This solution to the division of labour between semantics and pragmatics has already been in the air: Récanati indicates its possibility by saying that "...what is said is identified on an intuitive basis" (1989b: 309), or, more formally, proposes his Availability Principle:

"In deciding whether a pragmatically determined aspect of utterance meaning is part of what is said, that is, in making a decision concerning what is said, we should always try to preserve our pre-theoretic intuitions on the matter" (1989b: 310).

It is an intuitive prediction of the speaker's intentions that guides the hearer in interpreting the utterance (see also Carston 1988 and 1998).

2.7. Application to Numerals

I believe numerals can prove to conform to the rule of intentional, default interpretation sketched above. It is commonly held that e.g. 'three' in (12) is logically bound to mean 'at least three', and pragmatically bound to mean 'at most three' or 'exactly three' (cf. e.g. Horn 1985, Levinson 1988, Kempson and Cormack 1981).

(12) Max has three children.

The default interpretation can be postulated along the lines of POL enriched with an account of intentions. In this interpretation, 'three' means 'exactly three' as this is the case of the strongest informative intention -- and, of course, the highest informative content. By definition, quantifier expressions of this sort do not exhibit a primary intention. However, in my account all three types of intentions allow for degrees in agreement with the DI principle. Meaning 'at least three' is also possible; let us imagine a context where a parent who has three children qualifies for social benefits. The object of the predication is, however, only partially revealed and intentionality is weaker. Here we utilize PI principle to the effect that the primary intention takes precedence over the other two intentions; after all, it has a different cognitive status and a different methodological motivation. The same conclusion is easy to obtain on an intuitive basis. So, if the postulation of the strength of intentions incorporated into semantics is on the right track, then perhaps they can be seen as standing in a fixed inverse relation to processing effort. Sentence (13) would then gain the interpretation in (13a).

(13) Two examiners marked six scripts.
(13a) Exactly two examiners marked exactly six scripts.

This does not help much with the resolution of the relative scope of the quantifiers and with the collective and distributive reading. Still, this preliminary delimitation of the number of readings seems to be a plausible step. Next, the fact that the addition of 'each' would be meaningful seems to suggest, in agreement with the Cooperative Principle, that its omission is significant (see footnote 1) and points towards the collective reading. Kamp & Reyle (1993: 321) point out that any satisfactory paraphrase of a collective reading would have to refer to the fact that the effort of the individuals is combined -- either physically or in its outcome. They quote examples where the collective reading is obligatory or at least is strongly preferred and argue that the distributive reading is best regarded as obtained through an optional further step after the collective reading has been obtained (see ibid.: 320-330). The intention-based account succumbs to this strength of the collective interpretation and makes it a default by attributing a primary referential intention to a set, thus incorporating collective intentions which are not reducible to individual intentions. One can plausibly argue that collective intentions can be shown to interact with the other three dominant intentions in communication and strengthen them by strengthening the primary, referential intention. As a result, the default interpretation is produced and it is the collective one. The route to the default interpretation can be directly derived from the interaction of intentions presented in the case of definite descriptions, with the addition of the collective intention. Any further concerns about the status of the latter fall outside the focus of this discussion and are, after all, largely a matter of an epistemological commitment. At the moment, this is as far as the account leads, extending Searle's (1990b) view on collective intentions.

2.8. Indefinite Descriptions

Throughout this paper I have been discussing and in effect contesting the idea that underdetermined semantic representation always functions as a cognitively viable step in the analysis of interpretatively ambiguous expressions. The MOR dictates uniform semantics and accounting for any differences in meaning by means of pragmatic principles -- be it Gricean, neo-Gricean or relevance-theoretic. Definite descriptions, i.e. definite noun phrases in sentences of the type (14), allow for two interpretations: referential (about e.g. Jones) or attributive (about whoever Smith's murderer may be, see Donnellan 1966).

(14) The murderer of Smith is insane.

POL and DI guarantee the interpretation of the utterance which is either the default one or differs from it if the departure from the default is signalled in the context.

Indefinite descriptions are sometimes claimed to have even more readings, e.g. (15) is sometimes said to have a specific reading which is postulated as a reading intermediary between the referential and the attributive and it describes the case where the speaker does not intend the hearer to individuate the person (Ludlow & Neale 1991; Larson & Segal 1995).

(15) A friend of mine from Oxford paid me a visit.

The use of an indefinite signals that the identity of the object spoken of is not relevant for the purpose of the conversation. Indefinites do not induce the primary intention and are not logically ambiguous. If it is so, then the use of an indefinite serves the purpose of saying either that [1] the identity of the referent is unknown to the speaker or that [2] it is known to the speaker but irrelevant for this particular discourse either because the hearer would not know the referent anyway or because whoever it is, it is only the indefinite description that matters. Therefore, rather than postulating manifold ambiguities, MOR and POL suggest two interpretations: specific and nonspecific, which is after all a well acknowledged and established distinction (see Kamp & Reyle 1993). So, indefinite descriptions can be used specifically or nonspecifically. The specific use signals that the speaker is talking about a particular, known individual or object and that this individual or object is either taken to be new to the hearer or its identity is irrelevant for the conversation. The nonspecific use signals that the speaker is talking about an unknown individual or object. The difference between the specific and nonspecific use is truth-conditionally decisive in the examples where it triggers the ambiguity of scope as in (16) (from Kamp & Reyle 1993: 279), where on one reading there is a particular problem that preoccupies them all, and on the other reading every politician has some problem or other that preoccupies him or her.

(16) A problem about the environment preoccupies every serious politician.

However, we can say that the ambiguity of scope is engendered here by the quantified expression 'every politician'. Therefore the argument about the truth-conditional impact of the reading is not decisive. Hence, Ludlow & Neale's (1991) distinction into referential, attributive and specific use of indefinites appears rather contentious. In order to postulate it, the authors had to interweave a change of perspective from that of the recognition of the speaker's intentions by the hearer to that of the speaker's background knowledge. Attributive and referential uses are presented as straightforward cases of recognition of the speaker's intentions by the hearer. In the specific use more than the intentions has to be recognized: the hearer knows that the speaker knows the person talked about and that the speaker does not intend to communicate this information. Ludlow & Neale distinguish here the speaker ground, the proposition meant and the proposition expressed. However, if the account of communication is to be given in terms of PI and the communicative and informative intentions, then the change of perspective is unnecessary -- not to mention that it is methodologically inappropriate. The aim is to analyze the process of utterance interpretation and the establishment of meaning by the hearer and not the differences between the speaker's background knowledge and speaker's meaning. The perspective is that of the hearer recovering the speaker's intentions and thus of the language use rather than that of the totality of cognitive processes involved in utterance interpretation. In other words, there are only two readings that matter: specific and nonspecific seen as intended as specific and nonspecific. The totality of the speaker's thoughts and attitudes is irrelevant. The confusion is still rife in the literature but it is important not to run the two perspectives together.

In utterances with indefinite descriptions, when accounted for by means of primary, communicative and informative intentions, it seems natural to admit that the nonspecific interpretation is rendered as a default: indefinites are not referring expressions; the primary intention is missing and this fact allows the hearer to read more information off the semantic representation than would be the case when primary intention was in operation, i.e. in the case of definite descriptions. The semantics is not ambiguous: indefinites do not exhibit referential properties. They can, however, be specific. Even then their aim is not to secure a referent but rather by their means the speaker communicates to the hearer an event or state, at the same time making the hearer alert to the fact that no stronger statement was necessary or possible.

The specific and nonspecific readings can be explained as follows: the thought of the speaker exhibits intentionality because all acts of consciousness do. This intentionality can be stronger or weaker. However, the utterance does not exhibit the primary intention of securing the referent. The primary intention does not interact with the informative and the communicative intentions. This situation can be engendered either by the weakness of the primary intention of the mental act (the nonspecific reading of the indefinite) or by the conflict between the primary intention of the mental act and the informative intention (the specific reading; the referent may not be of interest in the conversation). In any case, the utterance is permeated with the primary intention of the speaker's thought and the specific or nonspecific interpretation is established between the speaker and the hearer. To repeat, the nonspecific interpretation occurs in the case where, according to the hearer, the intentionality of the mental state of the speaker is the weakest, i.e. it fails to reach the object. The specific reading is rendered when the speaker's thought is recognized as strongly intentional but the utterance fails to inherit this primary intention due to the interaction of the communicative and informative intentions. In other words, the underlying thought is either general and the primary intentionality is unrealized, i.e. does not reach the object, or concerns an intentional object but is not communicated at its face value.

2.9. Interim Conclusions

Let us take stock. Ludlow & Neale (1991: 184) assert that the uses of indefinite descriptions do not prove the existence of a separate referential interpretation. That is true. And so is the claim that 'donkey'-sentences of the form as in (17) are not ambiguous between the singular and plural interpretation of the description (cf. ibid.: 198).

(17) Every man who buys a donkey vaccinates it.

But proceeding to assigning one of the possible readings has to be more fine-grained, level-neutral, and allow for the dynamism of semantic representation and, a fortiori, for default interpretations.9 Sense need not be unspecified, it just needs to be constructed in situ as default sense or departures from the default, predicted by the recovered degrees of intentions. Only in this way will one avoid multiplying the levels of senses beyond necessity.

I have suggested that in the case of definite descriptions intentions are strongest when the speaker communicates a proposition about an individual known to the hearer (cf. 2.3). The hearer automatically processes the utterance as if it concerned a particular, known individual. In (1), this case corresponds to the referential reading. Nota bene, both the communicative and informative intention (of Sperber & Wilson 1986) are in their strongest form: the particular person is spoken of and it is intended that this information is recovered by the hearer. If, on the other hand, the speaker communicated a proposition about whoever fits the description, the informative intention would make the process of reference assignment stop at a certain stage: either at the very beginning, if the informative intention was strongly supported by the communicative one, or later on, if the communicative intention was weaker. This is due to the interaction of the primary, communicative and informative intentions. The strongest intentions correspond to the smallest effort in processing the message on the part of the hearer. We can say that this is the default situation: the message is optimally informative, minimally demanding in assumption recovery, and thus the situation is optimally advantageous for securing relevance. Intentionality, directedness at a referent, is secured without going through an intermediate, underspecified stage in utterance interpretation: the person or object is intended and this message 'automatically' gets through to the hearer.

To sum up: semantics remains uniform, unambiguous, but fully truth-conditional thanks to securing a unique reading of a sentence by means of intentions incorporated in the dynamic framework. Rather than underdetermination and underspecifiedness resolved later on by pragmatics, what we get is default interpretations. Only by accepting this standpoint can we explain the attributive use and referential mistakes of definite descriptions, as well as the multiple logical ambiguity of numerals (see Kempson & Cormack 1981).10 The alternative readings are simply the weaker cases of intentions: primary, referent-securing intention in the case of definite descriptions, and informative intention in the case of quantified expressions. They 'mix in' various degrees of information read off the semantic form and intentions in order to render a truth-evaluable representation. Indefinites are not ambiguous in this way and one need not resort to DI to discuss their reading. The semantics is thus flexible and dynamic: the semantic representation can be strengthened or weakened by intentions and the process is not temporal. In the default case there is no underdetermination as a step in the ladder. Kamp (1984) and Kamp & Reyle (1993) allow for the dynamism to be achieved by contextual specification; Récanati (1994) opts for Methodological Contextualism. And as it has been demonstrated here with a help from intentions, the dynamism can be taken 'higher up' in the 'semantics-proposition-implicatures' schema, to affect the semantic representation. As an additional bonus, we can now observe that the ambiguity-nonambiguity dilemma disappears and turns out to be merely a terminological mismatch. More about it in Section 3.

3. Unitary Default Semantics: Towards a Proposal
3.1. Discourse Representation and Cognitive Representation

In this section I am going to discuss the notion of a propositional form that constitutes part and parcel of an 'intrusionistic' dynamic semantic representation. I shall again take on board the theses of semantic ambiguity and underdetermination and demonstrate that, to a large extent, the polemic is driven by a terminological vagueness and confusion. Finally, I shall summarize the advantages of building a semantics around the idea of default interpretation.

The difficulty with defining semantic representation is engendered by the fact that there is no uniform connection between the output of grammar and the situation referred to by the linguistic string. The interpretation of the linguistic string does not undergo any strict semantic rules. But since neither does the situation itself, what is required is a restatement of the relation in more dynamic terms. The reality changes by the moves and actions of the participants and the interpretation of a linguistic string changes with more information becoming available, along the lines described by Heim's (1988) File Change Semantics where the task of the hearer is, so to speak, to construct a file and regularly update it so that it contained all the necessary information revealed by the speaker. So, for indefinite expressions the speaker starts a new file card (introduces a new discourse referent, cf. Karttunen 1976), and for a definite, an existing file card is updated, according to the conditions specified in the semantics. Files, interpretive devices that mediate between the language and the world, perform the role of a monitor that records all the changes of the situation, including both linguistic and non-linguistic information such as that from perception and background knowledge. They do not rely on logical forms (in Chomsky's sense) but rather overlap with them since logical form contains a lot of grammatical information which is irrelevant for constructing meaning. Lappin (1991) makes an even stronger claim that the logical form is redundant as a level separate from the S-structure to satisfy MOR and, as we can infer, POL as well. One of the main arguments for this standpoint comes from the ambiguity of the scope of quantified noun phrases that do not seem to require a unified LF in order for one or the other semantic representation to be assigned to them (cf. ibid.: 310-313. See also Kempson 1991/92 on how syntax and semantics are intertwined). So, even within the domain of syntactic analysis, logical form may be overly informative.

Changes in the situation may affect the truth value of an utterance. They cannot be recorded after the propositional form has been constructed because they affect the truth-conditional meaning of the utterance. Heim's solution evades this difficulty by introducing a level on which both utterances and other sources of information write down anything that is required. Essentially the same relation between the two components is proposed by Kamp's DRT. Levinson calls this perspective 'pragmatic intrusionism': "...there is a common slate, a level of propositional representation, upon which both semantics and pragmatics can write..." (Levinson 1988: 22). Pragmatic intrusionism is a more cognitively plausible model of utterance interpretation with its semantic and pragmatic aspects assigned an equally privileged status. Discourse representations are mental representations and are, so to speak, two-aspectual: they rely on model-theoretic semantics but also concern the meaning grasped by the hearer on hearing an utterance. The theory thus has to be composed of three stages: a generative syntax, rules deriving representations from syntactic constructions, and mapping from representations to a model, supplied with a definition of truth (cf. ibid.: 4). Logical ambiguities seem to vanish on the level of deriving a representation, thanks to the various possibilities of applying construction rules that account for the diversity of meanings (cf. Kamp & Reyle 1993). A discourse representation shows what the world would have to be like for the sentence to be true. Its construction rules account for the introduction of new referents and adding properties (or, generally, DRS-conditions, DRS being a Discourse Representation Structure, an interpretation of a sentence or a text) to the existing referents which is essentially based on the linguistic representation. When more information becomes available as the discourse proceeds, the representations are filled in with more properties and referents (see ibid.: 59).

3.2. Processing Options and Unambiguous Representations

A DRS is true if there are individuals in the universe of discourse that correspond to the discourse referents and the conditions of that DRS predicate something that can be truthfully said about these individuals. However, the interpretative ambiguity remains here as a full-blown ambiguity of processing options: in the case of indefinite NPs, for instance, the hearer can process the utterance to render the specific or the nonspecific reading, depending on the DRS in which a discourse referent for 'an x' is introduced. The appropriate reading can thus be achieved by means of choosing the appropriate order of rule application or by postulating into which DRS (in the case of embedded DRSs, or: opaque contexts) the referent is to be introduced (see Kamp & Reyle 1993: 288-293; Jaszczolt 1998). The theory does not predict under what circumstances the hearer will follow one or the other option, but neither does it postulate an ambiguous unit or an underspecified representation. Whether there are defaults or merely contextual preferences, the process of utterance interpretation relies on the whole discourse situation that is available and avoids multiplying stages through which this interpretation must proceed. Ambiguity is postulated because the sets of information contributing to each of the alternative DRSs are different in various respects, both semantic and pragmatic. But it is an ambiguity of the sentence rather than the utterance. Given a sentence, the interpretation of its various utterances can go either of the prescribed ways. Given an utterance, it proceeds along one particular way because of the default interpretation which is engendered by the joint commitment, i.e. by sharing of the discourse referents between the interlocutors (Kamp 1990) and, if applicable, because of overriding the defaults. According to Kamp, there are also devices of formal and external anchors for directly referential terms in the semantic interpretation: in order for a representation (DRS) to have truth conditions of a singular proposition, the DRS has to have a clear referent, i.e. be connected with an entity by an external anchor (ibid.). Kamp (1984) believes that the structures that people form during the cognitive process of utterance interpretation resemble the representations of Discourse Representation Theory. In cognitive processing, like in DRT, building an interpretation can proceed in various ways. For instance, anaphoric pronouns introduce referential ambiguities. Kamp says that background assumptions about the world help select referents in such problematic cases. Principles of inference in communication, forming intentions out of beliefs and desires, resemble principles of inference in formal logic and are applied to items with syntactic structure, i.e. logical formulas (Kamp 1990).

Now, it seems that these items do not yet constitute a semantic representation. Discourse understanding is incremental, proceeds bit by bit and in the process it accommodates changes of the context. In our account, some of these bits are inherently equipped with triggers of intentions, and intentions prevent the semantics from being ambiguous. Instead, they secure the reading either as the default, strongest one or a weaker one when intentions are weaker. For instance, who is meant by using a proper name depends on the context -- including the speaker's intentions (Kamp & Reyle 1993: 62). So, in (18), the discourse referent for a Porsche has to be introduced in the main, transparent, superordinate DRS which corresponds to the widest scope of the existential quantifier that handles 'a Porsche'.

(18) John doesn't like a Porsche. He owns it.

(ibid.: 106). But there are no strict conditions for when this should be done. Perhaps the issue would be less complicated if intentions were fully applied in a sense of reading information off the world or the inherent lack of this process. Then the hearer can be said to arrive at default readings and consecutively weaker readings corresponding to weaker intentions. Only if this schema is of no help, true communicative, interpretative ambiguity arises.

The problem also arises in sentences of the type (16), (19) and (20).

(16) A problem about the environment preoccupies every serious politician.
(19) Every boy in Mary's class fancies a girl who Mary doesn't know.
(20) Bill doesn't know a book that I have read during the past four weeks.

(ibid.: 279, 288, 303 respectively). Different readings of these sentences owe their existence to the different scope taken by the indefinite noun phrase. Kamp & Reyle suggest here two possibilities of accounting for the phenomenon: relaxing the order of application of rules for constructing DRSs or externally imposing a status of a quantifier or a referring expression on the indefinite, according to the hearer's judgement. Since such judgement is decisive in the theory, we can infer that it has a role to play in DRS construction and as such it prevents the occurrence of an ambiguity. It does so because the (cognitive) process of DRS construction intentionally aims at one DRS, the one that represents the reading recognized as intended by the speaker, or, to follow the doubly-dynamic perspective (Jaszczolt 1996b), the one [1] congruent with the intentions of the speaker but also [2] to be collaboratively achieved by the speaker and the hearer in the process of communication. So although "...the processing of indefinite noun phrases must allow for a considerable spectrum of alternatives" (Kamp and Reyle 1993: 292), these are alternatives to the default reading which is obtained by the ordinary, default application of construction rules. The default reading of indefinites always treats them as quantifiers. There is no reference-securing intention involved which would signal direct referentiality as is the case in the default reading of definites. But this intention may be present in other non-default readings, although in competition with the informative intention for a surface appearance. It is sometimes signalled by the presence of words such as 'certain', 'particular', 'given', and relative clauses. The authors say that in such cases indefinites are processed as definites and obtain their default, referential interpretation (see ibid.: 293). I have suggested that they are still processed as indefinites but the primary intention of the mental act affects the informative intention of the utterance.

Kamp and Reyle admit that they do not know how to distinguish between 'long' and 'short' indefinites and they have no good processing rule to account for this distinction. All they do is to postulate various processing options available to the hearer. However, the hearer 'chooses' the processing option automatically in the context of the conversation, not as one of the possible two. Since there are various reasons for scope ambiguities, one cannot fit the examples into one standard mould of logical variations of scope. There is no unique rule they would conform to. Any such rule would have to be psychologically plausible and the translations from natural language into first order logic are not sensitive to this plausibility (see Kamp & Reyle 1993: 304). The ambiguity is thus relegated to the level of utterance and resolved by utterance processing that relies on dynamic semantics allowing for pragmatic intrusionism. In this way the ambiguity-nonambiguity dilemma is exorcized for these cases. We can add, it is to be replaced with dynamic semantics, combined with contextualismand intentionality.

I have suggested in this paper that the multiplication of levels of representation in utterance interpretation can be avoided by means of recognizing intentionality and, consequently, default readings. This proposal requires a model of interaction of semantic and pragmatic information that would allow the two levels of analysis to build on each other as it is, for instance, proposed in Discourse Representation Theory and File Change Semantics (for pragmatic intrusionism see also Hausser 1981). Retaining truth-conditional semantics is methodologically advantageous because meaning is atomistic and we do not want to become involved in the even darker realms of the totality of a person's thoughts and the inner workings of mental states (Kamp 1990; Kamp & Reyle 1993; Fodor 1994). If, as in DRT, we allow nonverbal information to influence the DRSs, then the truth conditions of the formed propositions can be predicted correctly. There is even room for default interpretations of descriptions because interlocutors are taken to see themselves as sharing discourse referents (cf. Kamp 1990: 81). Although the inner-theoretical explanation of the phenomenon is weak there, we can propose intentions as the missing exegesis: intentions as acting out of one's beliefs and desires, and intending, at the same time, the particular object or person.

4. Conclusions

This paper is not intended to provide an ultimate solution to the problem of processing of interpretative ambiguities. Its aim is rather to draw attention to the fact that psychological plausibility and semantic parsimony should go hand in hand. Grice's principle of Modified Occam's Razor has been shown to be effective only in conjunction with a principle called here Parsimony of Levels. I suggested that, in the case of definite descriptions, the distinction between the semantic form and the truth-evaluable propositional representation need not be made, and provided criticism of some approaches that rely on underdetermined semantics. I suggested instead a Default Semantics where pragmatic information is naturally incorporated into semantic/cognitive constructs in a dynamic way, reflecting the dynamism of conversation. This approach supplements the existing dynamic semantics (e.g. DRT) by a principled account of intentions. It is argued that ambiguity can be accounted for by means of introducing degrees of intentions that help construct the intended DRS, i.e., to reach the intended interpretation. All in all, in the analysis of discourse, the need for sense generality seems to disappear: there are only defaults and overridden defaults, recovered in a principled way.

REFERENCES

Anderson, C.A. and Owens, J. (eds). 1990. Propositional Attitudes: The Role of Content in Logic, Language, and Mind. Stanford: CSLI.

Atlas, J.D. 1979. "How linguistics matters to philosophy: presupposition, truth, and meaning". In Oh and Dinneen (eds) 265-281.

Atlas, J.D. 1989. Philosophy Without Ambiguity: A Logico-Linguistic Essay. Oxford: Clarendon Press.

Atlas, J.D. and Levinson, S.C. 1981. "It-clefts, informativeness, and logical form: radical pragmatics (revised standard version)". In Cole (ed.) 1-61.

Bach, K. 1987a. "On communicative intentions: a reply to Récanati". Mind and Language 2: 141-154.

Bach, K. 1987b. Thought and Reference. Oxford: Clarendon Press.

Bach, K. 1984. "Default reasoning: jumping to conclusions and knowing when to think twice". Pacific Philosophical Quarterly 65: 37-58.

Bach, K. 1994a. "Conversational impliciture". Mind and Language 9: 124-162.

Bach, K. 1994b. "Semantic slack". In Tsohatzidis (ed.) 267-291.

Bach, K. and Harnish, R. 1979. Linguistic Communication and Speech Acts. Cambridge, MA: MIT Press.

Baumgartner, W. and Klawitter, J. 1990. "Intentionality of perception". In Burkhardt (ed.). 210-225.

Bird, G.H. 1994. "Relevance theory and speech acts". In Tsohatzidis (ed.) 292-311.

Brentano, F. 1874. Psychology from an Empirical Standpoint.Transl. by A.C. Rancurello, D.B. Terrell, and L.L. McAlister in 1973. London: Routledge and Kegan Paul.

Burkhardt, A. (ed.). 1990. Speech Acts, Meaning and Intentions: Critical Approaches to the Philosophy of John R. Searle. Berlin, New York: Walter de Gruyter.

Carston, R. 1988. "Implicature, explicature, and truth-theoretic semantics". In R. M. Kempson (ed.). Mental Representations: The Interface Between Language and Reality. Cambridge: CUP. 155-181.

Carston, R. 1996. "Metalinguistic negation and echoic use". Journal of Pragmatics 25: 309-330.

Carston, R. 1998. "Postscript (1995)" to Carston 1988. In A. Kasher (ed.). Pragmatics: Critical Concepts. Vol. 4. London: Routledge. 464-479.

Cohen, P.R., Morgan, J. and Pollack, M.E. (eds). 1990. Intentions in Communication. Cambridge: CUP.

Cole, P. 1981. Preface to Cole (ed.) xi-xiv.

Cole, P. (ed.). 1981. Radical Pragmatics. New York: Academic Press.

Donnellan, K.S. 1966. "Reference and definite descriptions". Philosophical Review 75: 281-304.

Donnellan, K.S. 1970. "Proper names and identifying descriptions". Synthese 21: 335-358.

Dummett, M. 1991. Frege and Other Philosophers. Oxford: Clarendon Press.

Escandell-Vidal, V. 1996. "Towards a cognitive approach to politeness".

In K. Jaszczolt and K. Turner (eds). Contrastive Semantics and Pragmatics. Vol. 2. Oxford: Elsevier Science. 629-650.

Fodor, J.A. 1994. The Elm and the Expert: Mentalese and Its Semantics. Cambridge, MA: MIT Press.

Grice, H.P. 1957. "Meaning". Philosophical Review 66: 377-388.

Grice, H.P. 1969. "Utterer's meaning and intentions". Philosophical Review 78. Reprinted in Grice 1989. 86-116.

Grice, H.P. 1975. "Logic and conversation" In P. Cole and J.L. Morgan (eds). Syntax and Semantics. Vol. 3. New York: Academic Press.. Reprinted in Grice 1989. 22-40.

Grice, H.P. 1978. "Further notes on logic and conversation". In P. Cole (ed). Syntax and Semantics. Vol. 9. New York: Academic Press. Reprinted in Grice 1989. 41-57.

Grice, P. 1989. Studies in the Way of Words. Cambridge, MA: Harvard University Press.

Harnish, R.M. 1990. "Speech acts and intentionality". In Burkhardt (ed.) 169-193.

Hausser, R.R. 1981. "The place of pragmatics in model theory". In J.A.G. Groenendijk, T.M.V. Janssen and M.B.J. Stokhof (eds). Formal Methods in the Study of Language. Part 1. Amsterdam: Mathematisch Centrum. 183-205.

Heim, I. 1988. The Semantics of Definite and Indefinite Noun Phrases. New York & London: Garland Publishing.

Horn, L.R. 1972. On the Semantic Properties of Logical Operators in English. University of California dissertation. Reprinted in 1976 by Bloomington: Indiana University Linguistics Club.

Horn, L.R. 1985. "Metalinguistic negation and pragmatic ambiguity". Language 61: 121-174.

Horn, L.R. 1988. "Pragmatic theory". In F.J. Newmeyer (ed.). Linguistics: the Cambridge Survey. Vol. 1. Cambridge: CUP. 113-145.

Horn, L.R. 1989. A Natural History of Negation. Chicago: University of Chicago Press.

Husserl, E. 1900-1901. Logical Investigations. Transl. by J.N. Findlay in 1970. London: Routledge & Kegan Paul.

Husserl, E. 1913. Ideas. General Introduction to Pure Phenomenology. Transl. by W.R. Boyce Gibson in 1931. London: George Allen & Unwin.

Jaszczolt, K.M. 1992. Belief Sentences and the Semantics of Propositional Attitudes. D.Phil. dissertation. University of Oxford.

Jaszczolt, K.M. 1993. "De re/de dicto: a semantics of belief sentences". Papers and Studies in Contrastive Linguistics 28: 39-64.

Jaszczolt, K.M. 1996a. "Reported speech, vehicles of thought, and the horizon". Lingua e Stile 31: 113-133.

Jaszczolt, K.M. 1996b. "Relevance and infinity: implications for discourse interpretation". Journal of Pragmatics 25: 703-722.

Jaszczolt, K.M. 1997. "The 'default de re' principle for the interpretation of belief utterances". Journal of Pragmatics 28: 315-336.

Jaszczolt, K.M. 1998. "Discourse about beliefs". Theoretical Linguistics 24: 1-28.

Jaszczolt, K.M. 1999. Discourse, Beliefs, and Intentions: Semantic Defaults and Propositional Attitude Ascription. Oxford: Elsevier Science.

Kamp, H. 1984. "A theory of truth and semantic representation". In J. Groenendijk, T.M.V. Janssen and M. Stokhof (eds). Truth, Interpretation and Information. Selected Papers From the Third Amsterdam Colloquium. Dordrecht: FORIS Publications. 1-41.

Kamp, H. 1990. "Prolegomena to a structural account of belief and other attitudes". In Anderson and Owens (eds) 27-90.

Kamp, H. and Reyle, U. 1993. From Discourse to Logic: Introduction to Modeltheoretic Semantics of Natural Language, Formal Logic and Discourse Representation Theory. Dordrecht: Kluwer.

Kaplan, D. 1989. "Demonstratives: An essay on the semantics, logic, metaphysics, and epistemology of demonstratives and other indexicals". In J. Almog, J. Perry and H. Wettstein (eds). Themes From Kaplan New York: OUP. 481-563.

Karttunen, L. 1976. "Discourse referents". In J. McCawley (ed.). Syntax and Semantics. Vol.7. New York: Academic Press. 363-385.

Kempson, R.M. 1975. Presupposition and the Delimitation of Semantics. Cambridge: CUP.

Kempson, R.M. 1977. Semantic Theory. Cambridge: CUP.

Kempson, R.M. 1979. "Presupposition, opacity, and ambiguity". In Oh and Dinneen (eds) 283-297.

Kempson, R.M. 1986. "Ambiguity and the semantics - pragmatics distinction". In C. Travis (ed.). Meaning and Interpretation. Oxford: B. Blackwell. 77-103.

Kempson, R.M. 1991/92. "Semantics vs syntax? Shifting perspectives on natural language content". SOAS Working Papers in Linguistics and Phonetics 2: 27- 40.

Kempson, R.M. and Cormack, A. 1981. "Ambiguity and quantification". Linguistics and Philosophy 4: 259-309.

Kripke, S. 1977. "Speaker誷 reference and semantic reference". In P.A. French, T.E. Uehling and H.K. Wettstein (eds). Contemporary Perspectives in the Philosophy of Language. Minneapolis: University of Minnesota Press. 6-27.

Lappin, S. 1991. "Concepts of logical form in linguistics and philosophy". In A. Kasher (ed.). The Chomskyan Turn. Oxford: Blackwell. 300-333.

Larson, R. and G. Segal. 1995. Knowledge of Meaning: An Introduction to Semantic Theory. Cambridge, Mass.: MIT Press.

Levinas, E. 1961. Totality and Infinity: An Essay on Exteriority. Transl. by A. Lingis in 1969. Pittsburgh, PA: Duquesne University Press.

Levinson, S.C. 1987. "Minimization and conversational inference". In J. Verschueren and M. Bertuccelli-Papi (eds). The Pragmatic Perspective. Selected Papers from the 1985 International Pragmatics Conference. Amsterdam/Philadelphia: J. Benjamins. 61-129.

Levinson, S.C. 1988. "Generalized conversational implicature and the semantics/pragmatics interface". Unpublished paper.

Levinson, S.C. 1995. "Three levels of meaning". In F.R. Palmer (ed.). Grammar and Meaning. Essays in Honour of Sir John Lyons. Cambridge: CUP. 90-115.

Ludlow, P. and Neale, S. 1991. "Indefinite descriptions: in defense of Russell". Linguistics and Philosophy 14: 171-202.

Lyons, W. 1995. Approaches to Intentionality. Oxford: Clarendon Press.

Neale, S. 1990. Descriptions. Cambridge, MA: MIT Press.

Oh, C.-K. and Dinneen, D.A. (eds). 1979. Syntax and Semantics. Vol.11. New York: Academic Press.

Récanati, F. 1989a. "Referential/attributive: a contextualist proposal". Philosophical Studies 56: 217-249.

Récanati, F. 1989b. "The pragmatics of what is said". Mind and Language 4: 295-329.

Récanati, F. 1993. Direct Reference: From Language to Thought. Oxford: B. Blackwell.

Récanati, F. 1994. "Contextualism and anti-contextualism in the philosophy of language". In Tsohatzidis (ed.) 156-166.

Sacks, H. and Schegloff, E.A. 1979. "Two preferences in the organization of reference to persons in conversation and their interaction". In G. Psathas (ed.). Everyday Language: Studies in Ethnomethodology. New York: Irvington. 15-21.

Schiffer, S. 1995. "Descriptions, indexicals, and belief reports: some dilemmas (but not the ones you expect)". Mind 104: 107-131.

Searle, J.R. 1983. Intentionality: An Essay in the Philosophy of Mind. Cambridge: CUP.

Searle, J.R. 1990a. "Consciousness, unconsciousness and intentionality". In Anderson and Owens (eds) 269-284.

Searle, J.R. 1990b. "Collective intentions and actions". In Cohen, Morgan and Pollack (eds) 1990. 401-415.

Searle, J.R. 1992. The Rediscovery of the Mind. Cambridge, MA: MIT Press.

Smith, B. 1990. "Towards a history of speech act theory". In Burkhardt (ed.) 29-61.

Sperber, D. and Wilson, D. 1986. Relevance: Communication and Cognition. Oxford: Blackwell.

Strawson, P.F. 1950. "On referring". Mind 59. Reprinted in Strawson 1971. 1-27.

Strawson, P.F. 1964. "Intention and convention in speech acts". Philosophical Review 73: 439-460.

Strawson, P.F. 1971. Logico-Linguistic Papers. (2nd edition) London: Methuen. 75-95.

Thomason, R.H. 1990. "Accommodation, meaning, and implicature: interdisciplinary foundations of pragmatics". In Cohen, Morgan and Pollack (eds) 325-363.

Tsohatzidis, S.L. (ed.). 1994. Foundations of Speech Act Theory: Philosophical and Linguistic Perspectives. London: Routledge.

Turner, K. 1991. Review of Atlas 1989. Journal of Literary Semantics20: 188-192.

Turner, K. 1997. "On 'Pragmatic considerations in semantic analyses' ". Pragmatics and Cognition 5: 163-176.

van Deemter, K. and S. Peters (eds). 1996. Semantic Ambiguity and Underspecification. Stanford: CSLI Publications.

Wilson, D. 1975. Presuppositions and Non-Truth-Conditional Semantics. London, New York: Academic Press.

Zwicky, A. and Sadock, J. 1975. "Ambiguity tests and how to fail them". In J.P. Kimball, (ed.). Syntax and Semantics. Vol. 4. New York: Academic Press. 1-36.

NOTES

* I am indebted to Peter Matthews, Kent Bach, Barry Smith and Victoria Escandell-Vidal for their invaluable comments on the previous draft of this paper and to Ken Turner and Stephen Schiffer for the discussion of various aspects of what follows.

[1] This idea of a unique but incomplete semantic representation is also said to account for the ambiguities of the sentences of the type (10) with the sentential connective 'and' which is ambiguous between logical conjunction and the temporal or consequential conjunction:

(10) They got married and had a baby.

(see e.g. Carston 1988). Here the semantic representation is not underspecified: it is truth-evaluable but, on the underdetermined semantics standpoint, the proposition requires further expansion. The underdetermined-semantics account has also been applied to other interpretatively problematic cases. Discussion on the status of the ambiguities of sentences of the type (2) and (13) which allow for more than one logical representation (see e.g. Kempson 1979; Kempson & Cormack 1981; Carston 1996) has evoked much disagreement over the past two decades and focused on exorcizing semantic ambiguity and applying the idea of underdetermined, including underspecified, semantics instead.

(2) The present king of France is not bald.
(13) Two examiners marked six scripts.

Sentences (2), (10) and (13) frequently appear in the literature on the subject. They are said to exhibit a mixture of cases including underspecified and complete semantics, i.e. a need for either completion or expansion. See sections 1.5 - 1.7.

[2] cf.: "A number of philosophers have contended that Grice completely overlooked the fact that inferential processes of essentially the same sorts as those involved in implicature enter into determining what is said." Bach (1994: 269-270). In other words, what is implicit in what is said differs from implicature sensu stricto.

[3] What Husserl called objectifying, meaning-giving acts. See Jaszczolt 1992 and 1996a.

[4] Victoria Escandell-Vidal (personal communication) says: "...I would probably still prefer a two-layered model without claiming any psychological reality (...), at least until a way out of the problems of the notion of speech act is found." and: "...what you gain in psychological verisimilitude gets somewhat lost in theoretical elegance; and vice versa, what you gain in theoretical elegance you lose in psychological reality."

[5] Bach & Harnish (1979: 93) say on this point that there is no standard strategy for recovering the illocutionary intention. My addition to this minimalism is to claim that although there is no recipe for interpretation, there is one for standard interpretation in sentences that involve the multiplicity of senses, secured by the interaction of intentions, and, where applicable, by the referential intention. Bach (1987b: 66) recognizes the referential intention in the case of indexicals: their meaning depends on the referential intention of the speaker. From here there is only one step to saying that if things get complicated in interpreting informationally fuller referring expressions (descriptions, proper names), the hearer also resorts to the intentions of the speaker.

[6] This is not a primary concern of the paper. For an account of the vehicles of thought see Dummett 1991, Jaszczolt 1992 and forthcoming.

[7] As Escandell-Vidal (personal communication) observes, pure decoding may have no place in this account.

[8] In one of my earlier papers (Jaszczolt 1993) I claimed that in Polish there are lexical and syntactic means of distinguishing between talking about a known and an unknown individual in belief reports (de re and de dicto). However, the use of this distinction is also pragmatically-driven and constitutes in itself strong evidence for the default status of the de re interpretation of belief statements and belief reports.

[9] There are serious implications here for the distinction between being a Russellian and being a referentialist. See Neale 1990. But the opposition may after all boil down to being founded on a terminological mismatch. For arguments in favour of such a reconciliation see Récanati 1993.

[10] N.B., in (13), the speaker could say:

(13') Two examiners marked six scripts each.

which helps to demonstrate that this reading is not the default one of (13). On the other hand, in (13''), 'between them' sounds redundant:

(13'') Two examiners marked six scripts between them.

The final version of this paper is published in: K. Turner, ed. 1999. The Semantics/Pragmatics Interface from Different Points of View. Oxford: Elsevier Science. pp. 199-232. Copyright: 1999 Elsevier Science Ltd.