Advertisement

Referring Expressions in Speech Reports

  • Kaja BorthenEmail author
  • Barbara Hemforth
  • Barbara Mertins
  • Bergljot Behrens
  • Cathrine Fabricius-Hansen
Chapter
  • 609 Downloads
Part of the Studies in Theoretical Psycholinguistics book series (SITP, volume 44)

Abstract

When choosing among various referring expressions, speakers typically choose a form that reflects the audience’s mental representation of the intended referent. For instance, a speaker will most likely use a definite rather than indefinite description when introducing an entity that the addressee can uniquely identify. However, also considerations other than referent accessibility and the mental state of the addressee may affect the choice of nominal form. For instance, in a speech report such as Mary asked whether he had seen a dog, the choice of the expression a dog is influenced by the speaker’s intention to truthfully report on what was originally communicated as well as considerations about the representation of the referent in the mental model of the present addressee—and more than one nominal form may be valid. This paper reports on a pen-and-pencil experiment conducted to test how specific indefinites are reported on in direct and indirect speech in the four languages Czech, English, German, and Norwegian. The experiment supports the claim that indirect speech allows for a wider range of nominal forms than direct speech when the speaker reports on a speech event that originally contained a specific indefinite. Nevertheless, the study shows that the subjects prefer to use an indefinite description to report on a specific indefinite in indirect speech, even though also other forms are valid. This suggests that speaker’s effort, and not only hearer’s processing cost, may be crucial for the choice of nominal form. The comparison of the four languages reveals that general cognitive constraints related to reference assignment interact with language-specific conditions; examples are constraints on discourse type and considerations of processing economy following from the language’s lexical and morpho-syntactic inventory.

Keywords

Indefinite expressions Specificity (In)direct speech Audience design Norwegian English Czech German 

1 Introduction

In dialogue situations, speakers have the choice of a variety of expressions when referring to relevant entities in the current situation or discourse universe. The choice of referring expression is not random but narrowed down by general constraints, which may interact among themselves as well as with language-specific properties. An example of a general constraint is that reference to a specific entity presupposes availability of the intended referent in the mental model of the hearer. A language-dependent property interacting with this constraint is the particular inventory of expressions available for performing the reference act (see e.g., Givón 1983; Ariel 1990; Gundel et al. 1993; Baumann et al. 2014). The general constraints examined in this paper relate in particular to questions of audience design, general processing and production economy, and effects of different discourse types.

We investigate constraints that play a central role in the choice of referring expression in narratives such as (1). Our main focus lies on the form of expression used in a speech report when reporting on a specific indefinite, be it by means of an indefinite noun phrase (NP), a definite NP, or a proper name. We are interested in whether and in which way the choice of referring expression in speech reports depends on the (in)directness of the report, as illustrated in (1a) and (1b).1

(1)

Over the weekend, Peter had to look after a parrot called Polly. Polly was green with some dark spots on her wings. On Sunday afternoon, Peter forgot to close the bird’s cage when he opened the window to let in some fresh air. Polly flew out of the window, and very soon, Peter did not see her anymore. He ran out on the street and asked every stranger he passed

a) “Have you seen ________?”

b) whether he had seen ________.

In these reports, we have portrayed a situation in which the protagonist of the story is looking for a specific referent that is unknown to the addressee. The entity being searched for in the original speech event is thus in privileged ground (part of the speaker’s knowledge), not in common ground in the sense of Clark (1992, 1996) and Clark and Marshall (1981). In situations/scenarios such as the one in (1), a cooperative speaker will have to take the availability or the non-availability of the referent in the addressee’s mind into account, hence performing audience design (Clark and Murphy 1982). Because of the type of event described in (1), and because of distributional constraints associated with various nominal forms, indefinites are predicted to be highly predominant in direct speech examples of the type in (1a) (see Sect. 2.2). Since direct and indirect speech differ with respect to whose perspective is taken into account (the original speaker/hearer or the present speaker/hearer), a further prediction is that the choice of referring expression is likely to differ in direct and indirect speech.

It has been proposed (e.g., Sæbø 2013) that a speech event that contains a specific indefinite can truthfully be reported on in indirect speech with what Sæbø calls ‘a referential term,’ for example a name or a definite description. With respect to (1), Sæbø predicts that if Peter asked the stranger (the addressee), “Have you seen a parrot?”, while looking for Polly, the indirect speech report in (1b) may be realized with a definite NP or a proper name referring to Polly. On the basis of Sæbø’s proposal, an increased occurrence of definite NPs and proper names is expected in narratives of the type in (1b) with indirect speech, compared to narratives of the type in (1a) with direct speech. Our main focus in this article will be the realization of NPs in indirect speech. A comparison with direct speech is necessary since such a comparison allows us to infer whether or not the NP realization is due to aspects of the original speech situation rather than the indirectness of the report. The role of indirect speech will however interact with other factors, which will be described below.

One constraint that has to be taken into consideration when looking at the use of referring expressions in direct and indirect speech is the fact that discourse type in general naturally guides the choice of expressions in languages. It is a plausible assumption that the distinction between direct and indirect speech in a narrative reflects the distinction of spoken versus written discourse. This means that if different nominal forms are predominant in one or the other discourse type, this may influence the choice of referring expression in direct and indirect speech as well. Whether or not such discourse-specific choices exist, and which forms are predominant in which case, differs across languages.

Language processing is generally guided by principles of economy (cf. the minimal everything principle, Inoue and Fodor 1995). We thus assume that the choice of a particular referring expression will be influenced by the respective complexity or length of the possible alternatives that exist in a certain language. More specifically, if two competing referring expressions are likely to yield equally satisfactory interpretations, it is expected that the one that requires least processing effort will be preferred (cf. Sperber and Wilson 1995). Again, which form is the most or least complex does not necessarily overlap across languages. Only cross-linguistic studies will allow us to shed light on the specific contribution of these constraints, which are tightly linked to the specificities of a particular language.

In this paper, we investigate how general constraints and language-specific properties interact when speakers select a nominal form in context, and how the selection is influenced by the distinction between direct versus indirect speech. The investigation is based on a cross-linguistic experimental approach, in which parallel pen-and-pencil questionnaires have been designed in Czech, German, English, and Norwegian. The difference between direct and indirect speech with respect to the narrative perspective is the same across the four languages. Furthermore, all languages can be assumed to follow the same general constraints on economy and audience design, and the four languages presumably exhibit some general correlations between the choice of referring expressions and discourse types.2 However, as we will see, differences in the languages’ structural properties and lexical inventory of referring expressions lead to differences in how the general constraints “surface.”

The structure of the paper is as follows: In Sect. 2, we present the distinction between direct and indirect speech, the notion of audience design, and linguistic properties of the four nominal forms that play a role in the experiment. In this section we also point out some important language-specific properties relating to noun phrase structure and determiner inventory. In Sect. 3, the hypotheses for the experiment are outlined, followed by the method and the experimental set-up in Sect. 4. The results are presented in Sect. 5, along with a follow-up study for Czech in Sect. 6. In Sect. 7, we discuss the results and consider some further theoretical and methodological issues relating to the study. Section 8 concludes the paper.

2 Background

2.1 Direct and Indirect Speech and the Four-Minds Model

The two sentences in (2a) and (2b) represent two different types of reported speech, that is, oratio recta (direct speech) and oratio obliqua (indirect speech).

(2)

a. Peter went straight up to Ann and said: “I love you!”

b. Peter went straight up to Ann and said he loved her.

As formulated in Coulmas (1986), direct speech evokes the original speech situation and conveys, or claims to convey, the exact words of the original speaker in direct discourse. Indirect speech, on the other hand, adapts the reported utterance to the present speech situation, that is, the given narrative situation. The main difference between the two lies in the perspective, or point of view, of the reporter. In direct speech the reporter lends his voice to the original speaker and repeats (more or less)3 what he said, thus adopting the original speaker’s point of view. In indirect speech, on the other hand, the reporter presents the speech event from his own point of view. This allows him to introduce information about the reported speech event on the basis of his own knowledge about the world.

Whereas languages differ with respect to the morpho-syntactic properties of indirect speech, most languages that have indirect speech share the same pronominalization strategies for the two constructions (see e.g., Li 1986; Güldemann and von Roncador 2002; Schlenker 2003; von Roncador 2010).4 In the case of simple reports (where only one verb of report is involved), one generalization that holds for the four object languages is that first- and second-person pronouns in indirect speech have to reflect the deictic center of the present speech situation, whereas first- and second-person pronouns in direct speech have to reflect the deictic center of the original speech event. Thus, the English pronoun ‘I’ and its oblique variant will always refer to the original speaker if part of simple direct speech, whereas the same forms in simple indirect speech will necessarily refer to the present speaker, here called the narrator. Figure 1 shows how the utterance “I love you” (as embedded in (2a)) may be reported on in direct and indirect speech.
Fig. 1

The four-minds model and the use of deictic pronouns

This figure illustrates how the narrator of the story has some original speech event in mind and how this original speech event may be reported on by the narrator, either through direct or indirect speech. The figure reflects the fact that there are at least four interlocutors that are potentially relevant in such reports. Since we will be dealing with written narrations in this paper, we label these four interlocutors ‘narrator,’ ‘reader,’ ‘original speaker,’ and ‘original hearer.’ As mentioned earlier, the main difference between direct and indirect speech is whether the point of view is that of the original speaker or the narrator.

The four languages under investigation, that is, Czech, English, German, and Norwegian, differ from each other with respect to some of the morpho-syntactic features of indirect speech. For instance, German and Czech, unlike English, have mood shifts in addition to regular tense shift in indirect speech. What is crucial to the present study is that all four languages share the property that in indirect speech the narrator is allowed to introduce information about the speech event on the basis of his or her own knowledge about the world, whereas this is not so in direct speech.5

In an indirect speech report, the possibility of using a nominal form that deviates from the nominal form used in the original speech event holds not only for pronouns, as illustrated in Fig. 1, but also optionally for other types of referring expression. An example from Coulmas (1986) illustrates this. Imagine that Oedipus said about his wife, who (unknown to him) was also his mother: “My wife is beautiful!” This utterance may be reported on as Oedipus said that his wife was beautiful or as Oedipus said that his mother was beautiful. In other words, indirect speech is quite flexible with respect to degrees of word-to-word iconicity with noun phrases in the original utterance.

A similar phenomenon is involved in our test items, illustrated in (1b) earlier, repeated here as (3):

(3)

Over the weekend, Peter had to look after a parrot called Polly. Polly was green with some dark spots on her wings. On Sunday afternoon, Peter forgot to close the bird’s cage when he opened the window to let in some fresh air. Polly flew out of the window, and very soon, Peter did not see her anymore. He ran out on the street and asked every stranger he passed whether he had seen a parrot/Polly/the parrot.

All three variants of (3) constitute coherent texts and truthful reports of a situation where Peter asked each stranger he met, “Have you seen a parrot?” Again, this illustrates the flexibility of noun phrase realization in indirect speech.

The fact that an indefinite description (here ‘a parrot’) can sometimes be reported on in indirect speech with a proper name or a definite description is captured by the ‘referential report’ postulate of Sæbø (2013, pp. 270–271):

(4)

Referential Report

Consider an utterance of a sentence [a P] Q.6

[a P] can be rendered by a referential term u (a name, a personal or demonstrative pronoun, or a definite or demonstrative description) in a report of this utterance if and only if either

a) the reporter has background knowledge that the referent of u is the only entity that satisfies P and/or Q, or

b) the original speaker had the referent of u in mind when uttering [a P] but the referent of u was not identifiable to the hearer.

Beware that the term ‘report’ in (4) is intended to mean indirect speech report. Given this, the postulate says that an indefinite NP that occurs in some original speech event can be reported on in indirect speech in terms of a referential term, for example, a definite description. The indefinite description referred to in (4) is according to many theories on specificity a specific indefinite; the speaker uses an indefinite description while having a particular referent in mind, because of the hearer’s lack of knowledge of the referent (see e.g., Ioup 1977; Fodor and Sag 1982; Farkas 1994). With this definition in mind, what the postulate in (4) says is that a specific indefinite (that occurred in some original speech event) may be reported on with a referential term in an indirect speech report. This predicts that definites and proper names may be used in (3) when reporting on the question “Have you seen a parrot?”. Like Coulmas, Sæbø is concerned with indirect speech reports where the narrator adds information that was not part of the original utterance. The difference between Sæbø and Coulmas’ work on indirect speech, is that Coulmas describes cases where the discrepancy between the indirect report and the original utterance concerns knowledge not shared between the original speaker and the narrator, whereas Sæbø describes cases where the narrator adds information that was not part of the original utterance because of the original hearer’s lack of knowledge about the referent.

Example (3) illustrates that a specific indefinite can be reported on with various nominal forms in indirect speech. A report on a specific indefinite in direct speech, on the other hand, will necessarily involve an indefinite expression in order to do justice to the wording used by the original speaker. The various possibilities of reporting on specific indefinites in direct and indirect speech in examples such as (3) are summarized in Fig. 2.
Fig. 2

Direct and indirect speech reports on specific indefinites

This figure reflects how direct and indirect speech contexts are assumed to differ with respect to which nominal forms are plausible reports on specific indefinites.7 While these assumptions seem intuitively correct, they need to be tested experimentally. For indirect speech in particular, where the choice of referring expression appears to be more open, there are to our knowledge no previous studies investigating which of the possible alternatives is preferred. Giving empirical evidence for the basic differences between direct and indirect speech with respect to the choice of referring expression, as well as testing general preferences concerning choice of nominal form, will be the most relevant outcome from our experiment.

2.2 Four Nominal Forms

In the main experiment, we gave our informants a choice of four different forms of referring expression to fill in for an open slot in a story of the type illustrated in (1) and (3). The forms are (i) proper names, (ii) definite NPs, (iii) possessive NPs, and (iv) indefinite NPs. In the present section, we present constraints that are assumed to hold for the four alternatives, according to current theories. While this section presents restrictions on the four designated English nominal forms, we will point out some relevant cross-linguistic differences among the four languages in Sect. 2.4.

Choice of nominal form is assumed to be crucially addressee-oriented. According to certain authors (e.g., Haviland and Clark 1974; Ariel 1990, 2001; Gundel et al. 1993, 2010), nominal forms contribute processing signals to the addressee that narrow down the set of possible referent candidates, thereby enhancing the ease of successful reference resolution. According to Gundel et al. (1993), unaccented personal pronouns in English, for instance, have as part of their encoded meaning a procedural instruction, namely that the addressee can expect to find a representation of the intended referent among a subset of the entities represented in her current short-term memory, that is, among those entities that are in her current focus of attention. Obviously, the speaker too needs to have a representation of the referent in working memory to be able to refer to it, but this is not a sufficient condition for felicitous use of a pronoun. The relevant constraint concerns the state of the situation model of the addressee (or the audience more generally).

Experiments have shown that speakers do not always perform audience design in this sense, but let their own mental state be reflected instead, in particular in case of cognitive pressure (see Arnold et al. 2004; Lane and Ferreira 2008). Notably, predictions from Gundel et al. (1993) theory and related ones require that the speaker performs audience design. According to these theories, lack of audience design in NP production is expected to lead to infelicity and processing difficulties.

In Sect. 2.1 it was argued that the difference between direct and indirect speech amounts to a difference in whether to reflect the perspective of the present speaker (the narrator) or the original speaker. Given that the choice of nominal form is strongly addressee-oriented, it may sound counter-intuitive that the difference between direct and indirect speech is expected to correlate with differences in the choice of nominal forms. This is only a seeming contradiction, though. The interdependency of direct/indirect speech and nominal form follows because nominal form reflects the speaker’s assumptions about his addressee’s mental model. This means that when the perspective changes from one speaker to another, this simultaneously leads to a change of perspective from one addressee to another, for example from the original hearer to the present reader, whose representations of the world are not the same.

Below, we outline conditions expected to affect the choice of nominal form in our experiment.

Proper names . According to Mulkern (1996), a simple proper name, such as Mr. Simpson or Peter, provides a processing signal that the intended referent is already familiar to the addressee (in the sense of Gundel et al. 1993). That is, for a simple proper name to be used appropriately, the addressee has to already have a mental representation of the referent available in either long- or short-term memory.8 From the hearer’s point of view, this processing signal helps to narrow down possible referent candidates to ones that she already has a representation of. In all the stories used in our experiment, the original hearer is a stranger to the original speaker, which makes it very unlikely that the original hearer already possesses a mental representation of the intended referent. Since the nominal form in direct speech is supposed to reflect the original speaker’s assumptions about the mental state of the original hearer, Mulkern predicts that simple proper names are disfavored in these stories with direct speech. That this is indeed the case can be observed in (5) when a simple proper name is inserted for the open slot.

(5)

Paula, a nurse in a geriatric ward, was worried because earlier that day, a dementia patient, Mr. Simpson, had evidently decided to go off on a walk on his own. He was nowhere to be found in the home. Paula rushed out on the street and asked the first person she came across: “Excuse me, you haven’t seen ________ around here, have you?”

an elderly man; the elderly man; my elderly man; Mr. Simpson

The use of a simple proper name is intuitively dispreferred in the given narrative, the reason being that the stranger is not likely to know who Mr. Simpson is. The unacceptability of a proper name in this context is exactly what Mulkern’s theory on proper names predicts.

When the conversation between Paula and the stranger is reported on with indirect speech, on the other hand, use of a proper name is intuitively much better. This is illustrated in (6).

(6)

Paula, a nurse in a geriatric ward, was worried because earlier that day, a dementia patient, Mr. Simpson, had evidently decided to go off on a walk on his own. He was nowhere to be found in the home. Paula rushed out on the street and asked the first person she came across whether she had seen ________ around there.

an elderly man; the elderly man; her elderly man; Mr. Simpson

In this case the speech report is indirect, which means that the narrator can reflect his own perspective, as opposed to that of the original speaker. He can therefore take the mental model of the present reader into account when selecting a nominal form. The reader in (6) has a representation of Mr. Simpson in short-term memory, because of prior reference, and is therefore familiar with the referent. Mulkern’s prediction is therefore that use of a simple proper name in reference to Mr. Simpson should be possible, which indeed seems to be the case.

Definites. According to Gundel et al. (1993), use of the definite article in English is felicitous if the descriptive content of the noun phrase enables the addressee to establish a unique representation of the referent, either because it is already represented in the addressee’s memory, or because the addressee can create a unique representation on the spot on the basis of the description in the phrase. With this in mind, consider the direct speech example in (5) once again, in which case the nominal form must reflect the mental state of the original hearer (i.e., the stranger on the street). The definite description ‘the elderly man’ in the question “Have you seen the elderly man?” is presumably not rich enough for the original hearer to create a unique representation of the intended referent, as there may be many elderly men that the addressee can think of. The definite noun phrase ‘the elderly man’ is therefore not expected to be acceptable in this case.9

When the story is presented with indirect speech, as in (6) above, the perspective changes to that of the narrator and the reader—in which case the nominal form does not have to reflect the mental state of the original hearer. In this case, Gundel et al. (1993) prediction is that use of the definite form should be felicitous. The reader of the story already has a mental representation of the referent, because of previous mention, and is therefore able to uniquely identify it on the basis of the description ‘the elderly man.’ The prediction is that the definite noun phrase should be acceptable in reference to Mr. Simpson, which seems intuitively correct.

Possessives. Possessives such as ‘my old man’ were also included among the four NP choices in our experiment. Possessives seem to undertake similar restrictions as definite descriptions. That is, also for possessives it can be expected that the addressee should be able to associate a unique representation of the referent. What makes possessives different from definite descriptions is that a representation of the possessor must be easily accessible, and this possessor is in fact a very useful processing cue in the determination of the unique referent for the whole phrase. A phrase of the type ‘my N’ in a direct speech situation will therefore often enable the addressee (as long as the possessor is known) to accommodate a unique representation without prior knowledge of the referent. However, in (5) and (6) above, and in the majority of our test items,10 there is no natural possessor relation between the original speaker and the entity (s)he is looking for. Since there is no natural possessor relation between Paula and Mr. Simpson in (5) and (6), the prediction is that it will not be felicitous to refer to Mr. Simpson with a possessive NP when Paula is the intended possessor. Considering the intuitive infelicity of possessive NPs in (5) and (6), this prediction is borne out.

Indefinites. Indefinite descriptions typically introduce new discourse referents into the discourse (see e.g., Prince 1981; Heim 1982; Gundel et al. 1993; Abbot 2004). That is, indefinites typically refer to entities that are not part of the interlocutors’ common ground and that cannot be uniquely identified on the basis of the descriptive content of the phrase. In direct speech examples such as (5), where the nominal form has to reflect the mental model of the original hearer, the prediction is that indefinites will be perfectly acceptable since the original speaker introduces a referent that is new to the original hearer.

Also in indirect speech examples such as (6), the prediction is that indefinites ought to be acceptable. There are at least two possible lines of arguments that lead to the conclusion that indefinites are expected to be acceptable. On one view, indefinites may in principle refer to entities that are uniquely identifiable, familiar, activated, or even in focus of attention, but less naturally so than corresponding definite expressions because such uses involve breaking Grice’s first quantity maxim, that is, to be as informative as required. This view is held by Gundel et al. (1993), for instance. On such a view, the indefinite NP an elderly man in (6) is seen as coreferential with the previous NPs referring to Mr. Simpson. Another view is to assume that the indefinite an elderly man, if inserted for the open slot in (6), is not used to refer to Mr. Simpson; rather, it is used to introduce a new discourse referent—some elderly man or other. As a second step of inference, the reader may (and most likely will) infer that the original speaker was looking for Mr. Simpson while asking his question. Thus, although a new discourse referent is introduced when the indefinite NP is processed, the discourse model is updated later, in the sense that the two discourse entities are identified with each other. Notably, this is a type of pragmatic inference different from that of coreference (see Borthen 2004 for more arguments). In this approach, it is felicitous to use an indefinite NP in examples such as (6) because the indefinite NP is used to introduce a new discourse referent to the reader, just as expected. For our discussion it is not important to decide which one of these two approaches is more plausible; what is crucial is that both views predict that an elderly man should be a possible NP alternative in indirect speech examples such as (6), just as well as in direct speech. That this prediction is plausible is supported by the intuition that an elderly man is a perfectly acceptable noun phrase in both (5) and (6). We will discuss this point further in Sect. 2.3 below.

In sum, this section has provided independent motivations for the ‘referential report hypothesis’ of Sæbø (2013), that is, the postulate that specific indefinites can be reported on by referential terms in indirect speech reports. On the basis of the theoretical assumptions presented in this section, the prediction is that possessives will not be felicitous in any of the two conditions tested in our experiment (direct vs. indirect speech), as long as there is no natural possessor relation between the protagonist of the story and the entity (s)he is looking for. The prediction is furthermore a clear preference for indefinites in direct speech examples such as (5), whereas indirect speech examples such as (6) are expected to allow for definites and proper names as well as indefinites. In our experiment, we test experimentally how far these predictions hold.

2.3 Two Perspectives in Indirect Speech

In the previous section, we argued that the mental model of the addressee is a crucial factor for the speaker’s choice of referring expression. But what, then, determines the choice when more than one form is acceptable from the point of view of the addressee? Consider once again the difference between (7a) and (7b) below:

(7)

Over the weekend, Peter had to look after a parrot called Polly. Polly was green with some dark spots on her wings. On Sunday afternoon, Peter forgot to close the bird’s cage when he opened the window to let in some fresh air. Polly flew out of the window, and very soon, Peter did not see her anymore. He ran out on the street and asked every stranger he passed

a) whether he had seen a parrot.

b) whether he had seen the parrot/Polly.

The three choices of nominal form in the indirect reports in (7) are all in accordance with the reader’s mental model; (s)he will have no problem interpreting either of these forms as intended by the narrator. One difference between (7a) and (7b) has to do with what aspects of the original speech event the narrator chooses to focus on. As argued above, the situation in the story is such that Peter most likely used an indefinite expression when he asked his question. Thus, with an indefinite NP, as in (7a), the report is most likely closer to, and more iconic with, the spoken words of the original utterance than the alternatives in (7b). Let’s call the choice of an indefinite NP in this situation ‘the iconic approach,’ bearing in mind that this nominal form reflects the original speaker’s assumption about the mental model of the original hearer.11

When a proper name or a definite description is used, as in (7b), the report is most likely less faithful to the words uttered in the original speech event, since Polly or the parrot would hardly be used in a question to a stranger with no previous knowledge of Polly. By reporting on the speech event with either of these referential terms, the narrator not only pays attention to the present communication situation, in which the reader is already familiar with Polly; he also adopts the perspective of the protagonist/the original speaker, since the original speaker had Polly in mind when asking his question. Let’s call this ‘the narrative approach.’ The two approaches are illustrated in Fig. 3 below.
Fig. 3

Two perspectives in indirect speech

This figure pinpoints the main difference between ‘the narrative approach,’ which emphasizes the present reader’s (and the original speaker’s) mental model, and ‘the iconic approach,’ which emphasizes the mental model of the original hearer. Both perspectives in Fig. 3 are compatible with the mental models of the narrator and the reader of the story, in the sense that they can produce/interpret the various nominal forms without problems. As for which approach will be the preferred one, there is more than one plausible hypothesis.

From the point of view of NP processing per se (abstracting away from the fact that we are dealing with speech reports), the narrative approach may be considered the less costly one with respect to cognitive effort. Since Polly has already been introduced into the story, reference to this accessible referent by way of a definite description or a proper name may be expected to be less cognitively demanding for the reader, since given discourse entities are not typically referred to by indefinite NPs (cf. Heim 1982; Ariel 1990; Gundel et al. 1993). Thus, considerations of NP processing alone would count in favor of a preference for definites and proper names as opposed to indefinites in our indirect speech examples. Similar predictions follow from Heim’s (1991) communicative principle maximize presuppositions, which is made use of in theoretical semantic literature (see e.g., Schlenker 2003; Sauerland 2008; Percus 2006).

However, if we take into consideration the effort involved in producing speech reports more generally, a different conclusion might be drawn. Bear in mind that an indirect speech report requires the narrator to (1) hold a representation of the communicated meaning of some original utterance, and (2) integrate this representation with the ongoing discourse, and produce the speech report accordingly. An “echo” of what the original speaker said allows the narrator to, grossly speaking, copy the representation he has of the speech event (or some aspects of it). The narrative approach, on the other hand, where the NP form is adapted to the present narrative situation, would require more contextual processing on behalf of the narrator, linking the indirect speech report cohesively to the previous discourse and take the reader’s perspective into account. With several persons introduced in the discourse, the narrator has to check the correspondence of a proper name with the one given in the discourse, or check that the correct link is made between a definite description and its antecedent. By choosing an indefinite NP, on the other hand (i.e., the ‘iconic approach’), the narrator can leave it to the reader to make the connection, if any. Thus, from the point of view of context integration and considerations of the narrator’s production effort, one can argue that the prediction is a preference for indefinites in our indirect speech examples.

While Sæbø (2013) pointed out the possibility of reporting on specific indefinites with either indefinites or referential terms in indirect speech reports, our experiment is designed to test the extent to which informants choose to use one form or the other. While assumptions about referent accessibility and the reader’s cognitive effort per se lead to the expectation that referential terms (definites and proper names) will be the preferred choice, assumptions about more general context integration of speech reports lead to the prediction that indefinites will be favored.

2.4 Cross-Linguistic Differences

As mentioned in Sect. 1, the experiment was run on the four languages Czech, English, German, and Norwegian. In most respects, we claim that the four languages possess a comparable lexical inventory of determiners and other nominal categories, and that they have a comparable nominal syntax. Similarities and some relevant differences are shown in Table 1 above. Proper names are not represented in the table.
Table 1

Comparison of determiners and noun phrase structure

 

Indefinite

Definite

Possessive

Eng

a white cat

the white cat

my white cat

Ger

eine weiße Katze

die weiße Katze

meine weiße Katze

Nor

en hvit katt

den hvite katten

den hvite katten min

a white.strong cat

the white.weak cat.def

the white.weak cat.def mine

Cze

nějaká bílá kočka

ta bílá kočka

moje bílá kočka

Indef.Deter. white cat

Def.Deter. white cat

possessive white cat

In Norwegian, a definite noun phrase is formed through a definite suffix on the noun as long as there is no premodifying adjective. However, when the noun is premodified, as in the example in Table 1, a preposed definite article usually appears before the adjective while the definite suffix is kept on the noun.12 This means that the definite version of a Norwegian premodified noun phrase can be seen as morpho-syntactically slightly more complex than the minimally distinct indefinite variant. A corpus search at http://www.tekstlab.uio.no/nps (see Borthen et al. 2008) shows that preposed definite articles in Norwegian are surprisingly infrequent, constituting less than 1 % of all noun phrases in this corpus. Furthermore, Norwegian preposed definite determiners have been found to be acquired by children later than the definite suffix and after premodifying adjectives start to appear (Anderssen 2007). Given that a property of language processing in general is least possible effort (cf. the minimal everything principle, Inoue and Fodor 1995), it can be expected that the relative complexity of premodified Norwegian definites leads to a preference for corresponding indefinites when indefinite and definite forms are otherwise equally plausible. No such asymmetry is expected for the other languages under investigation.13

The Czech determiner system differs in crucial respects from the Germanic languages under analysis, since a noun phrase in Czech often takes the form of a bare noun, whether a definite or an indefinite interpretation is intended. There is no obligatory indefinite article in Czech, but other indefinite determiners may occur. The Czech parallel to the indefinite alternative in our test material involves an indefinite non-obligatory determiner, nějaký. This determiner can be found in contexts where the existential quantifier some would have been used in English, in the function of marking non-generic indefinite reference.

We also need to pay some special attention to the Czech demonstrative determiner ten, which constitutes the definite alternative in the Czech data. It has been argued in previous studies that the demonstrative ten seems to have an article-like function (see e.g., Mathesius 1926; Berger 1993). One fact that counts in favor of this is that ten is neutral with respect to the feature ± distal, and as such differs from other demonstratives (cf. Meyerstein 1972; Komárek 1978). Some authors (e.g., Hlavsa 1975, p. 53; Hammer 1986) thus describe ten as a translation equivalent to the English definite article the. In our view, however, the Czech determiner ten is not a true equivalent to the definite article. It is rarely used anaphorically and it is not obligatory for interpretations normally associated with definiteness, as a bare noun can be used instead. Furthermore, it cannot be employed for generic interpretations (cf. Zubatý 1917; Krámský 1963).

According to Mathesius (1926), two specific functions of ten are the ‘reminding’ and the ‘accentuating’ functions. In the accentuating function, ten is used to express elevated emotionality of the speaker, or it can be used when speakers wish to intensify a particular content. In the reminding function, on the other hand, the speaker can refer to an entity that is already known, either objectively (to speaker and hearer) or subjectively (to speaker). An example from Berger (1993) that illustrates this function is Tak do té Prahy jsem nakonec nejel (‘So in the end I did not get to (that) Prague’). Also Berger (1993) mentions, among the various uses of non-anaphoric ten in spoken language, that it can be used in a pseudo-anaphorical way, in which case the use of it suggests that the hearer and the speaker have common knowledge about the entity in question. This is probably the same ‘reminding’ function as Mathesius mentions. In addition to these properties of ten, one should be aware that ten is much more frequent in spoken discourse than in written narratives, as shown in a recently published corpus-based grammar of the Czech language (cf. Cvrček et al. 2010).14 This property of ten, as well as its reminding function, will turn out to be of importance in our experiment.

In sum, although ten is not a definite article, it is a definite determiner in the sense that it seems to require that the referent be at least uniquely identifiable to the addressee (cf. the definition of ‘definite’ in Gundel et al. 1993). In this sense, ten is comparable to the definite articles in the Germanic languages. On the other hand, this determiner differs from the Germanic definite articles in that (a) it is used most frequently in spoken discourse, (b) it is rarely used anaphorically, and (c) one of its prominent functions is a ‘reminding’ function. Reminding someone of something presumably presupposes familiarity (i.e., it has been processed earlier). If correct, this means that this use of ten narrows down possible referent candidates to ones that are already familiar to the addressee. This is information that may be very useful for the addressee for the purpose of identifying the intended referent. For instance, there are many Chinese girls in the world, so you would not necessarily know who the intended referent of the phrase the Chinese girl is. But if the speaker uses a form that signals that you are already familiar with the intended referent (e.g., that Chinese girl you know), you may infer who the speaker has in mind. Following this line of reasoning, the prediction is that ten (when having a reminding function) may be used in cases where Germanic definite articles would not necessarily be appropriate. On the other hand, the dispreference for ten in written discourse and the constraint against anaphoric use raise the expectation that ten will be infelicitous in many cases where definite articles are preferred in the Germanic languages.

In sum, the language-specific properties of Norwegian and Czech discussed in this section lead to the expectation that we may find deviances between the use of definites in these languages compared to English and German.

3 Hypotheses

We started this article by introducing Sæbø’s ‘referential report hypothesis’ in Sect. 2.1, which says that referential terms can be used to report on specific indefinites in indirect speech. This hypothesis was further motivated in Sect. 2.2 where we looked at constraints associated with the various nominal forms that constitute the NP alternatives in our experiment. From this discussion we can formulate the following hypothesis for the test items exemplified in (1a–b) and (5) and (6) above:

Hypothesis 1

Definites and proper names will be chosen more frequently in indirect speech than in direct speech across the four languages.

Direct and indirect speech reports are assumed to differ in terms of which speaker’s point of view is presented, that is, the narrator or the original speaker. Choice of nominal form depends on the speaker’s assumptions about the hearer’s mental representation of the discourse. When the perspective changes from one speaker to another, this leads to a change in whose mind (which hearer) is taken into account, and consequently a difference in what referring expressions are acceptable. In our direct speech examples, the original hearer is most likely not familiar with, and cannot uniquely identify, the referent, which disfavors definites and proper names. In our indirect speech examples, on the other hand, the reader already has a representation of the given referent in short-term memory, which predicts that definites and proper names should be possible. Since this option is not available in direct speech, Hypothesis 1 follows as a consequence.

In order to test Hypothesis 1, the following two subquestions have to be answered:

Subquestions for Hypothesis 1

a) What is the distribution of nominal forms in the direct speech examples in each of the four languages?

b) What is the distribution of nominal forms in the indirect speech examples in each of the four languages?

On the basis of the theoretical assumptions presented in Sect. 2.2, the working hypothesis for Subquestion 1a is that there will be a clear preference for indefinites in our direct speech examples in all four languages, because the original hearer is most likely not familiar with, and cannot uniquely identify, the referent that the speaker is looking for. This hypothesis presupposes that the narrator performs audience design on behalf of the original speaker. Subquestion 1b is more of an open question, since the designated addressee in indirect speech, that is, the reader, can be expected to be able to properly interpret both indefinites and referential terms (here, definites or proper names). As argued in Sect. 2.3, the prediction based on referent accessibility alone is that referential terms will be preferred. This is what we called ‘the narrative approach’ in Sect. 2.3. The prediction based on more general context integration involved in speech reports, on the other hand, suggests a preference for indefinites (‘the iconic approach’): Assuming that the original speech situation is part of the mental representation of the narrator, it might be considered less cognitively demanding to simply “copy” and reproduce a representation of a speech event than to integrate and adapt it cohesively to the present narrative situation. We should note here that a preference for the iconic approach may actually counteract the choice of definite expressions in indirect speech and thus the prediction expressed in Hypothesis 1.

We saw in Sect. 2.4 that there are differences among the four languages with respect to the determiner inventory. The definite determiner in Czech (i.e., the demonstrative determiner ten) is mostly used in spoken discourse with a non-anaphoric function. Since a definite noun phrase in our indirect speech examples will necessarily be interpreted anaphorically and be part of written discourse, this raises the expectation that the Czech definite alternative will be less frequent in indirect speech than the Germanic definite alternatives, other things being equal. This is reflected in Hypothesis 2:

Hypothesis 2

In the indirect speech examples, definites will be chosen to a greater extent in English and German than in Czech.

Norwegian, too, exhibits properties that may affect a free choice between an indefinite and a definite expression, that is, the facts concerning complexity discussed in Sect. 2.4. As shown there, Norwegian pre-modified definite descriptions may be seen as morpho-syntactically somewhat more complex than the minimally distinct indefinite phrases. This means that we might expect Norwegian subjects to favor premodified indefinite descriptions to minimally distinct definite ones when both yield equally satisfactory interpretations. This expectation is represented as Hypothesis 3:

Hypothesis 3

In the indirect speech examples, definites will be chosen more frequently in English and German than in Norwegian.

The difference in complexity between definite and indefinite pre-modified NPs in Norwegian is very small, though, so we consider this a weaker hypothesis than Hypothesis 2 above.

It follows from Hypotheses 2 and 3 that we expect both Norwegian and Czech definites to be less frequent in indirect speech than English and German ones. However, although Norwegian and Czech may show superficial similarities, the reasons behind the similarities are not the same, thus the two separate hypotheses above.

There is an important inter-dependency between Hypotheses 2 and 3 on the one hand and Hypothesis 1 on the other. If constraints related to discourse type, discourse function, and processing economy are stronger determinants of nominal form in Czech and Norwegian than those that count in favor of ‘the narrative approach’ and thus the choice of a definite description (cf. Sect. 2.3), then Hypothesis 1 is not expected to hold for Norwegian and Czech, only for English and German. On the other hand, if discourse constraints, discourse function, and processing economy are weaker determinants of nominal form than those that count in favor of ‘the narrative approach,’ then Hypothesis 1 is expected to hold for all four languages. If, finally, a preference for iconicity determines the choice of referring expressions in general, we will have to expect Hypothesis 1 to be disconfirmed across all languages under investigation with a general strong preference for indefinites.

4 Method and Experimental Set-Up

The experiment was set up as a pen-and-pencil questionnaire in which 30 or more native speaker informants from each of the four languages Czech, English, German, and Norwegian participated. The informants were asked to read short narratives of the kind in (5) and (6) (see Sect. 2.2), and to mark their preference for one out of the four pre-given nominal phrases to fill in for the open slot in the story. The subjects were informed that we wanted to know what their feel for language intuitively told them to be the best alternative. They were instructed to read the story, choose from the given alternatives and underline their choice, go to the next item and not return to any item to make revisions or corrections. The instructions were kept equivalent across all four languages.

The narratives were set up in two ways, as we have already seen, one ending in a direct report, as in (5), the other as an indirect report, as in (6). In all the stories, the entity ultimately referred to by the target NP, was mentioned at least three times earlier in the story, both with a description and a proper name, as illustrated in (1) and (5)–(6).

The experiment was presented in six different (randomized) lists, each containing 16 critical items with three training items and 33 fillers. No subject was exposed to more than one item of each minimal pair (with either direct or indirect speech), and the order of the NP alternatives was altered across items. For cross-linguistic validity and variation, the test-narratives were constructed in parallel across our four object languages.

5 Results

We will present our results in two ways according to our research questions. We will first compare the four languages under investigation with respect to the distribution of referring expressions separately for direct and indirect speech. For these analyses we used frequencies of choice as the dependent variable. The experimental factors were Language with the four conditions Czech, English, German, and Norwegian as well as Referring Expression with the conditions ‘definite,’ ‘indefinite,’ ‘possessive,’ and ‘proper name.’ Language was realized as a between participants factor, Referring Expression as a within participants factor. For the comparison of choices, we analyzed log-odds of choices for each referring expression in direct and indirect speech following the formula in (8). The study included 134 speakers across languages, with 38 Czech speakers, 30 English speakers, 36 German speakers, and 30 Norwegian speakers.

(8)

logoddsdir/indir = log2(p(dir)/p(indir))

Analyses of variance were calculated for both independent variables generalizing across participants as well as across items.

5.1 Choices of Expressions in Direct Speech

Figure 4 shows the frequencies of the individual forms of referring expressions for the different languages in direct speech contexts. We established a reliable main effect of Referring Expression (Fs) but no main effect of Language. As predicted, participants chose indefinite expressions significantly more often than any other expression across languages (Fs). Unexpectedly, though, proper names were used reliably more frequently than possessives (Fs), although proper names were predicted to be infelicitous in direct speech. As shown in the graph, Czech speakers use more definites in direct speech than Germanic speakers. This can be related to the ‘reminding’ function of Czech ten, as well as to the tendency for ten to be used in spoken discourse.
Fig. 4

Frequencies of referring expressions in direct speech in Czech, English, German, and Norwegian

5.2 Choices of Expressions in Indirect Speech

For indirect speech, we found a main effect of Referring Expression just as for direct speech (see Fig. 5), but no main effect of Language. Indefinites were used more often than any other expression (Fs). Proper names were used reliably more often than possessives (Fs). German and English speakers chose more definite expressions than Czech and Norwegian speakers (Fs). The dispreference for definites in indirect speech for Czech speakers was predicted from the fact that the Czech determiner ten is less felicitous in written language and rarely used anaphorically. For Norwegian, the dispreference for definites in indirect speech is predicted from the slightly higher complexity of definites in comparison to indefinites (more on this in Sect. 7).
Fig. 5

Frequencies of referring expressions in indirect speech in Czech, English, German, and Norwegian

5.3 Comparing Choices of Referring Expressions in Direct and Indirect Speech

The log-odds analysis will allow a more direct comparison of the use of referring expressions across languages. In this analysis, a main effect of Language was found (Fs), but no main effect of Referring Expression. We did, however, establish a reliable interaction of Language and Referring Expression (Fs). This interaction results from the fact that while definites were used more frequently in indirect speech in German and English than in Czech and Norwegian (Fs), definites were used relatively more frequently in direct speech than in indirect speech in Czech. Possibly to compensate for the dispreference of definites in indirect speech in Czech, proper names were used more often in indirect speech compared to direct speech in Czech, while at least numerically the opposite was the case in English and German (Fs). Norwegian did not show any differences with respect to the choice of referring expressions across the two conditions (Fig. 6).
Fig. 6

Log-odds of direct over indirect speech for referring expressions in Czech, English, German, and Norwegian. Positive values correspond to a higher number of choices in direct speech, negative values correspond to a higher number of choices in indirect speech

6 Follow-Up Experiment for Czech

For cross-linguistic comparability reasons, a bare noun was not included as an option in our core experiment. Since a bare noun is definitely a qualified option in Czech, we asked 25 Czech native speakers for acceptability judgments of bare nouns and the definite determiner ten in direct and indirect speech. We did this in order to test whether ten is a relevant referring expression in the Czech examples.

All participants in this follow-up experiment were students at the University of South Bohemia in Budweis, Czech Republic. To make sure that this participant group was comparable to those taking part in the core experiment, none of them studied linguistics. There was an equal amount of male and female participants, and the average age was 23.7 years. Four scenarios, both in the direct and in the indirect condition, were selected from the main experiment and were presented to the participants in a randomized order in a within-subject design as a paper questionnaire. The participants were asked to judge the acceptability of the 24 items (four scenarios in either direct or indirect speech, each realized with three different NPs) using a rating scale from 1 (fully acceptable) to 8 (totally unacceptable). The three NP choices were bare nouns, definites (ten), and possessives. All participants had to judge all three referring expressions for each of the four items.

We ran an analysis of variance with the between-participant factors Linguistic Context (direct vs. indirect speech) and Referring Expression (bare noun, definite ten, and possessive). The results showed the following pattern: A reliable main effect of both experimental factors was established.15 Whereas possessives were less acceptable than bare nouns and definites in both contexts,16 bare nouns and definites did not differ in direct speech. There was, however, a reliable difference between these two options in the indirect reports where the Czech participants showed a preference for the use of bare nouns compared to the definites.17 Nevertheless, the use of the determiner ten was not completely impossible in the indirect reports; they were judged reliably better than possessives.18 As expected, possessives did not get a good score in either of the two conditions (Fig. 7).
Fig. 7

Mean judgments for bare nouns, definite ten, and possessives in direct and indirect speech

The follow-up study demonstrated that bare nouns can be used to refer to entities in both direct and indirect speech examples of the type in (5) and (6) in Czech. However, when bare nouns are available as an option, this does not lead to a complete disappearance of the definite determiner. On the contrary, both categories are equally suitable options for direct speech contexts. For indirect reports, bare nouns represent the preferred option (presumably because of the already stated dispreference for ten in written language and with anaphoric function); still, ten is not completely excluded in this context. These results show that the fact that the core experiment did not include bare nouns as an option did not skew the results for the Czech participants.

7 Discussion

This experiment has revealed results on how specific indefinites are reported on in direct and indirect speech reports of the type in (5) and (6) in the four languages Czech, English, German, and Norwegian.

7.1 Direct Speech

As shown in Fig. 4 in Sect. 5, indefinites are clearly preferred in all four languages in direct speech. This result is in accordance with the predictions on the relevant nominal forms presented in Sect. 2.2. Since direct speech reflects the original speech situation, in which the original hearer is most likely not able to uniquely identify the referent the speaker has in mind, indefinites are expected to be clearly predominant. Since the relevant referent is not in the interlocutors’ common ground in these cases, our study contributes to the investigation of privileged ground (see e.g., Keysar et al. 2000; Hanna et al. 2003; Barr 2008; Brown-Schmidt and Hanna 2011). The preference for indefinites in direct speech shows that the informants are aware of the original hearer’s lack of knowledge and produce a referring expression accordingly. The results are particularly interesting because the informants perform a complex version of audience design, that is, what might be called “embedded audience design”: On behalf of the narrator, they take the original speaker’s assumptions about the mental state of the original hearer into account when choosing a referring expression.

In some cases, however, indefinites are not chosen in direct speech, which is not directly predicted in Sect. 2.2. That is, there are some occurrences of definites, proper names, and possessives in the direct speech examples. The occurrence of these expressions in this context could in principle be due to the informants failing to take the original hearer’s perspective into account. It is assumed in the psycholinguistic literature that taking another person’s perspective is resource intensive (see e.g., Keysar et al. 2000), presumably even more so when performing embedded audience design as described above. Furthermore, it has been shown that partner-specific interpretation is most likely to occur in interactive dialogue settings (cf. Brown-Schmidt 2009), which was not the setting in our experiment. In light of this, the occurrences of definites and proper names in our direct speech examples could be attributed to the informants failing to perform audience design on behalf of the narrator.

However, a more plausible explanation in our view is that these occurrences are related to the degree to which we succeeded in designing the narratives as intended. We intended to design narratives with the following three properties:

(9)

 

1) The original hearer should not already be familiar with the referent (=> dispreference for proper names and definites in direct speech).

2) The original hearer should not be able to create a unique representation of the referent on the basis of the descriptive content of the phrase (=> dispreference for definites in direct speech).

3) The protagonist should not stand in a possessor relation with the entity that (s)he looked for (=> dispreference for possessives in both conditions).

In some of the stories, we failed on one or more of these intentions, and given this, the occurrences of some referential terms is not surprising. An example text that illustrates this is given in (10).

(10)

Upon entering the Pacific, the whaling ship Pequod and her master, Captain Ahab, were finally closing in on the white whale known to him as Moby Dick. Every whaling ship they met Ahab hailed with the following words: “Have you seen ___________?”

my white whale, the white whale, a white whale, Moby Dick

For this item, definites and proper names were chosen slightly more often than indefinites across the four languages, even though the target NP occurs in direct speech. In retrospect, this is not surprising: It is perfectly plausible that the white whale that Captain Ahab was hunting was well-known to the whaling hunters that he met, which means that both the definite description and the proper name could be used. A few informants even chose the possessive variant. This can be attributed to the fact that Captain Ahab was strongly obsessed with Moby Dick and did not want anyone else to catch him.

A general insight to be drawn from this is that although context includes previous discourse, it is not determined by it, as context is partly created by the interlocutors as processing takes place (see e.g., Sperber and Wilson 1995). Given that context is not a static feature, it is not surprising that informants may imagine different contexts than the ones intended by the researchers. Notice, however, that the occurrence of some referential terms in direct speech is of no obstacle to our main investigation, which is the comparison of NP realization in direct versus indirect speech.

Another insight to be drawn from Fig. 4 is that definite noun phrases are chosen significantly more often for Czech in direct speech than for the other languages. We can explain this, since the Czech determiner ten is typically used in spoken discourse and has a reminding function not found in the same sense in the Germanic languages. As argued in Sect. 2.4, one effect of the reminding function is that it may help the addressee to identify a familiar referent that has not been mentioned for a considerable amount of time. For the informants, the presence of this form may enhance access to a context where the original hearer, in spite of being a random person for the original speaker, is familiar with the referent the speaker is looking for—thus licensing the definite variant. A question that may be pursued in future research is whether the low number of indefinites in the Czech data could also be attributed to the fact that the indefinite marker in Czech is a less economical alternative compared to the canonical bare NP form.

7.2 Indirect Speech

Figure 5 in Sect. 5 shows that indefinite form is the predominant nominal form also in indirect speech. In light of NP processing and referent accessibility per se, this is a surprising result, as using an indefinite when a definite NP or a proper name would be plausible might be considered cognitively more demanding for both narrator and reader than the alternative (see the maximize presupposition principle, Heim 1991). On the other hand, as argued in Sect. 2.3, a preference for indefinites in indirect speech (cf. ‘the iconic approach’) might be due to considerations of production effort and context integration in reported speech. Assuming that the original speech situation is part of the mental representation of the narrator, iconicity (i.e., a greater degree of “copy–paste”) might cognitively require less effort than choosing a proper name or a definite NP, since the latter requires the narrator to link the indirect speech report cohesively to the previous discourse. The strong preference for indefinites in indirect speech observed in Fig. 5 suggests that considerations of context integration in reported speech has a strong influence on NP choice in this case and that the maximize presuppositions principle can be floated or overruled in indirect speech.

When we compare the results in direct and indirect speech, as revealed in Fig. 6 in Sect. 5, we see that the results of our experiment partly support and partly refine the ‘referential report hypothesis’ of Sæbø (2013) and thus Hypothesis 1 in Sect. 4.19 Hypothesis 1 says that more informants will choose definite descriptions and proper names in indirect speech than in direct speech. This hypothesis was supported for English and German with respect to definites, but not for Norwegian and Czech, and not for proper names.20

As mentioned in Sect. 4, Hypothesis 1 will necessarily interact with language-specific properties, for example the constraint against anaphoric use of ten in written discourse and the relative complexity of Norwegian premodified definites. We argued that if discourse function and processing economy are stronger determinants of nominal form in Czech and Norwegian than those that count in favor of choosing a definite description in indirect speech (i.e., considerations of referent accessibility, cf. Sect. 2.3), Hypothesis 1 is not expected to hold for Norwegian and Czech, only for English and German. The results in Fig. 6 show that this turned out to be the case. Whereas definites are used more often in indirect speech than in direct speech in English and German, this does not hold for Norwegian and Czech.

There was no difference in the use of nominal forms between the two contexts in Norwegian. The fact that Norwegian differs from English and German in this sense was a surprising finding to us. However, as pointed out in Sect. 2.4, premodified Norwegian definites might be considered morpho-syntactically more complex than minimally distinct indefinite phrases, and given that language use tends to follow a principle of economy, this raises the expectation that Norwegian indefinite descriptions are preferred to definite ones, other things being equal. Since this asymmetry is not found in English and German, this raises the expectation that Norwegian may show less use of definites than English and German when other options are available.

A reasonable objection to this explanation is that it might not be clear why the difference between Norwegian and the two other Germanic languages is not visible in direct speech as well as in indirect speech. We believe this has to do with whether the NP form is a matter of free choice or not. Clearly, the fact that Norwegian preposed definites are slightly more complex than corresponding indefinite ones does not mean they will never be used; they will necessarily be used, of course, when this is the only plausible option. In direct speech, indefinite form is obligatory if the informant imagines a context in which the three conditions in (9) hold. On the other hand, if the informant imagines a context in which the referent can be uniquely identified by the original hearer, only a definite NP would be appropriate. We assume that when the Norwegians choose a definite form in direct speech (and correspondingly in indirect speech, when reporting on a definite), this is because they have portrayed a context for the original speech situation in which the definite form is obligatory. In indirect speech, there is an additional possibility of using the definite description anaphorically, which allows English and German informants to use definites to a greater extent in this context. This possibility is, however, a matter of optionality, and thus exactly the kind of situation where we expect subtle (dis)preferences to show up. The Norwegians seem to avoid premodified definites when (and only when) there is a plausible, equally acceptable alternative to it that is less complex. This can explain why the alleged dispreference for premodified definites in Norwegian only shows up in indirect speech. These data can be taken to support recent findings in Batsbi (Harris and Samuel 2011), which suggest that double definiteness is a hamper with regard to processing.

The Czech definite form was used more often in direct than indirect speech (see Fig. 5), thus strictly contradicting Hypothesis 1. However, this fact can be related to properties of the Czech determiner ten. This element has many properties in common with the Germanic definite articles (cf. Sgall and Hronek 1992, p. 66), but also some idiosyncratic ones, as discussed in Sect. 2.4. This is reflected in Hypothesis 2, Sect. 3, which says that Czech definites are expected to be chosen less frequently in indirect speech than English and German ones. That the Czech definites are used less in indirect speech than in direct speech can be related to the fact that ten is rarely used anaphorically, and much more seldom in written narrations than in oral discourse (see Sect. 2.4). In the indirect speech examples, any referential term will be interpreted anaphorically and as part of a written narration. In the direct speech examples, on the other hand, the chosen NP is not used anaphorically, since there is no antecedent in the preceding discourse. Furthermore, although it is part of the written discourse, it is part of direct speech, which is to be taken as an exact reproduction of oral discourse. While we were aware of differences between the Czech and the Germanic noun phrase system prior to our experiment, we did not originally hypothesize that the differences between ten and the Germanic definite articles would be influential enough to affect the results for Hypothesis 1. This was therefore a particularly surprising and interesting finding of our experiment.

7.3 General Discussion

As outlined above, our experiment has supported the ‘referential report hypothesis’ of Sæbø (2013) in that we have found that definites are used to a greater extent in indirect speech than direct speech in English and German, but the same results were not found for Norwegian and Czech. Follow-up experiments will be useful in order to investigate this further. An experiment that is minimally distinct from the present one, except that the informants can choose freely which referring expression to fill in for the open slot, will show whether missing alternatives have affected the results in the present experiment. Also an acceptability judgment experiment may be an interesting follow-up study. If informants are asked to judge the relative acceptability of the four nominal forms used in our experiment, rather than picking out one, this may reveal that proper names and definites are judged as acceptable even though they are not the preferred option. In future experiments with NP alternatives, we suggest including a third-person anaphoric pronoun as one of the alternatives, as this form is less likely to be used in the kind of context presupposed in our direct speech examples. Including pronouns will thus increase the chances of revealing potential differences between direct and indirect speech.

An important insight of the experiment is that when one tests for informants’ preference in a case where the informants have a “free choice” (perhaps related more to communicative intention than cognitive and linguistic constraints), very subtle features of the response alternatives and the involved languages may have considerable effect on the results. That is, minor linguistic differences may lead to pulling the choice from one interpretive option to the other and consequently have a crucial influence. In such experiments, it will necessarily be hard to make precise predictions and to come up with plausible working hypotheses. On the other hand, exploratory experiments like these may be extremely useful in pinpointing differences among languages and linguistic items that at first glance are not readily visible but in fact play an important role in language production and comprehension.

8 Conclusion

In this paper, our focus has been on noun phrases used to report on specific indefinites in direct and indirect speech reports. We have presented experimental, cross-linguistic evidence regarding the degree to which speakers prefer, in indirect speech reports, a verbatim reproduction of the noun phrase used in the original speech event (an indefinite to report on an indefinite), or a slightly less accurate paraphrase (a definite or a proper name to report on an indefinite), which, however, reflects the writer’s and reader’s mental state more directly. The option of using indefinites to report on indefinites is preferred across the two conditions and the four languages investigated, but paraphrases with definites constitute an additional option in indirect speech under otherwise appropriate conditions. Indefinites are preferred also in indirect speech even though this could be assumed to be cognitively more costly than the alternative from the point of view of referent accessibility and a mental model perspective.

We have argued that the choice of referring expression involves considerations of audience design, is sensitive to discourse type and discourse functions, as well as being affected by general economy constraints, and we have shown that these constraints surface differently in different languages because of idiosyncratic properties of the individual language. This, in turn, leads to a difference among the investigated languages with respect to how specific indefinites are reported on in indirect speech. The experimental method and contrastive approach have proven very useful for highlighting differences among languages that are relevant for the choice of referring expressions in circumstances as complex as those investigated in our study.

Footnotes

  1. 1.

    The term ‘referring expression’ is here used in a maximally general sense, denoting any noun phrase (NP) that can in principle introduce a discourse entity.

  2. 2.

    For instance, full proper names seem to be used in more formal situations than first names across the four languages that we have investigated.

  3. 3.

    Even direct speech reports do not necessarily reproduce the original speech event exactly (see e.g., Wilson 2000 and Blakemore 2010), but they are typically more verbatim than indirect speech.

  4. 4.

    In some languages, such as Japanese, the direct-indirect speech distinction is not formally marked (see Maier 2009).

  5. 5.

    We will only be concerned with direct and indirect speech in the shape of an interrogative complement clause to a verb of saying in the indicative mood, to the exclusion of so-called free indirect speech.

  6. 6.

    Sæbø (2013) adds a footnote to the ‘referential report’ postulate, saying that the formulation is a bit simplistic: The indefinite description [a P] may be embedded, in which case the notation [a P] Q is inaccurate; and there is imprecision concerning whether the locution “can be represented” (what is intended is that the report is correct (true) under the substitution of u for [a P] if it is otherwise sufficiently faithful).

  7. 7.

    The NP alternatives in Fig. 2 are not supposed to be exhaustive, and they presuppose a context where the entity referred to by the NP has been previously activated.

  8. 8.

    According to Mulkern, the cognitive status associated with complex proper names is different from the one associated with simple ones. She shows that complex proper names such as Peter Simpson can be used in cases where the addressee is not previously familiar with the referent, whereas this is not the case for simple proper names such as Peter or Mr. Simpson. Thus, while simple proper names encode the cognitive status ‘familiar,’ according to Mulkern, complex proper names encode the cognitive status ‘uniquely identifiable,’ just like the definite article in English.

  9. 9.

    If the speaker in (5) asked, “Have you seen the elderly man who always sits on the bench over there?”, use of the definite article is intuitively okay. This is predicted by Gundel et al. (1993), since a rich description makes it much more plausible that the addressee is able to establish a unique representation of the intended referent.

  10. 10.

    (1) is an example of an item that allows for a possessive NP (‘my parrot’) in the target position.

  11. 11.

    The term ‘iconic approach’ is only intended to reflect the iconicity of the NP form, not other aspects of the speech event. Yao and Scheepers (2011) observe that the speaking rate of a reported speech event affected the reading rate during eyetracking when reported with direct speech, but not with indirect speech. This suggests that in contrast to direct speech, indirect speech is not mentally simulated and is thus ‘iconic’ to a lesser extent than direct speech.

  12. 12.

    Although the main rule is that a definite article should appear before the adjective in a Norwegian premodified definite noun phrase, the preposed article can be omitted if the adjective uniquely picks out the referent. This is the case, for instance, with many superlative adjectives.

  13. 13.

    Yet another difference between Norwegian and the other languages is that the possessive pronoun appears after the definite noun in Norwegian, as opposed to English and German, which place the possessive in front of the (simple or complex) noun. This difference does not play any role in our study, though.

  14. 14.

    The comparison of the use of ten with the use of the Russian demonstrative eto shows that they are not equivalents. In contexts where the Czech demonstrative ten is used for the expression of speakers’ emotionality, it can sometimes be translated with the Russian eto (Cze. T y děti, ty děti!—Rus. Ох уж е т и дети!—Eng. These children!). This, however, is not possible in intensifying contexts (Cze. Vem si jinou čepici, nebo ti ty uši omrznou—Rus. Другую шапку надэпь, а то ведь Ø уши отморозишь.—Eng. Take another hat or these ears of yours (‘your ears’) will freeze) (cf. Berger 1993, pp. 185–186). Significant differences between Czech and Russian are also evident in the use of ten in deictic and anaphoric contexts: Cze. Co potřebujete, abyste ten proces vyhrál? —Rus. Что вам нужно чтоды выиграть процесс?—Eng. What do you need in order to win the trial? (cf. Berger 1993, p. 187).

  15. 15.

    Linguistic Context: F(1,24) = 19.53, p < .001, Referring Expression: F(2,48) = 70.19, p < .001.

  16. 16.

    Scheffé tests: direct speech: definite vs. possessive: p < .001, bare vs. possessive: p < .001; indirect speech: definite vs. possessive: p < .04, bare vs. possessive: p < .001.

  17. 17.

    Scheffé test: p < .03.

  18. 18.

    Scheffé test: p < .04.

  19. 19.

    Beware that the correlation to Sæbø’s postulate on specific indefinites is only indirect, since we have not tested directly what utterance the informants imagine to report on when choosing a certain nominal form in indirect speech. The comparison of the results in indirect and direct speech allows us to infer, though, what forms are assumed to be the most natural ones in the original speech event.

  20. 20.

    The fact that there is no statistically significant increase in the use of proper names under the same condition is probably related to the fact that proper names (of human beings) are not used anaphorically as readily as definite descriptions (see e.g., Ariel 1990).

References

  1. Abbot, B. (2004). Definiteness and indefiniteness. In L. Horn & G. Ward (Eds.), The handbook of pragmatics (pp. 122–149). Oxford: Blackwell.Google Scholar
  2. Anderssen, M. (2007). The acquisition of compositional definiteness in Norwegian. In M. Anderssen & M. Westergaard (Eds.), Papers from the language acquisition workshop SCL 2006. Tromsø: CASTL.Google Scholar
  3. Ariel, M. (1990). Accessing noun-phrase antecedents. London: Routledge.Google Scholar
  4. Ariel, M. (2001). Accessibility theory: An overview. In T. Sanders, J. Schilperoord, & W. Spooren (Eds.), Text representation: Linguistic and psycholinguistic aspects (pp. 197–260). Amsterdam: John Benjamins.Google Scholar
  5. Arnold, J. E., Wasow, T., Asudeh, A., & Alrenga, P. (2004). Avoiding attachment ambiguities: The role of constituent ordering. Journal of Memory and Language, 51, 55–70.CrossRefGoogle Scholar
  6. Barr, D. J. (2008). Pragmatic expectations and linguistic evidence: Listeners anticipate but do not integrate common ground. Cognition, 109, 18–40.CrossRefGoogle Scholar
  7. Baumann, P., Konieczny, L., & Hemforth, B. (2014). Pragmatic expectations in anaphora resolution: Alternative constructions and referring expressions. In B. Hemforth, B. Mertins, & C. Fabricius-Hansen (Eds.), Psycholinguistic approaches to meaning and understanding across languages (Studies in theoretical psycholinguistics, pp. 197–212). Cham: Springer.Google Scholar
  8. Berger, T. (1993). Das System der tschechischen Demonstrativpronomina. Textgrammatische und stilspezifische Gebrauchsbediungen. Habilitation thesis, University of Munich, Munich, Germany. Retrieved October 2011, from http://homepages.uni-tuebingen.de/tilman.berger/Texte//texte.html
  9. Blakemore, D. (2010). Communication and the representation of thought: The use of audience-directed expressions in free indirect thought representations. Journal of Linguistics, 46, 575–599.CrossRefGoogle Scholar
  10. Borthen, K. (2004). Predicative NPs and the annotation of reference chains. In Proceedings of the 20th international conference on computational linguistics (COLING 2004) (pp. 1175–1178).Google Scholar
  11. Borthen, K., Brøseth, H., & Fretheim, T. (2008). Annotation manual for the Norwegian NP-form project. Working Papers ISK, 5, 81–126.Google Scholar
  12. Brown-Schmidt, S. (2009). Partner-specific interpretation of maintained referential precedents during interactive dialog. Journal of Memory and Language, 61, 171–190.CrossRefGoogle Scholar
  13. Brown-Schmidt, S., & Hanna, J. E. (2011). Talking in another person’s shoes: Incremental perspective-taking in language processing. Dialogue and Discourse, 2, 11–33.CrossRefGoogle Scholar
  14. Clark, H. H. (1992). Areas of language use. Chicago: University of Chicago Press.Google Scholar
  15. Clark, H. H. (1996). Using language. Cambridge : Cambridge University Press.CrossRefGoogle Scholar
  16. Clark, H. H., & Marshall, C. R. (1981). Definite reference and mutual knowledge. In A. K. Joshe, B. L. Webber, & I. A. Sag (Eds.), Elements of discourse understanding (pp. 10–61). Cambridge: Cambridge University Press.Google Scholar
  17. Clark, H. H., & Murphy, G. L. (1982). Audience design in meaning and reference. In J. F. LeNy & W. Kintsch (Eds.), Language and comprehension (pp. 287–299). Amsterdam: North-Holland Publishing Company.CrossRefGoogle Scholar
  18. Coulmas, F. (1986). Reported speech: Some general issues. In F. Coulmas (Ed.), Direct and indirect speech (pp. 1–28). Berlin: Mouton De Gruyter.CrossRefGoogle Scholar
  19. Cvrček, V., et al. (2010). Mluvnice současné češtiny. Prague: Karolinum.Google Scholar
  20. Farkas, D. (1994). Specificity and scope. In L. Nash & G. Tsoulas (Eds.), Actes du Premier Colloque Langues et Grammaire (pp. 119–137). Paris: John Benjamins.Google Scholar
  21. Fodor, J. D., & Sag, I. (1982). Referential and quantificational indefinites. Linguistics and Philosophy, 5, 355–398.CrossRefGoogle Scholar
  22. Givón, T. (1983). Topic continuity in discourse: A quantitative cross-language study. Amsterdam: John Benjamins.CrossRefGoogle Scholar
  23. Güldemann, T., & von Roncador, M. (Eds.). (2002). Reported discourse: A meeting ground for different linguistic domains. Amsterdam: John Benjamins.Google Scholar
  24. Gundel, J., Hedberg, N., & Zacharski, R. (1993). Cognitive status and the form of referring expressions in discourse. Language, 69, 274–307.CrossRefGoogle Scholar
  25. Gundel, J., Bassene, M., Gordon, B., Humnick, L., & Khalfaoui, A. (2010). Testing predictions of the givenness hierarchy framework: A cross-linguistic investigation. Journal of Pragmatics, 42, 1770–1785.CrossRefGoogle Scholar
  26. Hammer, L. B. (1986). Code-switching in colloquial Czech. In J. L. Mey (Ed.), Language and discourse: Test and protest (pp. 455–473). Amsterdam: John Benjamins.CrossRefGoogle Scholar
  27. Hanna, J. E., Tanenhaus, M. K., & Trueswell, J. C. (2003). The effects of common ground and perspective on domains of referential interpretation. Journal of Memory and Language, 49, 43–61.CrossRefGoogle Scholar
  28. Harris, A., & Samuel, A. G. (2011). Perception of exuberant exponence in Batsbi: Functional or incidental? Language, 87, 447–469.CrossRefGoogle Scholar
  29. Haviland, S. E., & Clark, H. H. (1974). What’s new? Acquiring new information as a process in comprehension. Journal of Verbal Learning and Verbal Behavior, 13, 512–521.CrossRefGoogle Scholar
  30. Heim, I. (1982). The semantics of definite and indefinite noun phrases in English. Ph.D. thesis, University of Massachusetts, Amherst, MA.Google Scholar
  31. Heim, I. (1991). Artikel und Definitheit. In A. von Stechow & D. Wonderlich (Eds.), Semantik: Ein internationales Handbuch der zeitgenössischen Forschung. Berlin: De Gruyter.Google Scholar
  32. Hlavsa, Z. (1975). Denotace objektu a její prostředky v současné češtině. Prague: Academia.Google Scholar
  33. Inoue, A., & Fodor, J. D. (1995). Information-paced parsing of Japanese. In R. Mazuka & N. Nagai (Eds.), Japanese sentence processing (pp. 9–63). Hillsdale: Lawrence Erlbaum.Google Scholar
  34. Ioup, G. (1977). Specificity and the interpretation of quantifiers. Linguistics and Philosophy, 1, 233–245.Google Scholar
  35. Keysar, B., Barr, D. J., Balin, J. A., & Brauner, J. S. (2000). Taking perspective in conversation: The role of mutual knowledge in comprehension. Psychological Science, 11, 32–38.CrossRefGoogle Scholar
  36. Komárek, M. (1978). Sémantická struktura deiktických slov v češtině. Slovo a slovesnost, 39, 5–14.Google Scholar
  37. Krámský, J. (1963). K probleme artiklja. Voprosy jazykoznanija, 4, 14–26.Google Scholar
  38. Lane, L. W., & Ferreira, V. S. (2008). Speaker-external versus speaker-internal forces on utterance form: Do cognitive demands override threats to referential success? Journal of Experimental Psychology: Learning, Memory, and Cognition, 34(6), 1466–1481.Google Scholar
  39. Li, C. N. (1986). Direct and indirect speech: A functional study. In F. Coulmas (Ed.), Direct and indirect speech (pp. 29–46). Berlin: Mouton de Gruyter.Google Scholar
  40. Maier, E. (2009). Japanese reported speech. Against a direct-indirect distinction. In New frontiers in artificial intelligence (Lecture notes in computer science, Vol. 5447, pp. 133–145). Berlin: Springer.CrossRefGoogle Scholar
  41. Mathesius, V. (1926). Přívlstkové ten, ta, to v hovorové češtině. Naše řeč, 10, 39–41.Google Scholar
  42. Meyerstein, Z. P. (1972). Czech deictics: Pronoun and articles? Linguistics, 91, 17–30.Google Scholar
  43. Mulkern, A. (1996). The game of the name. In T. Fretheim & J. K. Gundel (Eds.), Reference and referent accessibility (pp. 235–250). Amsterdam: John Benjamins.CrossRefGoogle Scholar
  44. Percus, O. (2006). Anti-presuppositions. In A. Ueyama (Ed.), Theoretical and empirical studies of reference and anaphora: Toward the establishment of generative grammar as an empirical science (pp. 52–73). Tokyo: Japan Society for the Promotion of Science.Google Scholar
  45. Prince, E. (1981). Towards a taxonomy of given-new information. In P. Cole (Ed.), Radical pragmatics (pp. 223–256). New York: Academic.Google Scholar
  46. Sæbø, K. J. (2013). Reports of specific indefinites. Journal of Semantics, 30, 267–314.CrossRefGoogle Scholar
  47. Sauerland, U. (2008). On the semantic markedness of phi-features. In D. Harbour, D. Adger, & S. Bejar (Eds.), Phi theory: Phi-features across modules and interfaces. Oxford: Oxford University Press.Google Scholar
  48. Schlenker, P. (2003). A plea for monsters. Linguistics and Philosophy, 26, 29–120.CrossRefGoogle Scholar
  49. Sgall, P., & Hronek, J. (1992). Čeština bez příkras. Prague: Nakl. H&H.Google Scholar
  50. Sperber, D., & Wilson, D. (1995). Relevance: Communication and cognition (2nd ed.). Oxford: Blackwell.Google Scholar
  51. von Roncador, M. (2010). Zwischen direkter und indirekter Rede. Berlin: De Gruyter.Google Scholar
  52. Wilson, D. (2000). Metarepresentation in linguistic communication. In D. Sperber (Ed.), Metarepresentation: A multi-disciplinary perspective (pp. 411–448). Oxford,: Oxford University Press.Google Scholar
  53. Yao, B., & Sheepers, C. (2011). Contextual modulation of reading rate for direct versus indirect speech quotations. Cognition, 121, 447–453.CrossRefGoogle Scholar
  54. Zubatý, J. (1917). Ten. Naše řeč, 1, 289–294.Google Scholar

Copyright information

© Springer International Publishing Switzerland 2014

Authors and Affiliations

  • Kaja Borthen
    • 1
    Email author
  • Barbara Hemforth
    • 2
  • Barbara Mertins
    • 3
  • Bergljot Behrens
    • 4
  • Cathrine Fabricius-Hansen
    • 4
  1. 1.Department of Language and LiteratureNorwegian University of Science and TechnologyTrondheimNorway
  2. 2.Laboratoire de Linguistique FormelleCNRS, Université Paris DiderotParisFrance
  3. 3.Institut für Deutsch als FremdsprachenphilologieUniversity of HeidelbergHeidelbergGermany
  4. 4.Department of Literature, Area Studies and European LanguagesUniversity of OsloOsloNorway

Personalised recommendations