2.1 Topicality

There is consensus in the literature that pronominal right-dislocation in Norwegian is connected to topicality (see Askedal 1987; Fretheim 1995; 2000; Faarlund et al. 1997; Vangsnes 2008; Borthen 2018). More precisely, unaccented, right-dislocated pronouns are assumed to encode the sentence topic.

A sentence topic can be described informally as “what the sentence is about”, as opposed to “what is predicated about the topic” (cf. Dik 1978; Reinhart 1981; Lambrecht 1994; Gundel & Fretheim 2004). As for what sentence topics are on a deeper level, and how to define the category more strictly, there are different classifications and definitions around. For the analysis of pronominal right-dislocation, the classical view on sentence topics proposed by Reinhart (1981) is both descriptively and intuitively appropriate, and will be adopted here. According to this account, sentence topics are entries under which the sentence’s proposition is stored in memory, if accepted to be true (Reinhart 1981: 80). The remaining non-topical part of the sentence is labelled comment.

Reinhart points out that a sentence topic is typically but not necessarily realized as the subject of the clause and as definite or pronominal, and it is typically but not necessarily already given in the discourse. The tendency for topics to represent given information (with potential effects on syntactic function and nominal form) can be seen as a secondary effect of its cognitive function as a “memory tag”: It is more efficient to successively store information under one memory tag which is repeated in subsequent sentences, than to constantly change the locus for the representation of the new information.

Reinhart (1981) acknowledges that some sentence structures may encode a specific topic-comment structure. Left-dislocation is one such construction. In the sentence (As for) Peter_i, he_i never arrived, Peter is necessarily the sentence topic. However, according to Reinhart (ibid.), most sentence structures may realize different topic-comment structures in different contexts. The sentence Max saw Rosa, for instance, has Max as the sentence topic when answering the question Who did Max see? whereas it has Rosa as the sentence topic when answering the question Has anybody seen Rosa yesterday?. Thus, information structure is a fundamentally pragmatic notion in Reinhart’s sense, and one way of testing whether a linguistic form marks the sentence topic, is to test what questions it answers in a natural way.

As described in Borthen (2018), judgments of such question-answer pairs suggest that pronominal right-dislocation in Norwegian is a topic construction in Reinhart’s (1981) sense. Consider the two alternative answers to A’s question in (4) and compare them to the parallel data in (5) (from Borthen 2018: 408). (4B2) is pragmatically unacceptable.

As Borthen points out, a crucial difference between the unacceptable answer (4B2) and B’s answer in (5), has to do with information structure. In (5), the referent of the right-dislocated pronoun and its correlate is part of the topic settled by A’s question, and pronominal right-dislocation is natural. In (4), the referent of the right-dislocated pronoun and its correlate is part of the utterance’s comment, and pronominal right-dislocation is not natural.

There are also syntactic pieces of evidence that suggest that pronominal right-dislocation in Norwegian is a topic construction. As first observed by Askedal (1987), the sentence-internal correlate must be either a main clause subject or a main clause fronted object. In contrast, a direct object in the canonical object position cannot function as the correlate of a right-dislocated pronoun. This is shown in (6) and (7) (from Borthen 2018: 408).

Since subjects and fronted objects are more typical topic expressions than are canonical objects (see e.g. Lambrecht 1994: 132; Ward & Birner 2004: 158–163), the data in (6) are in line with the hypothesis that pronominal right-dislocation is connected to topicality.

Other syntactic patterns point in the same direction. For instance, it is not possible to right-dislocate a Norwegian expletive pronoun (cf. Søfteland 2014: 34ff). As argued in Borthen (2018), this is compatible with the topicality hypothesis since an expletive is semantically empty and thus cannot represent “what the sentence is about”.

A construction that is compatible with pronominal right-dislocation is left-dislocation. As illustrated by the authentic example in (8) (Borthen 2018: 410), a right-dislocated pronoun and its correlate may be coreferential with a left-dislocated constituent.

Left-dislocation is assumed to encode the sentence topic in Norwegian (Faarlund et al. 1997: 907). Thus, Borthen (2018) argues that it is as expected that a right-dislocated pronoun and its corelate may be coreferential with a left-dislocated constituent, given that pronominal right-dislocation is a topic construction. One question that remains to be answered, is why a speaker would want to mark the topic twice. This question will be brought up in section 3.5.

The data in (4)–(8) above are as expected, given the hypothesis that a right-dislocated pronoun encodes the sentence topic. However, some authors claim that there is also some counter evidence. Askedal (1987) points out that a right-dislocated pronoun can be modified by a focus adverb such as også ‘also’. One example is (9) (cf. Borthen 2018: 410).

According to Borthen (2018), examples like this challenge the assumption that pronominal right-dislocation is a topic construction, since focus adverbs such as også ‘also’ are assumed to mark rhematic (and thus non-topical) information (cf. König 1991; Fretheim 1999; Faarlund et al. 1997: 915–920).⁴ Based on data such as these, it seems necessary to distinguish between unaccented, non-modified right-dislocated pronouns (that always represent sentence topics), and accented and possibly modified pronouns (that may represent non-topical information). Along these lines, both Askedal (1987) and Fretheim (2000) conclude that there are two kinds of pronominal right-dislocation in Norwegian. Such disjunctive analyses are not empirically wrong, but they are stipulative. In section 3, we will use Givón’s (1983a) notion of markedness to provide a rationale for why the information status of various types of right-dislocated constituents may vary.

Another unresolved question that the previous accounts do not solve, is the following: Since the correlate of the right-dislocated pronoun is necessarily a subject or a fronted object – which are prototypical topic expressions (see e.g. Lambrecht 1994) – why add the right-dislocated pronoun in addition? Why mark a default topic as the topic? As will be made clear in section 3, we believe the answer to this question points to the very essence of the construction, which is, in our view, not merely to encode the sentence topic, but to present it as marked.

2.2 Reference-related properties

2.3 Eventive breaks

Fretheim (1995: 34) points out that the use of a right-dislocated pronoun is sometimes related to breaks in the discourse. He uses the constructed examples in (15) to illustrate this. B1 and B2 are two competing follow-ups to A’s utterance. B1, with pronominal right-dislocation, is the more natural response of the two.

If person A had continued to talk about what happened when A met Axel Aarvoll, using an unaccented pronoun in subject position without a right-dislocated pronoun would have been the natural choice. But as Fretheim points out, when person B interrupts A’s story by mentioning that Axel has been married to his second cousin, using a right-dislocated pronoun is more coherent than not. According to Fretheim, it is the new turn in the story that triggers and favors the use of the right-dislocated pronoun. Similar authentic data are presented in Borthen (2018).

Data such as those in (15) illustrate that pronominal right-dislocation is particularly natural in cases of minor breaks in the discourse, but so far, it has not been clear how to connect this fact to the other properties of the construction, such as its assumed status as a topic construction. In section 3, we will argue that pronominal right-dislocation is not only a topic construction, but a marked topic construction that is used in cases of a lower degree of predictability and continuity than normally expected for topics, e.g. in cases of topic shift. This explains why pronominal right-dislocation is particularly natural in (15), since person A’s utterance is about what happened that day whereas B’s response is about Axel Aarvoll.

2.4 Contrast

Borthen (2018) shows that pronominal right-dislocation is often used when a contrastive meaning aspect is available in the discourse, and in such cases, the construction is particularly natural. This is illustrated in (16)–(18), from Borthen (2018).

In (16), the speaker’s view that it is ok to be separated opposes the addressee’s implicated view that it must be awful to be separated, and this contrastive meaning aspect is highlighted by the right-dislocated pronoun. Another example discussed in Borthen (2018), is (17).

(17) describes two facts that can be perceived of as contrasting. Whereas most of the authors arrive two days ahead of the performance, Ingvar Ambjørnsen arrives 3–4 days in advance. The use of a right-dislocated pronoun is particularly natural in this context, and its presence emphasizes the contrast between the behavior of Ambjørnsen and the other authors.

(18) below illustrates that a sentence with a right-dislocated pronoun may achieve a contrastive interpretation even in cases where this contrast is not licensed by the preceding context.

As reported on in Borthen (2018), the prohibition against sitting in the sun is presented as something that holds for the male character, as opposed to others. Borthen argues that since there is no context-driven bias towards a contrastive interpretation in (18), it must be the right-dislocated pronoun that makes this contrastive interpretation particularly highly accessible.

There are many types of contrast. According to Borthen (2018), pronominal right-dislocation gives rise to the particular type of contrast defined in (19).

The idea behind (19) is that a right-dislocated pronoun that is part of an utterance used to communicate a proposition p of the format [SUBJ₁ PRED₁], encourages the addressee to look for a contrasting proposition q of the format [SUBJ₂ PRED₂], with the same sentence predicate as in p, but with opposite polarity. For instance, (19) is intended to cover the contrast between the propositions ‘Peter is sleeping’ and ‘Mary is not sleeping’.

(19) underspecifies how accessible the contrasting proposition q is when a right-dislocated pronoun is used. It may e.g. be an entailment of the previous utterance (as in (17)), an implicature communicated in the previous discourse (as in (16)), or a contextual assumption established as part of the interpretation process based on the content of the utterance (as in (18)).

Borthen’s formulation of (19) is, however, not entirely accurate. Given that the sentence predicate is the locus for the negation, it is not correct to say that p and q have opposite polarity while PRED₁ and PRED₂ are identical. It is also not obvious from (19) whether it covers fronted objects as well as subjects. And finally, there is a more fundamental problem. Borthen claims that pronominal right-dislocation is a construction with an encoded procedural semantics in the sense of Blakemore (1987) and Wilson & Sperber (1993), and the contrastive meaning aspect in (19) is seen as part of this semantics. The problem is that contrast does not seem to be involved in all cases with pronominal right-dislocation. For instance, there is intuitively no contrastive aspect of the kind presented in (19) involved in (15), where the right-dislocated pronoun merely has a discourse structuring effect. This suggests that contrastiveness is not semantically encoded by the construction, although the construction makes such contrastive interpretations more accessible.

In section 3, we propose a refined version of Borthen’s (2018) definition of the contrast involved in pronominal right-dislocation, and we argue that marked constructions in general heighten the accessibility of contrastive interpretations. This means that the contrastive meaning aspect of pronominal right-dislocation will often, but not necessarily, play a role in the interpretation process.

2.5 Interpretational effects

The interpretational effects of pronominal right-dislocation have been described in various ways in the Norwegian literature. Askedal (1987) calls pronominal right-dislocation “a grammaticalized means for creating an atmosphere of intimacy that suits to further the cooperational intercourse between [the speaker] and the hearer.” Eiesland & Vindenes (2017) single out two interpretational effects, “emphasis” and “mitigation”, whereas Borthen (2018) shows that the interpretational effects range from a strengthening of the speaker’s emotional or epistemic attitude towards the proposition to mitigation of potentially face-threatening speech acts. Fretheim (1995; 2000) does not mention any such interpretational effects at all.

According to Borthen (2018), the interpretational effects are pragmatic inferences that arise as a result of the specific type of contrastiveness licensed by pronominal right-dislocation (see (19)). This contrastiveness does not involve mutually excluding propositions. On the contrary, the contrasting propositions p and q in (19) are compatible and comparable in the sense that p is interpreted in the light of q. The contrasting assumption q contributes to the context for the utterance, and consequently, pronominal right-dislocation is expected to affect pragmatic inferences.

Example (3) in section 1 illustrates this point. The utterance with a right-dislocated pronoun in (3) is repeated in (20a) below. The utterance’s propositional content p is represented in (20b) and the relevant contrasting proposition q in (20c).

One effect of the right-dislocated pronoun in (3) is that the speaker appears as more emotionally hurt than she would have without the right-dislocated pronoun. Borthen’s explanation for this is that the presence of a right-dislocated pronoun instructs the addressee to look for a contrastive interpretation as defined in (19), which leads to pragmatic inferences. In (3), p is ‘Torkild does not count’, and a contextually available contrasting proposition that leads to a relevant interpretation for the utterance as a whole, is ‘Other people count’. It is sad and potentially upsetting that a boy who has a handicap does not count in some people’s view, but the boy’s situation looks even sadder in the light of other people counting. This explains why the speaker appears as particularly emotionally hurt when she uses a right-dislocated pronoun in (3). This interpretation (that the speaker is upset) could have been reached also without the right-dislocated pronoun, but the instruction to look for a contrasting proposition gives it a “kick-start”.

In (17) and (18), the right-dislocated pronouns have a slightly different type of emphasis effect, but Borthen’s (ibid.) explanation is the same. The target utterances are represented in (21a) and (22a), together with the contrasting propositions in (21bc) and (22bc).

As argued in Borthen (2018), the propositions in (21b) and (22b) may not appear as particularly noteworthy in and of themselves, but this changes when interpreted in the light of (21c) and (22c). This explains why the contents of the target sentences in (17) and (18) are intuitively perceived of as being highlighted with the right-dislocated pronoun, and more generally why pronominal right-dislocation is sometimes claimed to contribute “emphasis”.

As mentioned earlier, pronominal right-dislocation can also have a mitigating function (cf. Askedal 1987; Borthen 2018; Eiesland & Vindenes 2017). For instance, in (16), Syns egentlig det er ok, jeg (‘In fact, I think it is ok, I’), the right-dislocated pronoun intuitively mitigates what the speaker says. Borthen argues that also this effect is a consequence of contrastive interpretations, in (16) between ‘The speaker thinks being divorced is ok’ and ‘The addressee does not think being divorced is ok’. However, although these intuitions seem correct, it is not entirely clear how the two seemingly incompatible pragmatic effects “mitigation” and “emphasis” arise. And more generally, it is not clear on Borthen’s account how to account for the fact that the pragmatic effects are sometimes, but not always, present. The analysis that we propose in section 3 solves this puzzle. By assuming that contrastiveness is not encoded by the construction (merely made more highly accessible), it follows that the pragmatic effects are often but not always present. And since the pragmatic effects are implicatures, it is as expected that the content of the target utterance may influence the direction of the implicatures to great extent.

2.6 Summary of previous research and desiderata

The previous research can be summarized by six observations and remaining questions. First, unmodified and unaccented right-dislocated pronouns have been convincingly argued to always pick out the sentence topic. However, it is not clear how this can be reconciled with the fact that right-dislocated pronouns can sometimes be accented and modified by focus adverbs, in which case they do not represent the sentence topic. Moreover, it is not clear why there is a need to mark a default topic (subject or fronted object) as the topic. Secondly, a right-dislocated pronoun can sometimes affect reference assignment, but only rarely so. We have argued that this cannot be due to encoded cognitive status, but an alternative explanation for the observed effects is missing so far. Third, it has been proposed that the construction marks the present and upcoming discourse topic, which may be hypothesized to affect subsequent mentioning, but this hypothesis has failed to be empirically verified. Fourth, the construction is sometimes, but not always, used in connection with discourse breaks. Although descriptively unquestionable, it is unclear how this fact is related to other properties of the construction. Fifth, the construction often, but not always, gives rise to a specific type of contrastive interpretation, and it has been argued that this may lead to various pragmatic effects. However, one main challenge is to connect contrastiveness to the construction’s remaining properties. And sixth, there is a need for a more explicit account of when the construction achieves the various interpretational effects, especially the emphasizing and mitigating functions, which at first glance seem incompatible. More generally, there is so far no account that successfully ties the construction’s many properties together.

In the next sections, we will leave the grammatical details of the construction to further investigations and take for granted that unaccented, nonmodified right-dislocated pronouns have to have subject or fronted object correlates, and necessarily represent the sentence topic. The main contribution of our analysis is twofold. First, we will show that there is a rationale behind the fact that right-dislocated unaccented and nonmodified pronouns encode the sentence topic whereas other types of right-dislocated constituents may have a different information structure status. And secondly, we will argue that there is a connection between the construction’s status as a specific type of topic construction and its effects on discourse coherence and interpretation.

3.1 Givón’s theory on marked constructions

Givón (1983a; b) aims at predicting the likelihood for a linguistic structure to represent the sentence topic. Inspired by assumptions within gestalt psychology, he argues that continuity in discourse, both at micro and macro level, is preferable cognitively speaking, and that since predictable and highly accessible information is easier to process than nonpredictable information, languages will tend to use their most simple and least marked constructions in cases of high degree of predictability and accessibility, whereas more marked constructions will be used in cases of lower degree of predictability and accessibility.

Givón (ibid.) defines markedness such that the more phonological material, the more semantic material, and the greater morphological or syntactic complexity a linguistic form involves, the more marked it is. In line with this, hierarchies such as (23) below can be derived and used to predict whether a construction will typically be used in cases of relative continuity and predictability or the opposite (cf. Givon 1983a: 17).⁸

According to (23), zero elements and unaccented pronouns will be used when a referent is highly accessible and the discourse highly continuous, whereas e.g. the cleft construction is predicted to be used in cases of low degree of accessibility and predictability, i.e. when the clefted constituent represents new, unpredictable information.

Givón (1983a; b) defines topicality in terms of givenness, which means that the hierarchy in (23) represents not only the likelihood for an expression to represent predictable and continuous information, but also the likelihood for it to represent the sentence topic. This view on topicality is partly at odds with the “aboutness” definition of sentence topic presented in section 2. However, one does not have to adopt Givón’s definition of topicality to make use of his notions of markedness and predictability. It is generally agreed upon that topics tend to be given and tend to be realized by certain forms. Thus, Givón’s model can be used to make more fine-grained predictions concerning these tendencies, and it can be used to explain the form-function correlations rather than stipulating them.

The degree of predictability and continuity in Givón’s model is measured relative to what would be the case for the language’s more unmarked constructions. For instance, the b-versions in (24) and (25) below, with right- and left-dislocated constituents, are more marked than the a-versions, since they contain more phonological and semantic material and involve greater syntactic complexity.

Being more marked, the prediction is that the b-versions will tend to occur in cases with a lower degree of continuity and predictability than the a-versions, for instance in cases where the subject referent has not been mentioned recently.

Givón (1983a) pinpoints that predictability and continuity in the discourse applies to several levels of the discourse, including not only participant continuity (i.e. successive continuity of sentence topics), but also event continuity and thematic continuity. Thematic continuity corresponds to continuity of the discourse topic (see section 2.2), which is claimed by Givón to not correlate with linguistic forms such as those in (23) to the same extent as participant and event continuity. Event continuity corresponds primarily to temporal sequentiality and adjacency within a thematic paragraph. Most commonly, within a thematic paragraph, actions are represented in the sequential order in which they naturally occur, and typically there is a small if any temporal gap between one action and the next. According to Givón (1983a: 8), language users often choose a marker of lower accessibility (e.g. a proper name instead of a pronoun) when there is an eventive break in the discourse, e.g. in connection with time passing by, change into a new sub event, or change of perspective (see also Ariel 1990).

3.2 Pronominal right-dislocation as a marked construction

According to Givón’s (1983a) model, pronominal right-dislocation creates a marked construction compared to the minimally distinct canonical version of the sentence.⁹ The scale in (26) follows from Givón’s definition of markedness.

Since pronominal right-dislocation is a more marked form than the corresponding structure without the right-dislocated pronoun, it signals a lower degree of predictability and continuity of the two. At the same time, it is a construction that encodes the sentence topic, and the right-dislocated pronoun is less marked than other right-dislocated constituents, due to its “light” phonology and rudimentary semantics. This is compatible with cases where the target referent is a sentence topic that is just less than maximally predictable and continuous, e.g. recently mentioned but not identical to the previous sentence topic.

In all the examples that we have presented in this paper pronominal right-dislocation involves a topic shift in the sense that the current sentence topic (identified by the right-dislocated pronoun) is different from the sentence topic of the previous sentence. There are some exceptions, but the same tendency is found for the whole data set represented in Table 1. Thus, contrary to Averintseva-Klisch (2006; 2008a; b; c), who argued that pronominal right-dislocation marks the discourse topic and is a forward-looking construction, we claim that pronominal right-dislocation is a backward-looking construction: It picks out the sentence topic and signals a lower degree of predictability than otherwise expected for sentence topics, e.g., that the topic is not the same one as in the previous utterance.

In what follows, we will show that this analysis accounts for the properties summarized in section 2.6, including right-dislocated pronouns’ (rare) co-occurrence with focus adverbs, the construction’s possible (but rare) effect on reference assignment, its use with minor discourse breaks, its contrastive aspect, and interpretational effects.

3.3 Differences among various types of right-dislocated constituents

The degree of nonpredictability and discontinuity signaled by linguistic forms on a scale such as (26) will be determined by the constituent’s degree of markedness, phonologically, morphosyntactically and semantically. Since an unaccented pronoun is not very complex, right-dislocated pronouns are expected to be used in cases of minor discontinuities. Right-dislocated accented pronouns and full nominal phrases, on the other hand, are semantically and phonologically more complex and thus expected to be used in cases of more serious discontinuities, possibly in reference to brand new discourse referents. These predictions are in accordance with the fact that lexical right-dislocations are often used for reference clarification due to a distant antecedent, whereas this is extremely rare for right-dislocated pronouns.

If a constituent in the right-dislocation position merely signals less degree of predictability and continuity than expected for the corresponding canonical structure, there is in principle nothing that prevents a right-dislocated constituent from being non-topical, as in the case of a right-dislocated pronoun being modified by a focus adverb (see (9)). Due to its form and content, a right-dislocated phrase consisting of a pronoun and a modifying focus adverb will be positioned towards the lower end of a scale of markedness (see (26)). On Givón’s givenness account of topicality, it follows from this that the phrase is extremely unlikely to be the sentence topic. On the aboutness approach to topicality, the exact information status of such a construction has to be accounted for in other ways. However, also on this account, Givón’s notion of markedness adds an explanatory aspect to the otherwise merely stipulated correlation between forms and information structure values: The correlation between a form and an information status is not random but connected to the form’s degree of markedness. Although Givón’s model of markedness in combination with an aboutness definition of topicality does not predict that a right-dislocated pronoun modified by a focus adverb represents non-topical information, it provides a non-stipulative rationale for why the information status is different from that of a right-dislocated unaccented pronoun.

3.4 Connection to discourse breaks

Since Givón (1983a) acknowledges that nonpredictability and discontinuity may appear at either the participant level or the eventive level, it is predicted that pronominal right-dislocation may be used in connection with eventive breaks, as mentioned in section 2.3. This is further illustrated with the authentic example in (27) (cf. Borthen 2018: 17).

In this discourse fragment, one and the same referent is the subject of three subsequent sentences, which seems to be a straight-forward case of topic continuity. However, there is an eventive break just before the sentence with the right-dislocated pronoun. Just as in (15) in section 2.3, the perspective changes from that of reporting on events that involve the given discourse referent during a certain period of time (answering the question What happened?) to that of reporting on a state that holds for the discourse referent. And intuitively, the presence of the right-dislocated pronoun compensates for the break.

It is possible to account for the connection between pronominal right-dislocation and eventive breaks with reference to information structure. Most of the sentence predicates in our data set (cf. Table 1) are stative predicates, i.e., predicates that are used to refer to properties that hold for, and are homogeneous and unchanged for, an individual over time (cf. Mittwoch 2019; Maienborn 2019). Only a few of the sentence predicates are eventive, i.e. with a lexical bias towards an interpretation where they refer to something that happens to an individual (ibid.). This tendency is not surprising given that the referent of an unaccented unmodified right-dislocated pronoun is always interpreted as the sentence topic. According to Jäger (2001), eventive predicates may supply a linguistically unexpressed topic in terms of a Davidsonian event argument while stative predicates do not. The topic of a sentence with an eventive predicate may therefore be an event variable (“concerning what happens”) or an individual variable, whereas the topic of a sentence that denotes a state has to be the subject referent (“concerning this individual”). From this, it is not surprising that we find fewer eventive than stative predicates among our data.

As pointed out by Mittwoch (2019: 34), the distinction between states and events is not fully determined by the verb that is involved; a predicate that may be described as denoting an event (or have a lexical bias towards event interpretations) may achieve a stative interpretation in a particular context. Indeed, the handful of sentence predicates in our data set that have a lexical bias towards an eventive interpretation achieve a stative interpretation in the given context when a right-dislocated pronoun is present. In (17), the arrival three to four days ahead of the event is presented as a habit (and thus a state) of Ingvar Ambjørnsen rather than an event. In this sense, our data suggest that right-dislocated pronouns enforce a stative interpretation of the sentence predicate.

This can be connected to the assumption that pronominal right-dislocation is a topic construction. Given that an event interpretation that answers the question What happened? has a Davidsonian (unexpressed) event argument as the topic (cf. Jäger 2001), it is just as expected that the use of a structure that identifies a different sentence topic is incompatible with it.

We now have an explanation for Givón’s observation that markers of low degree of predictability are not only used with referential discontinuity, but also eventive discontinuity. In examples such as (15B1) and (27), the same referent is referred to in subject position in two adjacent sentences, the second time with an accompanying right-dislocated pronoun. Our explanation for why pronominal right-dislocation is often used in such cases, is that the first sentence – despite the subject pronoun – has an eventive variable as its sentence topic (answering the question What happened?), whereas the second sentence has the referent of the right-dislocated pronoun as its sentence topic. This means that the sentence topic changes from one sentence to the other, there is discontinuity, and the use of a marked topic construction becomes highly appropriate.

3.5 Marked constructions, nonpredictability and contrast

Contrastiveness is not a central aspect of Givón’s (1983a) work, but both Givón (1983a) and Ariel (1990) do mention competition as one factor that affects referent accessibility. Moreover, it can be observed that marked constructions often give rise to contrastive interpretations. Consider the minimal pair in (28). The subject constituent of (28b) is accented and thus more marked than the subject in (28a) in Givón’s (1983a) sense.

Intuitively, a contrastive interpretation comes easier to mind for (28b) than for (28a). A natural context for (28b) is a situation with more than one activated female discourse referent, where the predication that holds for the subject referent does not hold for some competing referent candidate. The tendency for accented pronouns to license contrastive interpretations has also been observed in the literature (see e.g. Rooth 1992: 7; McClay & Wagner 2015).

A similar pattern holds for fronted objects. Being structurally more complex, (29b) is more marked than (29a), according to Givón (1983a).

As described in e.g. Ward & Birner (2004), topicalized phrases as in (29b) may have one of two functions in English, one of which is a contrastive topic interpretation (see also Roberts 1996; Krifka 2007).

Also left-dislocation is particularly natural in situations where the given discourse referent is being contrasted to other accessible discourse referents. (30) illustrates this.

Observations in the literature point in the same direction (see e.g. Chafe 1976), thus associating left-dislocation with contrastiveness.

Based on data such as (28)–(30) and the related literature, we conclude that other things being equal, marked constructions make contrastive interpretations more highly accessible than the corresponding unmarked ones.

There are at least two ways in which this follows from Givón’s (1983a) model. First, a contrastive interpretation often involves a context with competition among potential discourse referents, which means that the intended referent of the given expression is not maximally accessible. Some types of contrastiveness are also linked to marked constructions in a more general way. For instance, an assumption that contrasts with what is known, in the sense of having opposite polarity to it, is surprising and thus nonpredictable. And an assumption about a certain individual which contrasts with what holds for other individuals (see e.g. (16) and (17)) may also come as a surprise. Thus, contrastive interpretations constitute the kind of meaning that one might expect to be communicated through marked constructions. This means that there is no need to assume that contrastiveness is encoded in the semantics of each type of marked construction; rather, it is a meaning aspect that is made more highly accessible by any marked form compared to its minimally distinct unmarked version.

In terms of the type of contrastive interpretation it gives rise to, pronominal right-dislocation differs from the marked constructions discussed above. Compare the examples in (31).

In (31a), the markedness of the structure is evident already after the second constituent has been processed. Given that the addressee starts searching for a contrastive interpretation at this point, a likely option is to look for alternative referents against which Øystein can be contrasted. In (31b), on the other hand, the markedness of the structure is not evident until the whole utterance has been processed. In such cases, the addressee can use the content of the sentence (as well as pragmatically enriched versions of it) as a basis for identifying potentially contrasting entities. This means that the contrasting entity may be a proposition and it means that the contrasting proposition need not be contextually accessible prior to the utterance, as in (16) and (17); it may be derived through pragmatic inferencing based on the utterance content and whatever contextual clues are otherwise available, as in (18).

Since the two types of dislocation in (31a) and (31b) represent different degrees of markedness, and give access to different types of contrast, this explains why they are sometimes combined and mark the sentence topic twice, as in (8). The constructions are compatible as long as they pick out the same sentence topic. To add a left-dislocated phrase to a sentence with a right-dislocated pronoun may be relevant due to its descriptive content, that enables reference identification, and to add a right-dislocated pronoun to a sentence with a left-dislocated phrase may be desirable due to the specific type of propositional contrast that the construction gives rise to.

We define the type of contrast involved in pronominal right-dislocation as in (32), which is a revised version of Borthen’s (2018) definition discussed in section 2.2.

To illustrate this, in (16), based on the topic-comment structure of the sentence with the right-dislocated pronoun, the proposition “[I]_TOPIC1 [think it is ok (to be separated)]_COMMENT1” can be compared to “[Some people]_TOPIC2 [do not think it is ok to be separated]_COMMENT2”, where COMMENT₂ is the negation of COMMENT₁. Our definition is reminiscent of the type of contrast called semantic opposition in Lakoff (1971) and defined in Spooren (1989: 31) as “a relation between two conjuncts each having different subjects, to which properties are attributed that are mutually exclusive in the given context”.

Summing up so far, we assume with Givón (1983a; b) that marked constructions signal nonpredictability and discontinuity compared to minimally distinct unmarked forms, and since pronominal right-dislocation is a marked construction, it gives rise to such interpretations. However, differently from Givón (1983a), we have argued that marked constructions in general heighten the accessibility of contrastive interpretations. This means that for pronominal right-dislocation, signalling nonpredictability may instantiate itself in one of two ways, or both: Either the construction indicates that the sentence topic that is identified is less predictable than expected for sentence topics (as in the case of topic shifts), or it indicates a particular type of contrastive interpretation based on the topic-comment structure of the sentence, i.e. lower predictability of the entire sentence content.

3.6 Effects on reference assignment

We are now in the position to explain why the right-dislocated pronoun in (12) affects reference assignment. Recall the data: Whereas the subject pronoun in (11) Den har et siksakband over ryggen ‘It has a zigzag band down its back’ is interpreted as referring to the particular viper that the addressee just mentioned, the same pronoun in the same position and context is more naturally interpreted as the ‘species viper’ when a right-dislocated pronoun is added, as in (12). Fretheim (1995) explains these data by an analysis which focuses on reference assignment only and implicates considerable enumeration of the meaning of determiners and pronouns in the lexicon. Our explanation is more general. The target sentence in (12) is a marked topic construction, which means that the addressee can expect some kind of discontinuity or nonpredictability compared to what would have been the case without the right-dislocated pronoun, i.e. a topic shift and/or a contrastive interpretation as defined in (32). If the pronoun den is interpreted as the particular viper just mentioned, this expectation is not borne out, whereas it is borne out if the pronoun is interpreted as the species viper. In the latter case, the sentence topic shifts from the first to the second utterance. Moreover, (12) achieves a contrastive interpretation, a fact that Fretheim (1995) fails to observe. Also the contrastive aspect of the construction contributes to the given reference assignment, since interpreting the proposition ‘the species viper has a zigzag band down its back’ in the light of the contrasting assumption ‘the snake over there does not have a zigzag band down its back’ provides an explanation for why the speaker does not think the given snake is a viper. In sum, this explains the difference between (11) and (12) without the need of lexical enumeration.

3.7 Explaining pragmatic effects

Borthen (2018) argued that it is the contrastive aspect of pronominal right-dislocation that gives rise to interpretational effects such as a strengthening of the speaker’s epistemic or emotional attitude towards the proposition or a weakening of a potentially face-threatening speech act. What is not entirely clear from Borthen’s account, though, is how one single construction can result in diametrically opposite effects (“emphasis” or “mitigation”). In this section, we illustrate how pragmatic interpretations may be derived in context, with a special focus on this alleged interpretational paradox.

According to relevance theory, the overall comprehension process involves three basic subtasks: a) constructing an appropriate hypothesis about explicit content (the utterance’s explicatures), b) constructing an appropriate hypothesis about the intended contextual assumptions (implicated premises), and c) constructing an appropriate hypothesis about the intended contextual implications (implicated conclusions) (Wilson & Sperber 2004: 615; Sperber & Wilson 1986/1995; Carston 2002). An explicature is an ostensively communicated truth-evaluable proposition which is a development (“fleshing out”) of the encoded concepts of the utterance. Explicatures may be used as vehicles together with contextual assumptions for the generation of implicatures, i.e. communicated assumptions whose derivation is based solely on pragmatic inference. Importantly, context is not a static entity that exists prior to an utterance on this view; rather, context is a mental object, i.e. the set of manifest assumptions that are needed in order to interpret an utterance.

On the relevance-theoretic account of the pragmatic effects of pronominal right-dislocation that we propose, the so-called “emphasis” and “mitigation” effects are implicatures. More precisely, the contrasting proposition q in (32) can function as a contextual assumption in light of which the utterance’s most relevant (basic or higher-level) explicature p ought to be interpreted – which may in turn lead to implicated conclusions that contribute an extra “layer” of meaning.

The relevance-theoretic framework accounts for the fact that communication can be either strong, where the speaker’s informative intentions are clear because there is strong evidence for them, or weak, where the speaker produces an utterance that opens for a number of possible implications, expecting that at least some of them will be fruitful enough for the addressee to offset the needed processing costs (Sperber & Wilson 1995: 197). Correspondingly, implicatures can be either strong or weak, and some of the implicatures that pronominal right-dislocation give rise to, may be weak in this sense.

A sentence with pronominal right-dislocation underdetermines several meaning aspects, including i) whether the speaker intends a contrastive interpretation or not (the marked construction may also be used for the purpose of signaling topic shift only), ii) which propositions ought to be contrasted, and iii) which implicatures to derive from the contrasting proposition in combination with the utterance content and other contextual assumptions. Thus, one might wonder how the addressee is able to land on an interpretation at all. The relevance-theoretic explanation for this rests ultimately on The communicative principle of relevance (Sperber & Wilson 1987/1995: 158), which is a heuristics for utterance interpretation. This principle says that when we process an utterance, we expect the processing cost that we invest to be worthwhile, i.e. optimally relevant. Relevance is measured in terms of a balance between processing effort and positive cognitive effects, where a positive cognitive effect is a worthwhile change in a person’s representation of the world (Wilson & Sperber 2004: 608). In accordance with this, a right-dislocated pronoun raises the expectation that the addressee will either benefit a decrease in processing effort (as may be the case when signaling topic shift) or obtain extra cognitive effects that would not otherwise be (as) available.

Concerning the interpretational effects, the relevance-theoretic prediction is that the addressee will look for interpretations that yield the most cognitive effects for the least processing effort. Moreover, he or she will take into consideration the most accessible (least processing demanding) contextual assumptions first and stop the interpretation process as soon as enough cognitive effects are derived. Less accessible contextual assumptions may be accessed and more costly cognitive effects derived if that is needed to achieve the expected level of relevance.

We can now explain the interpretational effects that have been observed for pronominal right-dislocation, e.g. the fact that the speaker seems more emotionally hurt in (3) when a right-dislocated pronoun is present. The right-dislocation makes the utterance marked, which heightens the accessability of a contrastive interpretation. In (3), the explicature ‘Torkild does not count’, does contribute relevant information in and of itself, but much more so if interpreted in light of the contrasting proposition ‘Other people count’ and various assumptions about Torkild, such as Torkild being a special and vulnerable boy. From these and other contextual assumptions, the conclusion ‘The speaker is upset that Torkild is treated worse than other people’ may be derived, which provides extra cognitive effects to compensate for the extra effort imposed by the processing of the right-dislocated pronoun. The addressee might have landed on the same interpretation also without the right-dislocated pronoun, but with the pronoun, the addressee gets a hint to look for some extra cognitive effects, and the specific interpretation gets a kick-start due to the heightened accessibility of a contrastive interpretation.