Left Dislocation in Spoken Hebrew, it is neither topicalizing nor a construction

1 Introduction

Left Dislocation (or Detachment) (LD) is a well-studied construction attested across many genealogically unrelated languages (see Westbury [2016]; Villa-García [2023] for recent overviews in different theoretical perspectives). The term often encompasses a family of related structures that involve an initial detached constituent with different degrees of syntactic connectivity to the subsequent main clause (López 2016). The classical LD, illustrated for English in (1), consists of a non-clausal NP followed by a clause that refers to it with a resumptive pronoun.

An additional structure of this type is the Hanging Topic (HT) construction in (2). In HT the initial NP sets an interpretation frame for the subsequent clause with no syntactic cross-reference. Gundel (1988: 238) proposes that this structure is universal at least in unplanned discourse.

In addition, based on prosodic (but not syntactic) detachment, Netz and Kuzar (2007) add to this list the Subject Marking construction (SM), shown in (3). In this configuration the initial noun forms a separate Intonation Unit (IU), yet the speaker integrates it as the subject of the clause. As they show, the structure has indeed functional parallels with LD and HT.

The three structures in (1)–(3), dubbed here “LD-type”, gained substantial attention in the field of Information Structure (IS) and played a salient role in its design as a universal cognitive model of linguistic information processing. The initial NP is analyzed as announcing the topic or the interpretive frame for the subsequent clause (e.g., Erteschik-Shir 1997: 53; Lambrecht 1994: 188; among many others). In this view, the construction is used to introduce referents which are required as the topic of the upcoming proposition, but are not yet active at the current stage of the discourse (Lambrecht 2001: 1,073). The reason for resorting to this construction is considered to lie in the cognitive capacity of information processing dubbed the “Principle of Separation of Reference and Role” (PSRR), commonly phrased as “Do not introduce a referent and talk about it at the same time” (Lambrecht 1994: 185). It is argued that establishing the reference (or opening a cognitive file) for the newly introduced referent requires cognitive processing effort on the interlocutor’s side. As a result, it is allocated a processing unit of its own (Prince 1998). This unit is kept separate from the subsequent, similarly effort-requiring step of integrating the segment into a clausal structure and processing new information as about this referent (or storing information in the newly opened file) (Haselow 2017: 109; Kuzar and Netz 2010; Michaelis and Francis 2007). It has also been found (since Prince 1998) that LD-type structures have recurrent discourse functions, such as listing, contrast, parallelism, opening new discourse sequences, marking hedging disagreement, and some others. These are derived in many views from the unified topic-announcing, referent-activating function (Ariel 2008: 126; Gregory and Michaelis 2001; Kerr 2014). Importantly, these studies of LD-type structures commonly assume that the initial noun is detached prosodically, but otherwise pay little attention to phonological, interactional, and production-related phenomena. For instance, examples of LD-type structures often exhibit hesitations or planning difficulties, but are nonetheless accepted as typical LD-type cases (cf. e.g., Gregory and Michaelis [2001: ex. (26) and (31)]; Maslova and Bernini [2006: ex. (11)–(13)]; for Hebrew see examples in Netz and Kuzar [2011]).

Geluykens (1992) situates the IS-approaches in the interaction-oriented perspective, analyzing the function of LD as a collaborative process, where the initial NP negotiates the identity of the referent with the interlocutor. This view can be regarded as connecting the topic-oriented analyses with interactional usage-based approaches. In the latter frameworks, the primary functions of LD are argued to be specific interaction-managing instructions. The set of the instructions is not directly predictable from the purported topic-announcing function but rather comprises several conventionalized usages. Duranti and Ochs (1979) found that LD in Italian is primarily employed to gain access to the floor and block the access of others. An encompassing study of LD in French (Pekarek-Doehler et al. 2015) reveals a range of functions belonging to a few domains: turn-taking, sequence organization, interaction management (assessment, disagreement), and content-management (contrast, listing).

Typically, research on LD-type configurations takes it for granted that (1)–(3) form conventionalized syntactic constructions with systematic functions. Consequently, the examined data is based on examples of LD, HT, and/or SM collected as matching the structural criteria of NP+clause. However, the initial constituent has also been analyzed as an “unattached NP”, namely an NP which forms its own non-clausal unit, used “as part of the process of establishing or tracking discourse referents” (Ono and Thompson 1994: 405). This idea also echoes the view that the initial NP represents a monomial “topic-announcing” utterance (Bickel 1993: 50; Erteschik-Shir 1997: 53; Jakobson 1936).

Building upon the interactional analyses above and developing the monomial view further, the present study questions the assumption that LD, HT, and SM in Hebrew represent a construction/sentence type, let alone that it has dedicated information- or discourse-structuring functions. In a nutshell, the argument is as follows. The original goal of this study was indeed exploring the usage and functions of LD-type structures in natural interaction in spoken Israeli Hebrew. However, the data collection process suggested that this question is biased by the assumption that LD-type configuration constitutes a sentence type deliberately employed by speakers. Instead, the data revealed that the initial unattached NPs in LD-type examples form part of a broader phenomenon of unattached NPs, only a small share of which evolve into an LD-type configuration. Nonetheless, the same factors account for all the occurrences of the non-clausal NPs irrespective of the optional continuation. Consider in this respect the made-up examples in (4). Across all the examples, the speaker starts with an NP occupying its own IU with a final continuing intonation.

The example in (4a) represents the classic case of LD. In (4b), the structure initiated by the first NP is abandoned and the NP does not express the topic of the three subsequent clauses. However, the referent is nonetheless successfully introduced, and is referred to again with a pronoun in the fourth clause. Example (4c) is identical to (4a) except for the fact that the referent has already been active and therefore is referred to with a pronoun. However, although it similarly represents an unattached NP continued with a clause, such examples are dismissed by LD-oriented studies as false starts. Yet if the follow-up clause employs a resumptive pronoun in a non-subject role as in (4d), it does qualify as a LD. For instance, this configuration is attested with possessive clauses in Hebrew illustrated further in (6), which constitute a salient example of LD in the literature on Hebrew (Landau 2009; Netz and Kuzar 2011). In (4e) the initial NP is smoothly integrated in the unfolding clause forming a SM. Finally, (4f) is yet another example classically classified as a false start, although its initial introduction is identical to LD in (4a) and the abandoned introduction in (4b). In the examination of LD-type structures within this broader context, the assumption that LD forms a conventionalized sentence type appears to be questionable, as this syntactic structure would idiosyncratically be sensitive to the distinction between a lexical noun and a pronoun, where the former is considered to be grammatical, while the latter as representing a disfluency, unless there is a co-referential oblique pronoun in the follow-up clause.

From the perspective of dynamically unfolding syntax, the syntactic analyses of (4a)–(4f) are problematic, since identically produced stand-alone items receive dissimilar analyses based on the continuation. As interactional approaches demonstrate, speakers possess no “bird’s eye view” of the entire utterance (Auer 2005; Hopper 2011), and typically the continuation would not have been planned yet upon the delivery of the initial NP across these examples. In the dynamic perspective of unfolding syntax, the initial constituents represent the same phenomenon of a separately produced NP, after which the speaker enters the phase of seeking for a suitable continuation. As such, they are likely to perform a similar set of interactional moves or to be driven by similar basic factors. Selecting solely (4a) and (4d) because their structure fits the pre-assumed definition of LD is biased by the assumption that LD is a sentence type/construction and misses the actual factors underlying the dynamic formation of binominal structures in conversation (cf. Hopper and Thompson 2008). Importantly, (4a)–(4f) already suggest that the detachment is orthogonal to referent activation as (4c) has a pronoun in the initial slot. It also suggests that the interpretation of activating a sentence-level topic is epiphenomenal of cases where the initial slot is a full NP and is followed by a relevant clause, as this effect is not found in (4b), (4c), and (4f). If NP detachment is a process independent of the formation of the subsequent clause, LD-type configurations re-appear as compositional constructs of their more basic constituents. In this case, it would be erroneous to define a highly specific and fully compositional configuration (such as “NP followed by a clause with a resumptive pronoun”) as a basic sentence-structure and selectively study the examples that match it, ignoring configurations where the initial slot is occupied by a pronoun, or the relevant continuation is not a clause or merely absent.

This study shows that the initial NP in LD-type configurations is to be regarded as a constituent with its own contribution, structurally unrelated to and planned separately from the continuation. The local role of the NP belongs to a set of general factors from the domains of interaction management, utterance planning, and attention alignment. In particular, its analysis requires neither the domain-specific notion of topicality, nor coining cognitive constraints to account for it. As a result, this study contributes to the current research which argues for the shift away from the Information Structural categories of topic and focus suggesting that these represent vague interpretational effects based on pre-empirical models of communication (Matić 2022; Matić and Wedgwood 2013; Ozerov 2018). Since these effects do not form analytic primitives, the categories coined on their basis defy coherent definitions with non-metaphorical concepts (cf. “aboutness”, “newsworthiness”, “at-issueness”, “relevance of alternatives”), require domain-specific cognitive/mental categories (Endriss 2009; Lambrecht 1994; Zimmermann and Onea 2011), and are inadequate for the analysis of diverse language-specific phenomena. The proposed theoretical framework links language-specific marking to specific discourse-managing instructions and interactional moves, thus acknowledging and exploring the cross-linguistic diversity of information management (Däbritz 2023; Grzech 2020). The identified categories are linked further to discourse phenomena and domain-general processes, thus both enriching the linguistic analysis and achieving a more parsimonious cognitive model of information processing (Ozerov 2021a).

The paper is structured as follows: Section 2 introduces the framework of interactional syntax, and the methods and data of the study. Sections 3–6 present the findings. Section 3 deals with updating NPs and Section 4 with vocatives, regarded as occasionally entering LD-like configurations but not representing classic LD-type structures. Section 5 explores the interaction-managing and production-based usages of unattached NPs, which occupy a central position in the LD research. Section 6 addresses the reference introducing function of detached NPs. Section 7 summarizes the findings: it discusses the information managing principles and strategies revealed by the study (7.1), and applies these findings to explain the findings of previous studies (7.2). Section 8 concludes the paper.

2 Language background, theoretical framework, study methods, and data

Israeli Hebrew has a nominative-accusative alignment with subject-indexing on the verb. The subject constituent is unmarked for case, while the O-argument exhibits differential marking based on definiteness. The prosodic characteristics of Israeli Hebrew generally resemble Western European languages, such as English (Ozerov 2013). The constituent order is determined by pragmatic considerations with a tendency for SVO.

LD structures are well attested in the language and have been discussed in previous research (cf. Netz and Kuzar [2011: 42] and the references therein). A prototypical LD structure is illustrated in the context of listing in (5).^[1]

Similarly to the research for English and other languages outlined above, the common assumption across different frameworks is that LD structures in Hebrew form a sentence type (Henshke 2015; Landau 2009). For instance, the usage of LD with the expression of possession in (6) gained attention in the generative framework resulting in a debate about its analysis as a double/broad subject construction (Doron and Heycock 2010) or a regular LD (Landau 2009).

The discourse functions observed for LD in Israeli Hebrew resemble the topicality-oriented proposals for English (Landau 2009): topic-introduction, contrast/parallelism of topical referents, and side-sequencing. They are also found in emotive-argumentative discourse, although the reason for that is unclear (Netz and Kuzar 2011: 52).

The corpus for the study is a collection of audio recordings from the Corpus of Spoken Israeli Hebrew (CoSIH) (Izre’el and Hary 2003) (C514_1–2, C612_2–4, C711_0–4, C712_1–2, C714, D142, OCD_1–3, OCh) and six recordings from The Haifa Corpus of Spoken Hebrew, totaling 2:15 h of mostly continuous talk. All recordings represent a natural dialogic speech of multiple (from two to five) participants in their everyday settings.

This study is situated in the interactional “online-syntax” approach (Auer 2015; Hopper 2011). According to this framework, speakers do not possess a bird’s eye view on the entire utterance but plan it in increments as the structure unfolds. This progression through the utterance is accomplished on the basis of possibilities enabled by the current temporarily achieved outcome. For instance, starting with the definite article the, the speaker is forced to continue with a noun phrase, or to restart. A finite transitive verb in I met… foreshadows a (pro)nominal constituent, such as a friend/her/our teacher. These continuation options are very specific. On the contrary, an utterance initial NP leaves the speaker with an ample choice of continuations: for instance, the sun allows the speaker to proceed with a copula is… followed then by an adjective (red) or by a progressive verbal form (rising), to use a verb (shines), to elaborate on the named entity with a relative clause etc. The study examines Detached NPs, defined in this study as follows.

Detached NP: a NP/pronoun that is neither narrowly foreshadowed by the previous structure, nor marked explicitly for its expected role in the upcoming clause by a case marker or a preposition. This NP either occupies its own Intonation Unit(s) with no other material, or is the last constituent of an Intonation Unit, if it is the first and only constituent of a new co-/sub-ordinated structure.

From the perspective of online syntax, this definition reflects the uniformity of cases where the speaker produces a NP unmarked by a preposition or a case marker for its relation to other constituents. This NP initiates a new syntactic structure and yet terminates the Intonation Unit. It is only the following discourse that either leaves this NP as a detached non-clausal unit (Helasvuo 2019; Ono and Thompson 1994) or incorporates it into a larger structure, as in (1)–(3) (LD, HT, and SM). It should also be added that due to this design, this study does not include cases of LD that appear entirely within a single IU (Khan [2016] for Neo-Aramaic; Shor [2020] for apparent “copular sentence” in Israeli Hebrew) and NPs marked by a case or a preposition. Such examples require additional study but are briefly addressed in Section 7.2.

All detached NPs were manually extracted from the corpus, totaling 593 examples. These were coded for the morphosyntactic form (pronoun, noun, relativized noun), referentiality and the pragmatic status of the referent (given, inferable, new, re-activated, exophoric), final prosodic contour of the IU, content-managing and interactional functions (update, stance, turn-taking, contrast), prosodic and syntactic disfluencies (lengthening, prosodic cut off, pause over 0.2 s after the NP, coherent syntactic continuation vs. cut off vs. intervening material before continuation), and LD-related outcome. Thirty percent of the total examples (179 cases) evolved into LD/SM-type structures defined as follows:

1. LD (or “LD-proper”) – a detached NP followed by a clause with a resumptive pronoun

2. “LD-like” – detached NP that deviates from the prototypical LD by some parameters: the relevant clause is distant from the NP with intervening structurally unrelated material (as in [4b]) or with an ambiguous pronominal reference; the lexical form of the NP or the pronoun deviates from the classical case (e.g., the initial NP is a pronoun as in [4c], or the clause repeats the initial noun), or the pronoun does not refer to the same referent but to a relevant notion (such as “This boy – it’s a disaster” in [21])

3. HT – Hanging Topic

4. “HT-like” – as in HT, but the relevant clause is distant from the NP

5. SM – prosodically detached subjects as in (4e)

LD-proper, LD-like, HT, and HT-like cases are dealt below jointly as “LD-type”. However, as discussed extensively below, these distinctions are a product of a retrospective static analysis of the detached NPs jointly with their optional continuation. This analysis is neither available nor relevant for interlocutors in the dynamically unfolding discourse faced with a separately produced NP.

Preliminary findings of this study were published as part of a broader discussion (Ozerov 2021b). Two years after the first round of data coding, the same set of data was coded again by the author, resulting in the agreement rate of 94 % with the first round. Some aspects and definitions here differ from the published preliminary findings resulting in differences between the two studies.^[2] In brief, the findings are as follows. From the overall 593 detached NPs, 114 cases evolved into one of the LD types, and 65 cases into SM, accounting jointly for 30 % of all cases of detached NPs. The rest – 70 % of detached NPs – were neither continued by a clause, nor evolved into a subject of the clause. However, across all cases, the detached NPs have been found to be motivated by the same interactional and attention-managing factors. Table 1 preliminarily summarizes the functions of detached NPs: (1) locally updating NPs, (2) vocatives, (3) NPs that repeat previous material for interactional reasons, (4) NPs that exhibit disfluencies, and (5) NPs performing a local attention-aligning move. Some of the examined factors are independent of one another and overlap (row 5 in the table). Column (a) breaks down the total detached NPs into categories. As usages (1) and (2) do not align with classic definitions of LD, column (b) excludes this data. Column (c) shows the percentage of detached NPs that evolve into a LD-type/SM for each category. Column (d) breaks down the factors for the LD-type/SM, excluding the 70 % detached NPs left with no continuation. Column (e) does the same for LD-type/SM that are not a result of factors (1) and (2).

As can be seen from the Table 1, this small set of general interactional and cognitive factors accounts for the whole of the data. It applies to detached NPs irrespective of whether these are followed by a relevant continuation and result in LD-type structures or not. The same functions apply both to the overall detached NPs and LD-type structures, in particular within the non-updating/phatic usages (columns [b] and [e]), suggesting no need for postulating additional categories or functions, such as topicality, for the analysis of LD. The following sections present the analysis separately for each of the different functions.

3 Content-management by detached NPs

The content-managing, updating (Jakobson’s [1960] “referential”) function of language is the most studied one in linguistics. It is typically associated with a sentential structure and yet a felicitous, grammatically acceptable update in spontaneous language is frequently achieved by a non-clausal constituent (Ono and Thompson [1994]; see Izre’el [2012]; Izre’el [2018] for Israeli Hebrew). Updating detached NPs are commonly attested visualizing the speaker’s perspective or elaborating on a previously introduced referent. Having produced an updating detached NP, the speaker faces a broad choice of possible continuations. Among other possibilities, they may opt for adding more information regarding the referent introduced by the NP. This results in an LD-type outcome demonstrated in (7).

Although (7) appears to have the form of a LD, the initial detached NP spread across (7a) and (7b) contributes novel information, which provides an updating elaboration on a previously mentioned referent. The subsequent unfinished structure (it is in) in (7c) and the assertion (here it is) in (7d) provide an additional update regarding the same referent. Instead of representing the structure of [topic+update] associated with LD, (7) follows the classic Daneš (1974) discourse structure of [update+update]. Updating NPs form the largest group of detached NPs (44 % of total detached NPs). Of these, only 9 % are continued into LD-type collocations and none into SM. However, although such examples have the form of [NP+clause], they would not necessarily qualify as LD cases (Geluykens 1992: 24). The classical LD structure is characterized not only by its syntactic configuration but also by the forward-looking role of the NP, which introduces a referent for the sake of using it as the topic of the subsequent clause.

4 Vocative detached NPs

Vocative NPs are nearly inherently detached prosodically, and some regard detachment as their core property (Hanna and Sonnenhauser 2013). They are typically analyzed as belonging to the phatic domain of language, used to establish and maintain the communicative channel with the interlocutors, and as such located outside of syntax and Information Structure. However, some views linked vocative NPs to topics, and in particular to LD structures (Lambrecht [1996]; Portner [2007] for a formal analysis; Haddad [2020] for an overview of generative syntactic proposals). In these approaches, all detached NPs perform a local attention-regulating move: while topical expressions align attention on a referent, vocatives do so on the addressee. In particular, vocatives resemble LDs with imperative or second person indexing clauses, as can be seen in Hebrew (8) where (8a) and (8b) appear as cross-referenced.

Clearly, the usage of vocatives is not restricted to this very rare configuration, and by no means does a vocative determine the subsequent structure, as illustrated in (9). There is no co-reference or semantic link between (9a) and (9b) here, as happens in 95 % of vocatives.

Irrespective of the continuation, the vocative performs a local phatic function of triggering or monitoring the addressee’s attention, thus establishing or maintaining the communicative channel. In case the addressed referent is mentioned in the subsequent clause, there is a sense of “aboutness” – in which the definition of topic is rooted – that can be attributed to the first NP in a retrospective static analysis. However, this interpretive effect is neither available nor relevant for the interlocutors in the dynamically unfolding conversation. The interlocutors orient at the local phatic move performed by the vocative, and the notion of phaticity is sufficient for its analysis. This notion represents a broad range of family-related anthropological and ethological phenomena (Brudzynski 2021; Davidson and Clayton 2016). These extend beyond linguistic communication into multimodal attention management (through gaze, touch, gesture, and body posture) observed across the animal kingdom, and represent a gradient nature of involvement hierarchies (Zuckerman 2021). As such, their characterization along the lines of discrete exclusively linguistic instructions oversimplifies the nature of this phenomenon, while their linkage to Information Structure (namely, topicality) unnecessarily sophisticates its analysis with domain-specific categories of linguistic information processing. Relying directly on the concept of phaticity, which is independently observed in interactional and ethological research, we can account for the occasional linguistic interpretations of “aboutness”, which epiphenomenally emerge from the collocation of a phatic expression with a subsequent clause in (8) in their static retrospective examination.

5 Interaction management and discourse production

The rest of the paper will look into functions considered to be representative of LD-type structures, where detached NPs foreshadow the referent which can be regarded as having topical relations to the subsequent clause. However, as the analysis shows, this process is not a deliberate topic announcement driven by cognitive limitations on information processing. Instead, in all cases the detachment of the NP is driven by local interactional or attention-managing processes orthogonal to activation and topicality. The apparent LD-type configurations emerge as unplanned products in the process of interaction management. Topicality or aboutness interpretations are epiphenomenal of more basic factors, at which the interactants orient locally without the need for constructing their further mapping on such interpretations as topicality, aboutness, or frame-setting. These interpretive effects are observed only in a static retrospective examination of the entire discourse sequence, and as such are neither available nor relevant for the local meaning-making in interactional discourse.

This section explores interaction managing and production-related factors underlying the distribution of detached NPs. Section 5.1 demonstrates how linguistic material is recycled (Hopper 2011) from previous utterances to re-establish an abandoned direction or to align with the interlocutors’ discourse (“resonance” in Du Bois [2014]). The talk is then continued in a direction relevant to the recycled material. When the recycled item is a detached NP and the continuation is a sentence, the outcome forms an LD-type collocation, but looking solely at LD-type examples misses the interactional phenomenon and its breadth. Section 5.2 proceeds to demonstrate detached NPs that are a product of an online modification of the chosen structure: a speaker begins an utterance using one structure, hesitates, and finally opts for a different construction. NPs typically are a successful (and hence frequent) starting point in unplanned talk in Hebrew. As a result, LD-type outcomes are a frequent product of online restructuring, often found in circumstances where forward planning is hindered, such as turn competition.

5.1 Re-instantiation and resonance

Being a multilateral process, conversation is only partially controlled by each of the participants and can evolve in a new direction without exhausting – in the view of at least one of the interlocutors – a previously addressed issue. As a result, speakers evoke abandoned sequences for further elaboration. A common strategy for achieving this goal is to repeat material that echoes the relevant part. This is illustrated in (10), where this move is performed using a detached NP.

(10)

a.	A: ‘They came to my home yesterday, |’
b.	B: ‘OK /’
c.	A: ‘and asked | whether I would be ready to participate in an experiment | of Tel-Aviv University |’ …1.1…
d.	B: ‘Oh | do you at least get credit points (for that)? /’
e.	A: <laughter> …1.1… ‘[I should have bargained] for that. |’
f.	B: ‘[I can’t belie-]’ –
→ g.	A:		nisuj	ʃel=universitat	Tel.A’ viv |
			experiment	of=university.of	Tel.Aviv
h.	hem		ʁots-im	la.sim a’l-aiχ	em |
	3pl		want-pl	to.put on-2sg.f	uhm
		’ʁamkol-im ||
		loudspeaker-pl
‘An experiment of Tel-Aviv university, | they want to put uhm | loudspeakers on you. ||’
(C514_1_sp1_ 022-032, C514_1_sp2_ 017-19)

Speaker A cooperated first with the side-joke about credit points in (10d) but then resumed her previous account in (10h). The resumption is done by repeating the last constituent from (10c) uttered before the interruption, namely ‘an experiment of Tel-Aviv university’ and then proceeding with the story. This results in the structure of an HT. However, although contextually speaking the last sentence can indeed be interpreted as about ‘the experiment’, the function of this detached NP is neither re-introducing nor announcing that. This material, recycled from the interruption point in its entirety as an indefinite NP, resumes the speaker’s original sequence. With no interruption, Speaker A in (10) would not repeat the NP and the analysis of its referent as the topic of the sentence ‘They want to put loudspeakers on you’ would be unlikely.

The repeated material is not necessarily nominal and therefore detached NPs merely represent an instantiation of this broader phenomenon, as illustrated in (11) where (11c) is repeated from (11a).

(11)

The speakers converse while crossing the street. The side-remark is made as a joke to a driver.

a.	aχʃav	ha’ j-a	’l-anu |
	now	was-3sg	to-1pl
b.	…0.64…	ti-dʁos	o’t-anu |
		2-run.over.imp	obj-1pl
c.	…1.8…	ha’ ja	’l-anuːː |
		was	to-1pl
d.	’eze	’odem |
	some	lipstick
‘Now we had …0.64… Run over us! …1.8… we had | a lipstick.’
(C523: 263.9″–268.7″)

Evoking previously mentioned material reactivates an abandoned sequence, discussion, or stance associated with it. This can be described as setting an interpretation frame or context for the upcoming utterance – properties associated with the concept of topicality. However, this characterization is a by-product of the discourse-structuring function of the repeated material and not its basic property or motivation for usage.

A related phenomenon is echoing the material employed previously by the interlocutor (“resonance” [Du Bois 2014]). This material is re-used to construct the speaker’s talk in relation to their current interactive goals (Goodwin 2013). For instance, repeating the interlocutor’s wording can convey an explicit acceptance of the proposed discourse direction, as is salient in question-answer pairs (12)–(13). In the former, B fully echoes the wording and the construction of A’s question, although this material is fully redundant.

(12)

A:	ma	hu	a ma ʁ /
	what	3sg.m	said
	‘What did he say? /’
B:	hu	amaʁ	ʃe=ze	dej	sababa
	3sg.m	said	that=it	quite	OK
	‘He said it’s quite OK. ||’
	(C711_0_sp2_248, C711_0_sp2_248)

In (13) the same principle applies with the resonating material being a detached NP.

(13)

a.  A:		la’kaχ-nu	tʁemp	le=’moʁon |
		took-1pl	ride	to=pn
b.		…1.3… χa’zaʁ-nu	leː= |	ulan’bataʁ |
		returned-1pl	to=	pn
‘We took a ride (tʁemp) to Mörön, …1.3… went back to | Ulaanbaatar. |’

c. B:		ma	ze	tʁemp /
		what	it	ride
‘What is “a ride”?’

→d. A:		tʁemp	|	ze	ʃe=a’taːː	meda’beʁ	im=ana’ʃim
		ride		it	that=2sg.m	talk	with=people
b=a-e’ʁev		im	’miʃehu	no’sea	le=’Moron |
in=def-evening		if	somebody	goes	to=pn
‘“A ride”, it is that you talk to people in the evening (asking) if there is anybody going to Mörön (the next day). |’
(OCh_sp1_330–336, OCh_sp2_101) (Figure 1)

The arising collocation of a detached NP and the subsequent relevant talk in (13) can appear to follow the IS-principle of the ‘topic-comment’ configuration. However, collecting only [NP+clause] examples of this kind and then resorting to the IS-concept of topicalization to explain their effects (cf. Likhacheva [2010] for French) misses the full scope of the phenomenon and hence its actual function. As can be seen in (13), the interrupting question results in a substantial pause on the side of the addressee. This pause may be interpreted as a lack of cooperation, and for avoiding that Speaker A is forced to display his intent to answer with whatever material is available. He does so by echoing the NP in a deviant high-pitched voice, acknowledging the acceptance of the discourse trajectory suggested by B’s question. Evidently, no continuation has been planned at this stage. The speaker proceeds with the explanatory formula “it is that” and hesitates again on the generic 2sg pronoun. Altogether, he gains over 2 s of planning time, whereas the apparent LD structure emerges from the strategies of structuring unplanned discourse to which he resorts: pausing, resonance, explicative “it is that”, lengthening.

Figure 1:

Resonating detached NP in answer planning.

Finally, consider the usage of re-instantiating and resonating NPs in (14). These do not develop into LD-type structures, demonstrating further that it is an optional collocation and not a predefined construction for which the speakers opt in circumstances of resonance.

(14)

The speakers are trying to recall how much time has passed since their visit to the discussed ski resort.3
		Israeli Hebrew	Translation	Comments, action type
a.	A:	ze ha’ja bid’juk be=ʃmo’nim ve=χa’ me ʃ ||	it was exactly in (19)85. ||	assertion
b.		[ani zo’χeʁ.et]a ||	[I remember.]a ||
c.	B:	[ʃmo’nim ve=χa’ me ʃ]a ||	[Eighty-five.]a ||	resonance, acceptance
d.	C:	[ʃmo’nim ve=’kama]a /	[Eighty what?]a /	echo-question
e.	B:	ʃmo’nim ve=[χa’ me ʃ]b ||	Eighty [five.]b ||	answer, resonance
f.	A:	[ze ha’ja]b aχʁe=ha-ni’tuaχ ʃeli ||	[It was]b after my surgery. ||	providing evidence for the asserted statement
g.	B:	ze ’ʃva-esʁe ʃa’na ||	It’s seventeen years. ||	calculation of the time passed since 1985
h.	A:	ani zo’χeʁ.et ||	I remember. ||	side-sequence opening
i.		ha’laχ-ti ’kodem le=--	I went first to --
j.	B:	ʃmo’nim ve=χa’ me ʃ | ’ʃ va -esʁe ʃa’ na ||	Eighty-five | seventeen years ||	resonance
	(D142_sp1_180–184, sp2_030, sp3_092–096)

1. 3

   Particularly long examples are given only in the transcription and the translation. The rightmost column details the relevant characteristics of the Intonation Unit and the action performed by the speaker using it.

Speaker A recalls the year in which they visited the discussed resort in (14a), paving the way for calculating the time that has passed since. Speaker B resonates with A in (14c). He reasserts it in (14e) and then proceeds to calculate the years passed in (14g). As A attempts to develop the discourse further, B re-instantiates “eighty-five” and reasserts his calculation with another bare NP in (14j). Resonating and re-instantiating NPs here are repeatedly used by B to relate his calculation to A’s assertion but do not produce LD-type collocations.

To sum up, speakers employ re-instantiating and resonating material for a variety of reasons, some of which are forward-oriented and as such relevant for the subsequent talk: returning to an abandoned sequence, selecting a direction from the options enabled by the preceding talk, and explicitly accepting the interlocutor’s proposal for a discourse shift. When the resonating material is a detached NP, the outcome of these cases can have an LD-type configuration. This option is very prominent in the data: it accounts for 37.5 % of the non-updating/phatic detached NPs and 36 % of the non-updating LD-type/SM. A retrospective static examination of the LD/SM examples such as (13) makes indeed an impression of “aboutness” and could pass the usual tests for topicality and Information Structural functions of LD. However, understanding the interactional phenomena clustered under the principle of resonance and re-instantiation sheds light on the sources of these “aboutness” effects. The apparent topical interpretation of the initial NP is a side-product of (i) the broader principle of discourse-foreshadowing by repeated material, resorted to for interaction-structuring purposes, and (ii) the typical interpretation of discourse as relevant to the immediately preceding utterance/action. The two phenomena are known independently of the notion of topicality and account for a broader set of examples. Therefore, an analysis rooted in these phenomena is both richer and more parsimonious.

6 Attention alignment

This section addresses cases where the detached NP appears at first to function according to the classic LD-analysis: it constitutes a separate discourse move which aligns attention on a referent, often as a preparation for a further statement for which this referent is relevant. This appearance notwithstanding, this section argues that the topic-oriented analysis misses the specific function of the detached NP. In the examples below, the detached NPs perform a local function of aligning the interlocutors’ attention on a referent, either mental (O’Madagain and Tomasello 2021) or exophoric. This move enables an array of potential follow-up actions, including the discourse-level option of referring to it later (cf. Ono and Thompson [1994: 404] and the references therein). This view is not tantamount to announcing the proposition topic for various reasons:

1. The requirement of attention alignment constitutes a separate, local goal in the interaction. It is structurally unrelated to the follow-up discourse which is not yet planned at this stage. Only 50 % of attention-aligning NPs evolve into combined LD-type/SM, showing that attention alignment does not stipulate for the follow-up structure.

2. The subsequent action is not restricted to uttering a proposition about this referent but can be of various kinds, including physical (silent) actions performed on the referent, such as locating it in space or considering to buy it.

3. The effect of attention alignment is neither restricted to nor necessarily relates to the immediately following sentence, but can last indeterminately into the discourse.

4. Even if the NP is co-referential with a further constituent, this does not stipulate for its topical role in the proposition.

This section limits the scope of the phenomenon to instances of a deliberately separate move involving an attention shift, excluding cases characterized by prosodic disfluencies or where the NP is recycled for the purpose of maintaining attention on an already attended referent. Consider first exophoric (“situationally evoked” [Prince 1981]) referents introduced for the sake of attention alignment. Their continuation options are diverse and are pursued only after the interlocutors have aligned attention on the referent, as (18) demonstrates.

The speaker announces places which are relevant to his trip and hence can be used in the upcoming narrative. Uttering Yunnan in (18b), he activates a referent from the shared knowledge, as he is looking for it on the map. This act is preparatory for locating that referent and sharing attention on it, which can be minimally embodied by a pointing gesture or accompanied by the deictic presentative hine. The alignment of the joint attention on the announced referent can also be a self-sufficient local goal of the speaker, and not a preparatory step for the immediate continuation. This is the case with Guangxi in (18e) and Khatgal in (18i), as the speaker names places which his interlocutor is merely expected to identify on the map. These referents are potentially relevant for the later narrative but are not developed in the immediately following discourse. The introduction of a referent also allows the speakers to comment and provide more relevant information. Having introduced Mörön in the same way he introduced Guanxi and Khatgal, the speaker opts this time to begin a story about it and refers to the place again with the pronoun ‘from here’. This LD-type outcome is one of the many potential products of structurally and functionally identical detached NPs, which all align joint attention. An introduction of a referent identifiable from the shared knowledge and not exophorically follows the same principles, as (18b) above and (19) show.

The speaker initiates a new sequence foreshadowing by (19c) that it is a multiunit turn culminating in a question (cf. Schegloff 1980). After the interlocutor’s approval of the move in (19d) and a long hesitation (19e), the speaker introduces a detached NP in (19f) ‘the Lebanon war’ as the next prerequisite for this yet unformulated question. This is done with a final falling contour, as if this were an assertion. This utterance transfers the floor and is followed by the interlocutor’s acknowledgment (19g), hesitation (19h), and a meta-question (19i), which all occur before the speaker enters the phase of a very incremental and hesitant planning of the actual question, for which the referent introduced in (19f) must be somehow relevant. As the talk is evidently not yet planned when this detached NP is uttered, the speaker’s goal upon its introduction is to negotiate the joint attention on this referent, as a preparation for further actions. In the first seemingly final formulation in (19o) the pronoun ze ‘it’ appears as a resumptive constituent referring to the war (‘it is cool’), as if it were a long-distance LD. Yet, the speaker proceeds to elaborate the structure further into an evaluative ‘it is cool that Y’ construction in (19p), retrospectively leaving the initial NP with no co-reference.

Example (19) demonstrates how speakers can restrict the scope of their local planning to an initial NP. Aligning attention on its referent is their local goal, the achievement of which allows for a range of desired follow-up actions. However, the nature of these upcoming actions, let alone the syntactic configuration of the subsequent discourse and the referent’s role in the proposition, can remain entirely unplanned at this stage. In (19) we see how the speaker cedes the floor, pauses, hesitates, and recurrently restructures the utterance following the NP ‘the Lebanon war’. The planning of the subsequent discourse starts only after the NP had been delivered. Aligning attention on a referent relevant for the yet unplanned upcoming discourse is sufficient for the analysis, and relies on the domain-general factors of attention and relevance. These factors also account for occasional “aboutness” interpretations, without the need for the cognitive domain-specific notion of topicality (cf. Tomlin [1995] for English).

The speaker in (19) cedes the floor for the interlocutor’s approval of the introduced referent in a nearly perfect accordance with Geluykens’s analysis of LD (1992). However, this combination is very rare in the data (8 % of attention-aligning NPs). Attention-aligning referent introductions often (52 %) occur with a continuing tone as the speaker directly proceeds with the talk. This can be seen in (20): the speaker is driving a car with her friends inside. She breaks the silence by introducing an inferable referent, which similarly to (19), successfully opens a new discourse sequence lasting for the next 40 Intonation Units. However, the speaker immediately incorporates this NP as a subject into a larger syntactic structure, resulting in a SM.

Examples (19)–(20) may also suggest that sequence-initiation is a discourse role of detached forward-looking NPs (cf. Netz and Kuzar 2011; Pekarek-Doehler et al. 2015). However, this effect is again epiphenomenal of an attention shift towards a referent, which cannot be interpreted as relevant to the ongoing discourse. It is not present in (21) where the activated referent links directly to the preceding discussion. This example also shows how achievement of attention alignment can emerge through a multi-step interactional process (“recalibration” [Lerner et al. 2012]).

B introduces a new referent dubbed ‘this child’ as the new attention center relevant to the discussed issue. Since this introduction follows his claim that it is people who tear the fruits down, this move already implies the child’s responsibility. Aligning attention on the child is the local goal of the speaker, while the continuation is yet unplanned. Following this move, he engages in the elaboration on the child’s identity, while simultaneously expressing stance, as he opts for the pejorative construction of a proper name with the definite article and a demonstrative (Averintseva-Klisch 2016). Finally, he proceeds to evaluating assertions in (21c) and (21d). Hence, the first two detached NPs (21a)–(21b) perform separate local moves of gradual attention alignment and evaluation.

To summarize, this section examined LD examples with discourse-new forward-looking NPs. It focused on the local contribution of the NP, the structural and interactional aspects of the online structuring of the utterance, the role of the co-referential expressions in continuation possibilities, as well as the absence of a continuation. All these suggest that the role of the initial NP is to locally align attention on a referent, enabling a range of possible (potentially non-verbal) continuing actions. This kind of analysis may appear to echo the view of the initial NP in LD as a thetic, all-focus structure, recurrently proposed in previous literature (Bickel 1993: 50; Erteschik-Shir 1997: 53; Jakobson 1936). However, the present analysis extends to a broad range of detached NPs, only some of which are continued by a further relevant clausal structure, either as LD-type or as SM. Fifty percent of such cases are not followed by a relevant clause at all, as they either perform a local self-sufficient move or are abandoned. The interpretation of sentence-level topic activation arises only where the attention-aligning move is immediately followed by a sentence to which it is relevant. This interpretive effect can be analyzed directly through the domain-general concepts of attention and relevance, applicable to a broader set of examples. Such LD cases constitute around a third of the overall examples of detached non-updating/phatic NPs (32 %) and of the non-updating LD-type/SM cases (29 %).

7 Discussion

8 Conclusions

This study originally aimed at exploring the structure and function of Left Dislocation/Detachment and Hanging Topic constructions (jointly called here “LD-type”) in spontaneous Israeli Hebrew conversation. However, the dynamic study of the structures as these unfold in natural conversation suggested the need for examining the whole set of initially detached NPs, as these are not restricted to LD-type sentences and represent the same interactional phenomenon. As a result, the selection of examples that end up evolving into LD and the omission of similarly detached NPs that do not do so re-appeared as a procedure biased by the pre-empirical assumption that LD structures form a predefined sentence type. This procedure examines dynamic structures in a retrospective static perspective, irrelevant and unavailable to interactants, and ignores prosodic factors. As a result of these considerations, the study examined NPs that initiate a new syntactic structure and are detached prosodically (defined here as detached NPs) in over 2 hrs of natural Israeli Hebrew conversation. It explored cases with no continuation after the NP, and cases continued into LD-type configurations (NP+clause). The latter cases constitute only a third of all detached NPs examined in this study. After analyzing the functions of the detached NPs, the study proceeded to explore whether LD-type structures justify the analysis of representing a separate sentence type with consistent functions.

The study demonstrated that across all the cases the usage of the NP can be accounted for by its local contribution to discourse or by production reasons. The primary motivations for using a detached NP are commonly known factors of cognition and interaction management: (i) local updating contribution, (ii) repetition of previous talk, (iii) structurally unsuccessful starting points, (iv) attention alignment, (v) phaticity. The same motivations account for all the cases of detached NPs, irrespective of whether they are continued into an LD-type structure or not. Consequently, all the LD-type examples and their functions can be explained through their compositional account as a collocation of the initial NP and of the subsequent talk. Table 4 summarizes the local contributions of the detached NPs and the role of the subsequent clause, shedding light on the principles of interaction management and revealing the sources of effects, topical or otherwise, associated with LD-type collocations. This small set of general interactional and cognitive functions is sufficient for the analysis of detached NPs and LD-type structures, with apparent Information Structural effects of topicality being a by-product of these more general factors.

The attention-aligning function resembles the topicalizing accounts of the LD-oriented research. However, attention alignment is a local move that should not be conflated with sentence-level topicality and which can be accounted for directly by the domain-general phenomenon of attention. Against a sentence-level analysis of topicality, this study finds that the act of attention alignment (i) can have an exclusively local contribution with no subsequent speech, (ii) can be followed by a physical (and not a verbal) action, and (iii) has effects lasting into the upcoming discourse and not necessarily applying to the immediately following clause.

Both the form of prototypical LD-type structures and their functions are compositional constructs of their more basic constituents: a lexical NP activates a referent (Ariel 1990), and a clause communicates a relevant proposition. Due to this compositionality, it is erroneous and unnecessary to regard the overall collocation as a basic construction (Goldberg 2006: 5). Moreover, both LD-proper and HT-proper are very rare in the corpus, making the frequency-based assumption of their conventionalization untenable. The cooccurrence of NP+clause shows no evidence for representing a conventionalized sentence type, which is planned as a single whole and has a coherent function. The opposite is the case, as the initial NP exhibits properties of performing a local interaction- or attention-managing task, before any continuation is planned. The topicality interpretation with respect to the subsequent clause can be derived from the basic pragmatic principle of relevance (Sperber and Wilson 1996). As a result, the overall product of NP+clause is vaguely interpreted as “about it”, or being its topic.

This study adds to the recent approaches that question the validity of, and the need for, the Information Structural categories of topic and focus, and the domain-specific cognitive models based on them. This research argues instead that topical and focal effects are high-level interpretive by-products of language-specific categories, and general discourse-level and cognitive processes (Matić 2022; Ozerov 2021a). The present study disentangles a prototypical example of topical effects and an apparent construction dedicated to their expression. It accounts for a broader range of data (LD-type structures, detached NPs with no continuation, atypical LD-type structures where the detached constituent is a pronoun, and more) and employs a more parsimonious theoretical model. This model relies on general and independently established factors of cognition (attention alignment), pragmatics (relevance), and discourse (recycling and resonance). As such, it dispenses with the need for a dedicated category of topicality rooted in domain-specific models of cognition and a vague concept of “aboutness”.