Discourse Markers in French Belgian Sign Language (LSFB) Dialogues and Their Translation into French: A Corpus-Based Study

Abstract

Discourse markers have been extensively studied in spoken languages from different perspectives, covering monolingual, contrastive, and translation studies. However, research on these items remains limited for signed languages, with only a handful of scattered publications. Following a corpus-based approach, this paper aims to investigate discourse markers in French Belgian Sign Language (LSFB), including their types, functions, and translation/s into written French. An 18 min sample of three dialogues and six signers was analyzed using a two-level independent taxonomy (domain and function) previously applied to spoken and signed data. Overall, 251 discourse markers were identified in the LSFB sample. They can be manual, nonmanual, or a combination of both, the latter type being the most frequent. In contrast to the previous literature, discourse markers cannot be spatial in LSFB. Regarding their functional spectrum, most discourse markers belong to the sequential domain (i.e., they are mostly used to structure discourse) and express ‘addition’ (i.e., providing more information) or ‘monitoring’ (i.e., keeping control over one’s turn or over the interaction). When examining the translation of DMs, most are either omitted or substituted by other non-discourse marking items in the target texts. Although these results are generally similar to previous studies on DMs in spoken languages, more research on these items in other signed languages is needed to obtain a precise overview of their role in human communication.

1. Introduction

Sign language linguistic research has been marked by two major milestones. The first was Stokoe’s (1960/2005) groundbreaking paper, which proved that signed languages are natural languages because they share properties with spoken languages. The second milestone was the advent of signed language corpora in the 2000s, which would not only advance the study of signed languages but also set the scene for the study of signed-to-written translation. Signed language corpora can be defined as large machine-readable datasets fixed in size, which are representative of the languages and their users, recorded in a studio setting, documented with relevant metadata, and made open-access (Fenlon & Hochgesang, 2022; Johnston, 2010; McEnery & Wilson, 2001). In contrast to their written/spoken language counterparts, the specificity of signed language corpora is that, in addition to recordings of signed language data and linguistic annotations, they also contain translations of the productions of signers into the environmental spoken language. Hence, signed language corpora could be defined as unidirectional parallel corpora (Hodge et al., 2019).

Existing signed language corpora mostly contain dialogues between deaf signers guided by a deaf moderator. Data are annotated by trained deaf annotators using ID-glosses, i.e., written labels in capital letters and in the environmental spoken language consistently used to identify signs throughout the corpus, regardless of their meaning in the context (Johnston, 2010). Data are also translated by (teams of) deaf and hearing professionals into the environmental (majority) spoken language. Annotation is a very time-consuming and difficult-to-fund task, which is why some of the largest annotated signed language corpora have at best hundreds of thousands of tokens (Fenlon & Hochgesang, 2022; Schembri & Cormier, 2022). The amount of translated data varies widely from one signed language corpus to another.

So far, signed language corpora have been used as monolingual datasets to study linguistic aspects of signed languages, leaving their parallel dimension unattended. This paper aims to investigate discourse markers (DMs) in a signed language by including both the monolingual and the parallel dimensions. Following a corpus-based approach, the objectives of this paper are as follows: (i) to describe the type of DMs in LSFB (French Belgian Sign Language), (ii) to identify their functions using a taxonomy that has previously been used to annotate signed data (Crible & Degand, 2019), (iii) to examine whether these DMs are translated as other DMs into French, and (iv) to create a first inventory of DMs in LSFB, along with their functions and possible translations. The description of DMs from the perspective of a signed language will provide us with a more comprehensive picture of how these items are used in human communication. From a more practical point of view, this research will also be informative for LSFB learners, translation and interpreting trainees, teachers, and practitioners.

The remainder of this paper is structured as follows. Section 2 summarizes the background of this study, including previous research on DMs in signed languages, existing models to approach the functions of DMs, and some considerations related to translation between two (or more) spoken languages. Section 3 outlines the dataset and the annotation procedure. Section 4 presents the type of DMs found in our dataset, their functions, and how they were translated. Section 5 discusses the findings and proposes future research avenues. Finally, Section 6 concludes the paper.

2. Background

Since the first publication about DMs almost half a century ago (Schiffrin, 1987), research on these items in written and oral spoken language data, covering different aspects and adopting various perspectives, has not stopped growing. Unfortunately, this topic remains largely unresearched in the signed language literature despite some studies outlined below.

2.1. Overview of Discourse Markers in Signed Languages

The first papers on DMs in a signed language were published soon after Schiffrin’s seminal work, i.e., Roy (1989) and McKee (1992) on ASL (American Sign Language). However, these initiatives were only followed by some sparse publications on the same language by Metzger and Bahan (2001) and Hoza (2011), on LSV (Venezuelan Sign Language) by Pérez (2006), on Auslan (Australian Sign Language) by Johnston and Schembri (2007), and on LSE (Spanish Sign Language) by Villameriel (2008, 2010). All these descriptions are based on data produced by deaf signers in narratives or lectures (i.e., monological data). Pérez (2006) and Villameriel (2008, 2010) examined all DMs in their datasets, whereas the other works focused on specific DMs. Furthermore, Villameriel, drawing on his experience as an interpreter, also proposed Spanish translations for DMs in LSE.

More recently, some cross-linguistic and cross-modal corpus-based studies have been conducted (e.g., Gabarró-López, 2018; Crible & Gabarró-López, 2021). Gabarró-López (2018) describes three DMs—namely list buoys (lbuoy),1 the palm-up gesture, and the sign same2 (see Figure 1)—in LSC (Catalan Sign Language) and LSFB, drawing on two samples extracted from the reference corpora of these signed languages (Institut d’Estudis Catalans, 2025; Meurant, 2015).

These three DMs share the same form in LSFB and LSC, but they present cross-linguistic differences and similarities. In the two languages, palm-up is the most frequent and polysemous DM. It is also underspecified as its functions are highly dependent on linguistic context. The second most frequent DM is same, and the least frequent is lbuoy. The three DMs share most functions with their counterparts in the two signed languages, but other functions seem to be language-specific, e.g., same is most often used to express addition in LSFB, but this function does not exist in LSC because there is a dedicated DM with this meaning.

Crible and Gabarró-López (2021) carried out the first known research comparing the use of additive relations between a spoken and a signed language (French and LSFB) using comparable samples taken from oral and signed corpora (Dister et al., 2009; Meurant, 2015). Their study shows that both languages have dedicated additive connectives (en plus ‘in addition’; plus, and, add) and highly frequent polyfunctional ones (et ‘and’; same). However, their functional spectrum is different:

“et in French is monosemous, that is, it only encodes addition even though it can be pragmatically enriched in context (for instance, to express a consequence). By contrast, same in LSFB is polysemous and encodes addition and comparison, which leads to other meanings such as alternative”

(Crible & Gabarró-López, 2021, p. 75)

Crible and Gabarró-López (2021) conclude that the differences and similarities between additive relations across LSFB and French resemble other comparative studies between two spoken languages. The modality only triggers differences in the form; i.e., the DMs investigated in their study are produced with the voice in French and the hands in LSFB. However, the role that DMs play in discourse is similar.

All papers presented in this section only involve manual DMs, and some of them include co-occurring nonmanual elements (e.g., Gabarró-López, 2018; Hoza, 2011; McKee, 1992; Villameriel, 2010). Nevertheless, this description remains incomplete, as Metzger and Bahan (2001) state that DMs in signed languages are not only manual but can also be nonmanual and spatial. This paper is the first to explore these three potential types combined with their translation using dialogical data and a corpus-based approach to move away from partial descriptions and provide a comprehensive picture of the category. Before presenting these data, some background on the existing models to approach DM functions and on the translation of DMs is provided in the next sections.

2.2. Discourse Marker Models Used in Signed Language Research

The long research tradition concerning DMs in spoken languages has developed along with different models for the annotation of their functions. Various research questions, goals, and types of data lie behind each model, resulting in diverse approaches that hinder interoperability. This situation is particularly challenging for DMs in signed languages because research is not only scarce, but the models used are different, so that results cannot be compared cross-linguistically. For instance, Roy (1989) follows Schiffrin (1987), whose model “[is not] specifically designed for systematic corpus application and remains qualitative in nature” (Crible & Degand, 2019, p. 4), and Pérez (2006) and Villameriel (2008, 2010) used taxonomies developed from Spanish (Domínguez Mujica, 2004; Portolés, 2007), which have not been applied to other languages (either spoken or signed).

More recently, Gabarró-López (2018) drew on the three-level annotation model developed by Crible (2017). It was first designed to tackle oral English and French corpus data and later used with written data from these languages (Crible & Zufferey, 2015) as well as oral data from other languages such as Kinshasa Lingalá (Nzoimbengene, 2016) and Slovene (Dobrovoljc, 2016). The model consisted of (i) the type of DM (relational or non-relational), (ii) the domain (ideational, i.e., connecting real-world events; rhetorical, i.e., expressing subjective relations or metacommentary; sequential, i.e., structuring discourse; or interpersonal, i.e., managing discourse exchange), and (iii) the function (30 different possibilities). Each domain had a closed set of possible functions, resulting from a detailed revision of the literature on DMs. Each domain and function came with a fine-grained description with examples and paraphrases.

In the first cross-modal research on DMs between LSFB and French, Crible and Gabarró-López (2021) used the revised version of Crible’s (2017) model (Crible & Degand, 2019). The main differences between the first model and its revised version are that the number of functions is reduced to 15 (addition, agreeing, alternative, cause, concession, condition, consequence, contrast, disagreement, hedging, monitoring, quoting, specification, temporal, and topic), and they are independent of the four domains. In other words, each function can theoretically combine with any of the four domains. According to Crible and Degand (2019), the revised version provides better results in terms of inter-annotator agreement and allows for more options, as there can be up to 60 combinations of domains and functions.

2.3. Discourse Markers and Translation

As mentioned earlier, DMs in a signed language and their equivalent in translation/interpreting in a spoken language have only been addressed by Villameriel (2008). However, this study did not draw on corpus data. By contrast, the translation of DMs between two or more European spoken languages has been notably examined using parallel corpora (see

Table 1. Summary of selected papers on the translation of DMs.

Authors	DMs or Discourse Relations	Language Direction	Data
Crible et al. (2019)	All DMs with a focus on ‘and’, ‘but’, and ‘so’	Unidirectional from English to Czech, French, Hungarian, and Lithuanian	TED Talks (Cettolo et al., 2012)
Cuenca (2008)	‘Well’	Unidirectional from English to Catalan and Spanish	The film Four Weddings and a Funeral and its dubbed versions
Cuenca (2022)	Connective DMs	Unidirectional from Catalan to English	Papers from the Catalan Historical Review and their translations
Dupont and Zufferey (2017)	Concessive relations	Bidirectional between English and French	The News Corpus (Dupont & Zufferey, 2017), Europarl Direct (Cartoni et al., 2013), and TED Talks (Cettolo et al., 2012)
Hoek et al. (2017)	‘Because’, ‘so’, ‘also’, ‘in addition’, ‘although’, ‘but’, ‘if’, and ‘unless’	Unidirectional from English to Dutch, German, French, and Spanish	Europarl Direct (Cartoni et al., 2013)
Zufferey (2016)	Dans la mesure où (‘insofar as’), en effet (‘indeed’), and or (‘now’, ‘but’)	Unidirectional from French to English, German, and Spanish	Europarl Direct (Cartoni et al., 2013)
Zufferey and Cartoni (2012)	Causal relations	Bidirectional between English and French	Europarl Direct (Cartoni et al., 2013)

for a selection of studies). Some studies have focused on a connective (e.g., Cuenca, 2008) or specific discourse relations (e.g., Dupont & Zufferey, 2017; Zufferey & Cartoni, 2012), whereas other studies have delved into more than one connective and more than one discourse relation (e.g., Crible et al., 2019; Cuenca, 2022; Hoek et al., 2017; Zufferey, 2016). Despite the different items examined, these papers have in common that English is one of the languages studied. Except for Cuenca (2008, 2022), most papers worked on data that were first produced orally, then transcribed onto paper, and finally translated.

The types of data and the discourse genres may have triggered differences regarding translation strategies. Cuenca (2022) analyzed scientific texts originally produced in written form and found that literal translation is the most frequent strategy used for DMs. By contrast, the authors who worked on the transcribed productions of European deputies and TED talk presenters, or the dialogues of a film, found that the omission of DMs is the preferred translation strategy (e.g., Crible et al., 2019; Cuenca, 2008; Dupont & Zufferey, 2017; Hoek et al., 2017; Zufferey, 2016). Therefore, it can be concluded that the latter strategy is related to oral interactive texts and underspecified markers such as and, but, and so (Cuenca, 2022).

This paper adds to this body of research by analyzing a completely different multimodal dataset that includes LSFB interactive dialogues on various topics and their translation into written French. Consequently, this study contributes to having a more diverse representation of genres in which all DMs are analyzed, including interpersonal ones. Because of the type of data, a high proportion of DM omission is expected, particularly with DMs such as palm-up, which is both underspecified and highly frequent in interaction (Lepeut & Shaw, 2022).

3. Methodology

The sample used for the present study was extracted from the LSFB Corpus (Meurant, 2015). Featuring 100 signers from different parts of French-speaking Belgium (i.e., Wallonia and Brussels), the LSFB Corpus is the reference dataset for linguistic research on this signed language. Signers were invited in pairs to the University of Namur studio, where they discussed different topics proposed by a deaf moderator. The setting and the instructions provided to participants ensured that the data were highly interactive. Before the metadata collection and the recordings, participants signed an informed consent presented to them in LSFB and written French.

The LSFB Corpus contains 84 h of dialogue, of which 28 have been annotated for the manual activity in ELAN (Sloetjes & Wittenburg, 2008) by a team of deaf annotators. The annotation guidelines are based on the first large-scale signed language corpus, namely the Auslan Corpus (Johnston, 2019). Following the trend of other corpora, the LSFB Corpus also contains the translation of some videos into French. The team of professional deaf and hearing translators has so far translated “25 h of LSFB data […], encompassing 18,000 French sentences or 220,000 words” (Gabarró-López et al., 2024). Since translations are not only to be used for linguistic research but also in a broader context, including deaf students’ instruction or a non-expert audience interested in LSFB, the team was instructed to produce target texts in which the French remained as natural as possible.

3.1. The Sample

Three dialogues produced by three pairs of signers were selected for the present study. The criteria that guided the selection of this sample, which totals 18 min, are the diversity of discourse genres (see

Table 2. Summary of the LSFB sample.

Task	Topic of the Dialogue	Genre	Duration	Signer’s Code
3	Childhood memories	Narrative—explicative	4′53″	S055
				S056
4	Deaf or hearing: advantages and disadvantages	Argumentative—explicative	3′14″	S040
				S041
15	Your hobby, job, passion: equipment, actions, movements, rules, etc.	Descriptive—explicative	10′45″	S005
				S048

), the availability of a translation of the dialogues (i.e., not all annotated videos have been translated yet), the (future) possibility of comparing the same signers’ dialogical productions with other monological productions available, and the (future) possibility of comparing the translations with available interpreted renditions.

This sample contains 2335 manual tokens for the LSFB dialogues and 3283 words for their translation into written French.

3.2. Identification of Discourse Markers and Annotation

In line with previous research on LSFB, all DMs were manually identified in the selected sample following Crible (2017, p. 58), who defines them as

“a grammatically heterogeneous, syntactically optional, multifunctional type of pragmatic markers. Their specificity is to function on a metadiscursive level as procedural cues to constrain the interpretation of the host unit in a co-built representation of on-going discourse. They do so by either signaling a discourse relation between the host unit and its context, expliciting the structural sequencing of discourse segments, expressing the speaker’s meta-comment on their phrasing, or contributing to the speaker-hearer relationship.”

Thus, videos were inspected, and any manual (i.e., signs or gestures produced with the hands), nonmanual (i.e., movements of the head or body, and facial expressions), or spatial use (i.e., signing in different loci in front of the signer’s body) that matched this definition was annotated as a DM in ELAN. A deaf native LSFB signer, who has worked as an LSFB teacher for more than 20 years and has led various projects on describing this language, was consulted to verify that the annotation decisions on what was a DM and what was not were accurate. Besides the existing three tiers of basic annotation for each signer (one for the right hand, another for the left hand, and another for the translation), six additional tiers were added:

• Type_DM: This is a parent tier used to annotate whether the DM was manual, nonmanual, spatial, or a combination without any a priori distinction between the different possibilities.
• Articulators_DM: This is a child tier of the Type_DM tier used to annotate the specific articulators that produced the DM. When several articulators (of any type) were used, they were annotated in the same line and separated by “+”.
• Domain_DM: This is a child tier of the Type_DM tier used to annotate the domain attributed to the DM. A controlled vocabulary with the four possible domains (i.e., ideational, rhetorical, sequential, and interpersonal, see Section 2.2) was created and associated with this tier.
• Function_DM: This is a child tier of the Type_DM tier used to annotate the function attributed to the DM. A controlled vocabulary with the fifteen possible functions (addition, agreeing, alternative, cause, concession, condition, consequence, contrast, disagreement, hedging, monitoring, quoting, specification, temporal, and topic, see Section 2.2) was created and associated with this tier.
• Translation_DM: This is a child tier of the Type_DM tier used to annotate whether the DM was translated as another DM in French (in which case the DM in French was copied from the translation tier) or not (NO-DM).3
• Comment_DM: This is a parent tier in which comments related to any of the previous tiers were introduced, if necessary.

Once the three files were annotated, the next step was extracting data in Excel sheets to analyze them and shed light on the type of DMs used in the sample, their functional spectrum, and how they were translated.

4. Results

There are 251 DMs in the sample produced by the six signers in the three dialogical genres. Their form is described in Section 4.1 below.

4.1. Types of Discourse Markers in LSFB

After applying Crible’s (2017) definition, most DMs identified in the sample are produced with the main articulatory channel of signed languages, namely the hands (see

Table 3. Distribution of the types of DM in the sample.

Type of DM	Absolute Frequency	%
Manual + nonmanual	210	83.7
Nonmanual	22	8.7
Manual	19	7.6
Total	251	100

). In most cases (210 tokens), the DM is produced by a combination of the hands with nonmanual elements. These nonmanual elements are mostly mouthings (i.e., a word in spoken French articulated with a sign) and, to a lesser extent, head movements. Other nonmanual items such as body movements, eyebrow activity (i.e., furrowed or raised brows), eye activity (e.g., closed or squinted eyes, blinking), changes in eye gaze direction, and mouth actions (i.e., movements of the mouth that do not reproduce words) were seldom used with DM. Only 19 DMs were exclusively produced by one or two hands. Altogether, manual DMs (combined with nonmanual marking or not) amount to 25 different forms illustrated in Appendix A.

An example of a manual DM combined with nonmanual marking is presented in (1).4 The signer is recounting a memory related to speaking and signing during her time at school. She produces one DM, namely mais, which co-occurs with the mouthing mais (‘but’ in French). It expresses concession, defined as “a logical counter-expectation between two facts” (Crible & Degand, 2019, p. 26).

Example (1)—CLSFB2703_S056_00:01:15.655–00:01:21.192
pt:loc	ecole	integration	bien	ami	parler
pt:loc	school	integration	good	friend	speak
bien	mais	soir	savoir	aller	parents
good	but	evening	know	go	parents
avoir	sl	la-bas
have	sign-language	there
‘When I was integrated into a mainstream school, it was good because I could speak with my friends. However, when I went home in the afternoon, I knew I could sign with my parents.’

Regarding the other two types of DMs mentioned in the sign language literature, i.e., nonmanual and spatial DMs (Metzger & Bahan, 2001), almost no occurrences related to LSFB were found. Some clarification is in order. On the one hand, there are 20 cases in which the signers produced a mouthing that was not associated with a sign, resulting from contact between LSFB and French (i.e., alors ‘then’, mais ‘but’, ok ‘ok’, oui ‘yes’, parce que ‘because’, si ‘if’, and voilà ‘that’s it’). Although these cases cannot be strictly considered instances of nonmanual DMs in LSFB, as they are not mouthings used to disambiguate the meaning of a sign, they were taken into account as they contributed to the unfolding of communication between deaf signers. There are only two cases in which a head nod and puffed cheeks were used as DMs. On the other hand, no uses of space were found to be discourse-marking. Space has been said to express contrast in different signed languages, and LSFB is among them (Lombart, forthcoming). However, the excerpts where space is used in the sample do not match Crible’s (2017) definition of DMs. When expressing a contrast between two chunks of discourse and other related functions, such as concession, the use of space is mandatory. Therefore, it does not meet the criteria of syntactic optionality.

In example (2), the signer talks about the friends she used to have while she was at school. She marks a contrast between her deaf family and her hearing friends. This contrast is expressed with a body lean and a change in her signing space; i.e., her family is placed on the left side (from the second to the seventh picture), and her friends are placed on the right side (from the ninth to the sixteenth picture). Even though this contrast could be emphasized by adding a DM such as mais, the use of space is necessary. As a matter of fact, the signer of example (1) produces this manual marker to express a concession, but she also uses her signing space, i.e., talking at school is placed on her left side (from the first to the sixth picture), whereas signing at home is placed on her right side (from the ninth to the fifteenth picture).

Example (2)—CLSFB2703_S055_00:02:55.696–00:03.00.960
pt:pro1	famille	palm-down	sourd	palm-down	ami
pt:pro1	family	palm-down	deaf	palm-down	friend
sourd	palm-up	ami	entendant	oui	ecole
deaf	palm-up	friend	hearing	yes	school
pt:loc	avoir	plusieurs	entendants
pt:loc	have	several	hearing
‘I had my family, whose members are deaf, but I didn’t have deaf friends. I had hearing friends, I had some hearing friends at school.’

In the studied dataset, space was also employed to express a sequence of events, a consequence, or a specification. In these cases, the use of space is optional as signers may express the same meaning using other devices (such as manual DMs). However, these uses of space do not indicate how the host unit should be interpreted with respect to the other units. The content of the units is what lets the addressee know the relationship between the units. Therefore, these cases do not fully meet Crible’s (2017) definition of DMs either.

In example (3), the signer describes the paleo diet. He uses the space on his right side (from the first to the eighth picture) to say that this diet consists of eating what people in the past ate, and his central space (from the ninth to the fourteenth picture) to specify that they got this food from hunting. The two chunks of discourse could be produced in the same space, and the second could have been introduced by the DM exemple (‘example’, see Appendix A). However, we know that the second part is a specification of the first because of the propositional content of the two, not because of the use of space.

Example (3)—CLSFB2315_S048_00:01:08.861–00:01:18.051
home	avant	habiter	grotte	dans	suivre
man	before	live	cave	inside	follow
manger	avant	chase	animal	fleche	ds:lancer-fleche
eat	before	hunt	animal	arrow	ds:throw-arrow
mange	animal
eat	animal
‘One must follow the diet of the men who lived inside caves long ago. They chased animals using arrows and ate these animals.’

After describing the types of DMs, the following section focuses on the functional spectrum of the 251 instances of DMs (manual, nonmanual, and manual combined with nonmanual items) found in the LSFB sample.

4.2. Domains and Functions of Discourse Markers in LSFB

The sequential domain is the most frequent in the data, followed by the ideational, rhetorical, and interpersonal domains (see

Table 4. Distribution of domains in the sample.

Domains	Absolute Frequency	%
Sequential	93	37.0
Ideational	88	35.1
Rhetorical	39	15.5
Interpersonal	31	12.4
Total	251	100

). The sequential and ideational domains, on the one hand, and the rhetorical and interpersonal domains, on the other, have similar frequencies. However, the latter are much less frequent than the former. This distribution suggests that DMs in LSFB are mainly used to structure discourse (sequential domain) and to relate real-world events (ideational domain), rather than conveying the signer’s attitude or metacommentary (rhetorical domain) or contributing to the management of exchange (interpersonal domain). The low frequency of the latter category may appear unexpected in the highly interactive sample analyzed and may indicate that signers draw on other mechanisms for the smooth unfolding of dialogues, such as sign holds (Lepeut, 2022) or backchanneling cues (Lepeut & Gabarró-López, in press).

All the functions of Crible and Degand’s (2019) taxonomy were found in the sample (see

Table 5. Distribution of functions in the sample.

Functions	Absolute Frequency	%
Addition	35	13.9
Monitoring	34	13.5
Concession	24	9.5
Agreeing	23	9.1
Condition	22	8.8
Specification	20	8.0
Topic	19	7.6
Cause	18	7.2
Alternative	16	6.4
Temporal	15	6.0
Consequence	10	4.0
Hedging	10	4.0
Contrast	3	1.2
Disagreement	1	0.4
Quoting	1	0.4
Total	251	100

). The most frequent ones are ‘addition’ (mostly expressed by lbuoy ‘list buoy’) and ‘monitoring’ (mostly expressed by palm-up/voila [‘palm-up/that’s it’]), followed by concession (almost exclusively expressed by mais [‘but’]), and agreeing (mostly expressed by oui [‘yes’] or c-est-ca/palm-up/voila5 [‘that’s it/palm-up/that’s it’]). Hence, signers used DMs to provide additional information; keep control over the turn or over the interaction; express counter-expectation, nuance, or contradiction; and express a conforming response (see Appendix B for a list of all the functions and markers that can express them).

Thirty-five different domain–function combinations were identified (see Appendix C). Two functions, namely ‘consequence’ (i.e., indicating that the segment following the marker is the effect or result of the previous one) and ‘specification’ (i.e., giving more details), have variants in the four domains. The following examples (from 4 to 7) illustrate the function of specification across the four domains.

Example (4), ideational domain—CLSFB2315_S048_00:02:05.016–00:02.12.390
nom	exercice	titre	convenir	mouvement	corps	convenir	exemple
name	exercise	title	match	movement	body	match	example
utiliser	ds:barre	ds:disque	ds:lever
use	ds:bar	ds:disc	ds:lift
‘The exercise consists of body movements. For instance, we lift barbells.’

Example (5), rhetorical domain—CLSFB2315_S048_00:00:57.797–00:01.00.951
avoir	dire	titre	developper	dix	titre	dix	point
have	say	title	develop	ten	title	ten	point
programme
program
‘There are 10 capacities, so 10 points in the program.’

Example (6), sequential domain—CLSFB1904_S041_00:03:08.521–00:03.12.606
pt:pro1	savoir	oui	palm-up	dependre	niveau	social	variable
pt:pro1	know	yes	palm-up	depend	level	social	variable
‘Yes, I know. In fact, it depends on the social level.’

Example (7), interpersonal domain—CLSFB1904_S040_00:00:16.653–00:00.25.556
<S041>	besoin	vie	battre	objectif	vie	avancer	titre
	need	life	fight	objective	life	advance	title
aussi	developper	<S040>	ca-veut-dire	monde	sourd	toujours	developper
also	develop		it-means	world	deaf	always	develop
toujours
always
‘<S041> [Deaf people] need to fight for their lives so that they can move forward. They sort of need to develop. <S040> So the deaf world is always developing.’

Example (4) is taken from S048’s description of the types of exercises practiced in Crossfit. The segment of discourse introduced by exemple (‘example’) “gives more detailed information about the previous segment” (Crible & Degand, 2019, p. 27). In example (5), produced earlier by the same signer, the segment of discourse introduced by titre (‘title’, second token in the excerpt) provides details that the signer deems appropriate about what he said earlier. Example (6) is extracted from the dialogue about the advantages of being deaf or hearing. S040 says that deaf children from deaf families do not fight for their rights, and S041 (an LSFB activist) answers that he was born into a deaf family. Then, S040 replies that he is a different case, which is followed by S041’s excerpt in (6). The segment introduced by palm-up specifies a referent that was introduced earlier, i.e., deaf children born into deaf families. Example (7) was produced earlier in the same dialogue as (6). S040 uses ca-veut-dire (‘it-means’) to introduce a segment that specifies what S041 says about deaf people’s need for continuous development.

Most functions in the sample have two or three domain variants, whereas ‘topic’, ‘quoting’, and ‘disagreement’ are domain-specific, i.e., the first two are only found in the sequential domain, and the latter in the interpersonal domain. The other functions may combine with a preferred domain. For instance, ‘agreeing’ is mostly interpersonal; ‘alternative’, ‘condition’, ‘contrast’, ‘specification’, and ‘temporal’ are mostly ideational; ‘cause’ is mostly rhetorical; and ‘consequence’ and ‘monitoring’ are mostly sequential. However, three functions have more than one preferred domain: ‘addition’ is equally ideational and sequential; ‘hedging’ is equally rhetorical and sequential; and ‘concession’ is equally ideational, rhetorical, and sequential (see Appendix C). Regardless of the number of possible variants and the preferred combinations, the most frequent domain–function combinations are ‘sequential monitoring’ (28), ‘ideational condition’ (21), ‘sequential topic’ (19), and ‘interpersonal agreeing’ (19). These combinations are exemplified below (from 8 to 11).

Example (8)—CLSFB2703_S055_00:02:08.737–00:02:08.971
pt:pro1	monde	sourd	oui	c-est	aller	maman
pt:pro1	world	deaf	yes	it-is	go	mother
‘For me, the deaf world was… yeah, when I went with my mum [to the museums].’

Example (9)—CLSFB2315_S005_00:08:31.024–00.08.31.112
si	moins	onze	point	jeter	cartes	jeter
if	less	eleven	point	throw	cards	throw
‘If they have less than eleven points, they must discard the cards.’

Example (10)—CLSFB2315_S005_00:05:42.159–00:05:44.880
palm-up	pt:pro1	adorer	pt:pro1	two	lbuoy(2):two
palm-up	pt:pro1	love	pt:pro1	two	lbuoy(2):two
‘So I love two things.’

Example (11)—CLSFB1904_S040_00:00:47.702–00:00:52.406
<S041>	pt:poss1	frere	soeur	voir	pt:pro3	different
	pt:poss1	brother	sister	see	pt:pro3	different
<S040>	palm-up	oui	reserve	pt:poss1
	palm-up	yes	reserved	pt:poss1
‘<S041> My brother and my sister are different. <S040> That’s it, yes, they are more reserved.’

In (8), the signer explains what the deaf world meant to her when she was a child. She pauses to think about what comes next and afterward produces the DM oui (‘yes’) as an instance of self-monitoring, i.e., keeping control over her turn. In (9), the signer summarizes the rules of a card game. She presents a possible scenario (i.e., the condition introduced by si, ‘if’) followed by its logical effect. In (10), the same signer says the two things she likes doing the most. Before this excerpt, the other signer (S048) had the floor. He finished his turn, and S005 took the floor and opened a new topic with palm-up. In (11), S041 talks about the attitude of his brother and sister towards deaf activism. S040 expresses a conforming response by articulating two DMs, namely palm-up and oui.

If we turn to specific DMs, the most polyfunctional ones in the sample are also the most frequent, namely aussi (‘also’, 20), ca-veut-dire (‘it-means’, 12), mais (‘but’, 21), and palm-up (30, including c-est-ca and voila). aussi can express six functions, which can be combined with three domains (see Appendix A). Conversely, the other three markers can be used across the four domains. ca-veut-dire and mais can express four functions, whereas palm-up is the most polyfunctional DM with eight different functions (see examples 6, 10, and 11). It comes as no surprise that aussi, mais, and palm-up are so polyfunctional as the first two convey basic discourse relations, and the latter is underspecified. Therefore, the three can be enriched in context. ca-veut-dire does not express a basic discourse relation, but it can have a variety of uses related to its core meaning of reformulation.

Other frequent but less polyfunctional DMs can express two domains and one function (e.g., apres [‘afterward’], ideational and sequential domains, and temporal function) or one domain and two or more functions (e.g., gsign/palm-down, sequential domain, consequence, monitoring, and topic functions). Finally, some DMs are dedicated to a particular domain and function (e.g., ou.ou [‘or’], five ‘ideational alternative’, or suivant [‘next’], three ‘ideational temporal’). These dedicated DMs are amongst the least frequent in the sample. Still, they can be as frequent as other more polyfunctional markers (e.g., titre [‘title’], five occurrences of the rhetorical domain, four expressing hedging and one specification, or plus [‘plus’], three occurrences expressing addition, one of the ideational domain and two of the sequential domain).

4.3. Translation of Discourse Markers from LSFB to Written French

Only one-third of DMs in the LSFB sample were translated into French using other DMs (i.e., 73 occurrences). The translation could be literal or not (i.e., the target DMs could be more general, specific, or semantically different compared to the source DMs). As a general trend, DMs were most often translated literally (i.e., 54 cases), whereas divergent translations were not frequent (i.e., 15 cases). In four cases, the translation could not be considered literal or divergent as the DM was palm-up, which is polyfunctional, and its different meanings are varied and sometimes idiosyncratic. Since DM–DM translation is the least frequent option and its analysis is quite complex, the remainder of this section focuses on the cases where DMs were not translated by other DMs (i.e., 178 occurrences). The frequency and percentage of these two possibilities are presented in

Table 6. Distribution of DMs in terms of translation.

Translation of DMs in LSFB	Absolute Frequency	%
Not translated by another DM in French	178	71.0
Translated by a DM in French	73	29.0
Total	251	100

.

Among the NO-DM translations, most DMs were omitted in French (149 occurrences), and the others were substituted by other non-discourse-marking items such as conjunctions, adverbs, emphatic pronouns, etc. (29 occurrences). These two possibilities are illustrated below. Example (12) is the translation of example (8), whose glosses are provided again for the reader’s convenience. oui is omitted (see ‘Ø’ in the place where it would appear if translated), and the translator divides the source sentence into two. Example (13) contains the translation in French of example (10) and the glosses provided earlier. palm-up is translated by the emphatic pronoun moi (‘me’), which is used to signal that the topic of discourse shifts to what the other signer likes.

Example (12)
pt:pro1	monde	sourd	oui	c-est	aller	maman
pt:pro1	world	deaf	yes	it-is	go	mother
Il y avait, d’un côté, le monde des sourds. Ø Je me rappelle encore quand j’allais au musée avec ma maman.
‘There was, on the one hand, the deaf word. Ø I remember when I went to the museum with my mum.’

Example (13)
palm-up	pt:pro1	adorer	pt:pro1	two	lbuoy(2):two
palm-up	pt:pro1	love	pt:pro1	two	lbuoy(2):two
Moi, j’adore deux choses.
‘Me, I love two things.’

When examining the type of relations most frequently omitted/substituted, we find that sequential markers go first, followed by ideational, rhetorical, and interpersonal markers (see

Table 7. Distribution of DM omission/substitution in translation across domains.

Domains	Absolute Frequency of Omission/Substitution	% Regarding the Total	% Within the Domain
Sequential	73	41.0	78.4
Ideational	55	30.9	62.5
Rhetorical	32	18.0	82.0
Interpersonal	18	10.1	58.0
Total	178	100	----

). This distribution is related to the general frequency of DMs in the sample (see Table 4). However, the order changes if we focus on the percentage of omission/substitution within each domain. Rhetorical DMs are most frequently not translated by a French DM, followed by sequential, ideational, and interpersonal DMs. In other words, the translation of a DM by means of another DM is the strategy used in 40% of cases of the ideational and interpersonal domains, whereas this strategy is only employed in 20% of the cases belonging to the sequential and rhetorical domains. The lowest percentage of DM omission/substitution in the interpersonal domain goes against the expectation that it would be the most frequently not translated domain, as it is closely related to face-to-face communication but not to written texts (Cuenca, 2022).

Non-connective functions (‘monitoring’, ‘hedging’, ‘disagreement’, and ‘quoting’) were most often omitted/substituted (see

Table 8. Distribution of DM omission/substitution in translation across functions.

Functions	Absolute Frequency of Omission/Substitution	% Regarding the Total	% Within the Function
Monitoring	31	17.4	91.2
Addition	26	14.6	74.3
Concession	18	10.1	75.0
Agreeing	10	5.6	43.5
Condition	14	7.9	63.4
Topic	14	7.9	73.7
Specification	13	7.3	65.0
Cause	13	7.3	72.2
Alternative	12	6.7	75.0
Hedging	9	5.0	90.0
Consequence	8	4.5	80.0
Temporal	7	3.9	46.7
Contrast	1	0.6	33.3
Disagreement	1	0.6	100.0
Quoting	1	0.6	100.0
Total	178	100	----

). This may be explained by the fact that ‘monitoring’ and ‘quoting’ are mostly expressed by palm-up, which tends to be omitted or implied in LSFB-to-French translation (Gabarró-López & Heylens, 2025) due to its underspecified nature (Crible et al., 2019), see below. Moreover, ‘monitoring’ includes both interactive and punctuating uses, the latter being frequently omitted in multilingual parallel corpora (Crible et al., 2019). As for ‘hedging’, the preferred translations are phrases such as c’est un peu comme si (‘it’s like’) or on pourrait dire que (‘it could be said that’) instead of a DM. Conversely, ‘agreeing’, another non-connective function, is one of the functions most frequently translated by a French DM in the sample. The function of ‘agreeing’ is mostly expressed in the LSFB source data by the sign oui, which is not polyfunctional and whose direct equivalent is oui (‘yes’). According to Crible et al. (2019) and Hoek et al. (2017), the less polyfunctional DMs do not tend to be omitted. Overall, connective functions present a higher percentage of translation from LSFB into French DMs (35% on average) than non-connective functions (15% on average). However, there is a lot of variation depending on the connective function. A closer look at translation strategies is outside the scope of this paper but could shed light on this variation.

As regards the different DMs in the sample, frequent underspecified ones such as palm-up and aussi (‘also’) tend to be omitted/substituted (3 translations/30 tokens and 2 translations/20 tokens, respectively). However, other frequent underspecified DMs, such as mais (‘but’), show a higher translation rate (9 out of 21) similar to dedicated DMs, such as si (‘if’, 8 out of 21) and oui (‘yes’, 12 out of 22). This finding is in line with what Crible et al. (2019) and Hoek et al. (2017) found for the translation of English ‘but’ into other languages “as a result of [its] strong adversative meaning” (Crible et al., 2019, p. 149). Therefore, the fact that underspecification does not equally affect all DMs, their functions, and their translations is not only related to spoken languages but also applies to signed languages.

5. Discussion

This paper has described the type of DMs in LSFB, their functional spectrum, and their translation into French using another DM or not. After applying Crible’s (2017) definition of DMs to an 18 min LSFB sample containing six signers and three dialogues, it has been shown that DMs can be manual, nonmanual, or a combination of both. However, they cannot be spatial in LSFB. Although space is part of the grammar of this language, space is used to provide discourse cohesion (in the sense of referential or anaphoric continuity) rather than coherence, understood here as the signaling discourse relations between units. This finding, which partially contradicts Metzger and Bahan (2001), may be extended to other studied Western signed languages such as LSC. Still, further research should be conducted to determine whether other signed languages, particularly non-Western and village signed languages, behave similarly or not in this respect.

DMs in LSFB can express all domains and functions of Crible and Degand’s (2019) taxonomy. From a contrastive perspective, some aspects of the present study on LSFB are similar to what these two scholars reported about their French sample and what Crible (2017) found for English. For instance, the sequential domain is the most frequently used in the datasets, whereas the interpersonal domain is the least frequent. This indicates that both signers and speakers mainly use DMs to structure their discourse. The lower frequency of interpersonal DMs in these highly interactional samples may indicate that both signers and speakers use other devices for the management of discourse flow (e.g., commentaries/backchannel signals, see Lepeut & Gabarró-López, in press; Crible et al., 2024). The main difference in terms of domain distribution is that the rhetorical domain is the second most frequent in the French and English samples, but this position is occupied by the ideational domain in LSFB.

Regarding the functions, ‘addition’, ‘concession’, and ‘monitoring’ are the three most frequent in LSFB and French (even though the distribution is different); ‘contrast’, ‘disagreement’, and ‘quoting’ are the least frequent; and ‘alternative’, ‘cause’, ‘specification’, and ‘topic’ are in between the two extremes. The most striking difference is related to ‘agreeing’ and ‘condition’. They stand as the fourth and fifth most frequent functions in LSFB, but the former did not appear in French (although it has been attested in other samples, e.g., Crible, 2017), and the latter was only used twice (0.5%). In both datasets, there are more than 30 combinations of domains and functions. Some functions are domain-specific, i.e., ‘disagreement’ is always interpersonal, and ‘quoting’ and ‘topic’ are always sequential. The most frequent combination in both samples is ‘sequential monitoring’, so both signers and speakers frequently use DMs to “keep control over the turn/discourse […] usually in contexts of hesitation, stagnation” (Crible & Degand, 2019, p. 27).

The differences between the French, English, and LSFB samples may be related not only to language but also to other features such as the register (the French and English samples combine formal and informal conversations, while dialogues in the LSFB sample are informal), the genre (argumentative–narrative, descriptive–explicative, and narrative–explicative dialogues in LSFB, but no information about the French sample used in Crible and Degand (2019), and a mix of “free conversations, phone calls, face-to-face interviews, radio interviews, classroom lessons, sports commentaries, political speeches and news broadcasts” used by Crible (2017, p. 87) in English), and the speakers or signers. Moreover, further intra-modal and cross-modal differences could be found if both prepared and non-prepared monologues were analyzed. An interesting future avenue for research would be to compare this LSFB sample with a sample from the FRAPé Corpus (a comparable corpus of multimodal French, Meurant et al., forthcoming) to achieve direct comparability.

When focusing on specific DMs, basic DMs (i.e., highly frequent DMs with an associated function such as ‘and’ expressing addition in English) are not always the most frequent type, as reported in other studies (e.g., Crible & Degand, 2019; Crible et al., 2019). mais (‘but’, 21) and aussi (‘also’, 20) are the only basic connectives among the most frequent DMs; the others are, in decreasing order, c-est-ca/palm-up/voila (‘that’s it/palm-up/that’s it’, 30), lbuoy (‘list buoy’, 30), oui (‘yes’, 22), and si (‘if’, 21). The results related to palm-up are similar to previous studies on LSFB in which this DM is the most frequent and polyfunctional (Gabarró-López, 2018, 2019). However, the results of these two papers related to aussi (annotated as same, see endnote 2) and lbuoy are slightly different compared to the present paper. aussi is the second DM in terms of frequency, and lbuoy stands the third in Gabarró-López (2019), whereas this order is reversed in the present research. In terms of polyfunctionality, lbuoy is slightly less polyfunctional in Gabarró-López (2019) than in the present paper (three functions vs. five domain–function combinations), whereas aussi is more polyfunctional in Gabarró-López (2019) (15 functions vs. 9 domain–function combinations). The different datasets and taxonomies used may motivate these small differences.

One of the limitations of this paper is the reduced dataset analyzed, which may not include all existing DMs in LSFB. For example, no occurrences of et and ajouter (annotated as and and add, respectively, in Crible & Gabarró-López, 2021) were identified. Therefore, future research should include a larger sample with more discourse genres and signers with diverse profiles. This variety would allow us to carry out sociolinguistic studies about the distribution of DMs across age groups, sexes, and linguistic and educational backgrounds, which were beyond the scope of this paper. Other unexplored topics include DM collocations (e.g., in the sample studied, pourquoi ‘why’ and parce-que ‘because’ often co-occur as two continuous DMs) or the position of DMs in discourse (the latter has only been partially investigated in Gabarró-López (2018, 2020) for lbuoy, palm-up, and same).

Finally, most DMs in LSFB were not translated into French by another DM, in line with previous research on DM translation between two spoken languages (e.g., Crible et al., 2019; Cuenca, 2008; Dupont & Zufferey, 2017; Hoek et al., 2017; Zufferey, 2016). DMs were mostly omitted or substituted by a non-discourse-marking conjunction, phrase, etc., in the target texts. As Cuenca (2022, pp. 224–225) nicely summarizes,

“The omission of a marker is related to several factors: the marker, and specifically its meaning and domain, the type of relation (e.g., continuous relations favor deletion), cross-linguistic differences (e.g., lack of equivalent, avoiding stylistic awkwardness), and extralinguistic factors such as the translation process per se (not taking the risk of omitting a word if there is no good reason to do so) or even the genre, e.g., subtitling is known to increase the deletion of linguistic material, especially when lexical content is low”.

Indeed, non-connective functions (e.g., ‘hedging’ and ‘monitoring’) were most frequently omitted/substituted, as well as some continuous relations such as ‘addition’. Other functions, such as ‘agreeing’ and ‘temporal’ in LSFB, were translated more often using other DMs in French because of the DM or the existing equivalents between the languages. The translation of DMs deserves further analysis, including a fine-grained description of strategies, both when the DM is omitted/substituted and when the DM is translated by another DM (either more general, specific, or semantically equivalent). Also, it would be worth studying the cases where no DM was used in the source text but one was added in the target text, to obtain a complete picture of how DMs are translated between a signed and a spoken language.

6. Conclusions

This paper is the first attempt to describe all DMs in a signed language using a parallel corpus of dialogues. As such, it portrays the use of both connective and non-connective DMs in spontaneous conversations, the latter type having been excluded from previous research focusing on monological (frequently prepared) data. The study reports on the type of DMs in LSFB, their functional spectrum, and their translation into written French. Its main contributions are (i) the evidence proving that space is not a DM in LSFB, as has traditionally been said, and that DMs are most frequently a combination of manual and nonmanual elements; (ii) the fine-grained two-level description of DMs; (iii) the finding that DMs in LSFB are mostly omitted or substituted in French translations; and (iv) the creation of an inventory of DMs with their domains, functions, and possible translation/s. These contributions will hopefully be useful not only to linguists and translation scholars but also to other LSFB professionals, such as LSFB, teachers, translation lecturers, and translators, and students at different educational levels and settings.

The first contribution related to space sheds new light on signed language discourse, and the other three point out typological similarities with previous monolingual and contrastive research on DMs in spoken languages, both in oral and written form. This paper pushes translation studies forward and paves the way for the creation of new tools. On the one hand, this research has shown that DMs in LSFB are not only produced drawing on its main channel of articulation (i.e., the hands) but also on nonmanual elements. The fact that signed languages rely on more than one channel to produce DMs suggests that the same may be true for spoken languages; that is, DMs may not only be uttered with the voice, but other bodily actions may contribute to their production. On the other hand, the annotation of more DMs in LSFB and their alignment with their French translation could contribute to machine translation systems between these two languages.

Eventually, it is expected that the interest DMs have generated over the past 40 years in spoken languages will extend to (visual and tactile) signed languages with equal strength, to account for the diversity of modalities in human communication. Research should cover multiple markers, typologically diverse languages, and various research questions to test existing theories and frameworks (exclusively based on oral and written data) and maybe create new ones. The results will uncover how DMs structure discourse within each language, including language-specific and shared uses. This type of research will also facilitate cross-linguistic intra-modal and cross-modal comparisons, ultimately contributing to a more precise overview of the role these elements play when humans (hearing and deaf, sighted and blind) communicate.