(PDF) Discourse annotation guideline for low-resource languages

Natural Language Processing (2025), 31, pp. 700–743

doi:10.1017/nlp.2024.19

SURVEY PAPER

Discourse annotation guideline for low-resource

languages

Francielle Vargas1, Wolfgang Schmeisser-Nieto4, Zohar Rabinovich2,ThiagoA.S.Pardo

1and

Fabrício Benevenuto3

1Institute of Mathematical and Computer Sciences, University of São Paulo, São Carlos, Brazil, 2Viterbi School of

Engineering, University of Southern California, Los Angeles, USA, 3Computer Science Department, Federal University of

Minas Gerais, Belo Horizonte, Brazil, and 4Department of Linguistics, University of Barcelona, Barcelona, Spain

Corresponding author: Francielle Vargas; Email: francielleavargas@usp.br

(Received 30 November 2022; revised 19 February 2024; accepted 27 February 2024)

Special Issue on ‘Natural Language Processing Applications for Low-Resource Languages’

Abstract

Most existing discourse annotation guidelines have focused on the English language. As a result, there

is a signiﬁcant lack of research and resources concerning computational discourse-level language under-

standing and generation for other languages. To ﬁll this relevant gap, we introduce the ﬁrst discourse

annotation guideline using the rhetorical structure theory (RST) for low-resource languages. Speciﬁcally,

this guideline provides accurate examples of discourse coherence relations in three romance languages:

Italian, Portuguese, and Spanish. We further discuss theoretical deﬁnitions of RST and compare differ-

ent artiﬁcial intelligence discourse frameworks, hence offering a reliable and accessible survey to new

researchers and annotators.

Keywords: low-resource languages; rhetorical structure theory; corpus annotation; annotation guideline; discourse

processing

1. Introduction

The word “discourse” is derived from the Latin preﬁx dis- meaning “away” and the root word

currere meaning “to run”, and according to word etymology, the discourse describes thoughts

in words, responsible for the interpretation of the communicative events in context (Nunan

1993). Discourse analysis comprises a dynamic notion that embraces various linguistic com-

ponents of texts (morpho-lexical, syntactic, semantic, and pragmatic), as well as paralinguistic

and extralinguistic components (punctuation, prosody, paragraphing, links to the contextual set-

ting, short- and long-term memory) (DuBois 2003). As claimed by Hovy (1993b), the content of

a speaker’s discourse derives from different sources, and a surface-level structure is required to

comprise them all. The major sources for the content of discourse are the semantics of the mes-

sage, the interpersonal speech acts, the “control” information included by the speaker to assist the

hearer’s understanding (namely, information signaling theme, focus, and topic), and knowledge

about stylistic preferably (Grosz 1987). Considering this scenario, the very assumption underlying

the deﬁnition of discourse may be applied in different areas. Indeed, discourse analysis comprises

Warning: This paper may contain examples of offensive stereotypes and fake news.

RST community and additional resources:http://www.sfu.ca/rst/index.html

C

The Author(s), 2025. Published by Cambridge University Press. This is an Open Access article, distributed under the terms of the

Creative Commons Attribution licence (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted re-use, distribution and

reproduction, provided the original article is properly cited.

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 701

concerns across humanities and social sciences (Fairclough 2003), besides computer science, more

speciﬁcally the artiﬁcial intelligence (AI) dealing with its sub-area called computational linguis-

tics (CL) (Braud et al. 2020,2021,2022; Strube et al. 2023; Fawcett and Davies 1992;Hovy1993a;

Jurafsky and Martin 2009; Androutsopoulos, Lampouras, and Galanis 2013; Trnavac, Das, and

Taboa d a 2016,Rohdeet al. 2018; Fawcett and Davies 1992).

In AI literature dealing with CL, Jurafsky (2020) argues that language does not normally consist

of isolated and unrelated sentences, but instead, it is a collocated, structured, coherent group of

sentences. Moreover, a coherent structured group of sentences is named discourse. Ramsay (2005)

deﬁnes discourse in CL as an extended sequence of sentences produced by one or more people to

convert or exchange information. According to Moore and Wiemer-Hastings (2003), the models

of discourse structure and processing are crucial for constructing computational systems capable

of interpreting and generating natural language. The authors also argue that discourse research in

CL and AI encompasses spoken and written discourse, monologues, and dialogue (both spoken

and keyboarded), and the context created by prior utterances affects the current one regardless of

which participant uttered it. Lastly, research on discourse focuses on two fundamental questions

within CL and AI. First, what information is contained in extended sequences of utterances that

go beyond the meaning of the individual utterances themselves? Second, how does the context in

which an utterance is used affect the meaning of the individual utterances or parts of them (Moore

and Wiemer-Hastings 2003)?

Ramsay (2005) suggests that discourse-aware models must be built incrementally. For exam-

ple, a wide range of discourse sentences accomplishes reference to concepts of the world, which

were already known to the reader in the discourse. The incremental function of natural language

to knowledge construction provides beneﬁts to referring expressions for following tasks. These

tasks are issues into CL related to discourse-level phenomena, such as (i) anaphora,whicheven

though there is notoriously difﬁcult to ﬁnd a deﬁnition, according to Castagnola (2002), con-

sists of a reference to entities mentioned previously in the discourse; pronouns, which are used

to refer to items that have been mentioned very recently, and also it may be recognized based

on very simple characteristic properties; (ii) coreference (also called co-reference), deﬁned by

Jurafsky and Martin (2009) as two noun phrases that refer to the same entity, and coreference

resolution, hence, is the task of identifying the sets of noun phrases that refer to the same entity;

(iii) theme, according to Halliday (1995), consists of a sentence as its ﬁrst phrase, being the rheme

the remainder; (iv) rheme consists of an element responsible for providing the cohesion of a sen-

tence as a communicative whole Bussmann (1998); (v) topic and (vi) focus, such as claimed by

Ramsay (2005), are different devices than theme and rheme, although they achieve similar effects,

mainly if we consider intonation (or its typographical equivalents); (vii) presupposition and (viii)

entailment are types of inference. Presupposition presents a wide range of computational and

linguistic proprieties not exhibited by the general class of inferences (Mukherjee and Joshi 2013).

On the other hand, recognizing textual entailment requires automated systems to identify when

two spans of text share a common meaning; (ix) implicature (also called conversational implica-

ture) is a pragmatic concept introduced by British philosopher (Grice 1975), which showed how

meaning expressed by the speaker (speaker meaning) in conversation, not directly encoded in the

words, maybe inferred (recognized) by the hearer; (x) discourse coherence, as deﬁned by Jurafsky

(2020), consists of a coherent structured group of sentences, in which the word “coherence” is used

to refer to the relationship between sentences that makes real discourses different than just ran-

dom assemblages of sentences. Furthermore, according to Mann et al. (1992), theories of discourse

coherence, which were broadly deﬁned as “the sense of overall unity of a text,” are often used as

descriptive tools in the analysis of coherence.

Discourse and pragmatic literature have been providing a set of frameworks in natural lan-

guage processing (NLP), such as discourse representation structure, which consists of a set of

discourse referents—corresponding to the set of individuals mentioned in the sentence being

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

702 F. Vargas et al.

interpreted—and a set of conditions which are propositions involving those referents. In addi-

tion, the penn discourse treebank (PDTB) (Prasad et al. 2008), is a discourse framework

proposed on the simple idea that discourse relations are grounded in an identiﬁable set of explicit

words or phrases (discourse connectives) or simply in sentence adjacency, has been taken up and

used by many researchers in the NLP community and more recently, by researchers in psycholin-

guistics as well. As claimed by Jurafsky (2020), PDTB consists of labeling corpus lexical-aware

grounded composed of annotation of discourse coherence between text spans; they were given a

list of discourse connectives, words that signal discourse relations, such as like,because,although,

when,since,oras a result, among others. Finally, the most common discourse organization frame-

work used in CL (Jurafsky 2020), the rhetorical structure theory (RST)(MannandThompson

1987), is deﬁned as a theory to help us understand texts as instruments of communication.

The RST consists of a discourse framework for investigating relational propositions, which are

unstated but inferred propositions that arise from the text structure in the process of interpret-

ing texts (Mann and Thompson 1987). Although there is a lack of comprehensive RST literature

reviews, Vargas et al. (2022) proposed a survey on RST applied to fake news and fake reviews

detection, and Hou et al. (2020) provided an RST comprehensive review, in which theory, parsing

methods, and applications are well-discussed and presented.

Lastly, as claimed by Passonneau and Litman (1997), natural language systems rarely exploit

discourse structure, and cultural devices due to scarce consideration in the development of data

resources and methods concerning diferrent languages. This predominance results in a lack of

research and advances for low-resource languages (LRLs), as well as a lack of linguistic diver-

sity. To ﬁll this relevant gap, we introduce the ﬁrst discourse annotation guideline using the RST

framework for LRLs. We also present a summarized survey that comprises the main deﬁnitions

concerning discourse in AI dealing with CL and discuss theoretical aspects related to the RST

framework. Speciﬁcally, our guideline describes 24 (twenty-four) discourse coherence relations

in three romance languages: Italian, Portuguese, and Spanish.

2. Low-resource languages

According to Khan et al.(2023), natural languages may be classiﬁed into two main categories:

LRLs and high-resource languages (HRLs). More broadly, HRLs consist of languages with a

wide range of available data resources that enable machines to learn and understand natural

languages. For instance, English is a well-resourced language as compared to other spoken lan-

guages (Khan et al.2023). On the other hand, LRLs consist of languages with scarce or no

resources available, which is deﬁned by Cieri et al.(2016) as less studied, resource-scarce, less

computerized, less privileged, less commonly taught, or low-density languages. For instance,

most of the European languages are severely under-resourced (Khan et al. 2023). Furthermore,

Latin American languages including indigenous languages are also considered under-resourced

languages.

According to Cieri et al. (2016), terms such low density,less commonly taught,under-resourced,

less resourced,andlow resource are commonly used to refer to LRLs. For instance, low density

consists of languages with few online or computational data resources exist (Megerdoomian and

Parvaz 2008). Differently, less commonly taught are world languages that are underrepresented in

the education system.aLastly, the terms under-resourced,less resourced,orlow resource seem to

have similar means by literature.

In addition, LRL is also deﬁned via the metric of Gross Language Product (Wiemerslage et al.

2022). It is the product of the number of native speakers of the language in any country and the

country’s per capita Gross National Product. Notice that this deﬁnition suffers inﬂuence from

aNational Council of Less Commonly Taught Languages (NCOLCTL).

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 703

the social-political and economic context of language once that “low” deﬁnition is most likely to

be relative to some expectation based on the “importance” of the language (Megerdoomian and

Parvaz 2008).

Finally, It is a well-known fact that the LRLs are neglected in NLP literature. As a result, most

discourse-aware data resources are produced in English. Hence, we introduce the ﬁrst discourse

guideline using RST for LRLs. Speciﬁcally, this guideline provides examples of RST coherence

relations in three different romance languages: Italian,Portuguese,andSpanish.

3. RST annotation guideline

Here, we introduce the ﬁrst discourse annotation guideline using RST focused on LRLs. This

guideline focuses mainly on Italian, Portuguese, and Spanish. We used three different datasets:

(i) DETESTSb(Ariza et al. 2022) composed of comments in Spanish from news articles seg-

mented into 5,629 sentences containing hate speech against immigrants; (ii) Deceiverc(Vargas

et al. 2021) composed of 600 news articles and fake news in Portuguese and English; and (iii)

Italian-RST-Newsdextracted from relevant media outlets in Italy, as well as the Wikipedia.it.

Note that we extracted examples from datasets in different languages and domains (e.g., news

articles, fake news, and hate speech comments containing stereotypes) in order to show the

feasibility of the RST framework to be applied in different languages, domains and discourse

structures. Nevertheless, each language and domain present its challenges, mainly the domain of

web comments due to different noise and grammar and/or vocabulary errors or inadequacies, and

idiosyncrasies. Hence, we aim to provide a well-deﬁned and structured discourse guideline to sup-

port researchers and annotators to annotate discourse structure using RST in different languages

and domains.

The RST was originally proposed for computer-based authoring and natural language genera-

tion; however, this powerful framework has been used ever since in a wide variety of NLP tasks.

RST is a descriptive theory of a major aspect of the organization of natural text. It is a linguisti-

cally useful method for describing natural texts, characterizing their structure primarily in terms

of relations that hold between parts of the text (Mann and Thompson, 1987). Also, it provides

a model that represents a document as a tree of discourse units recursively built by connect-

ing smaller units through rhetorical relations (Mabona et al. 2019). Furthermore, according to

Mann and Thompson (1987), the fundamental mechanisms to annotate text and generate RST

trees are segmentation,nuclearity,schemas,andstructures (coherence relations), described as

follows.

3.1 Segmentation

Text segmentation is a fundamental task in NLP. In RST, there are two main deﬁnitions for text

segmentation: text spans and elementary discourse unit (EDU) (Asher and Lascarides 2023).

A text span is an uninterrupted linear interval of text. On the other hand, an EDU is usually

deﬁned as minimal building blocks for forming a discourse tree, and it is mostly used to desig-

nate a minimal speech act or communicative unit, where each EDU corresponds roughly to a

clause-level content that denotes a single fact or event (Prevot et al. 2023). Furthermore, the dis-

course segmentation task consists of segmenting an uninterrupted linear interval of text into a

sequence of EDUs (Li, Sun, and Joty 2018). Each utterance of an EDU contributes to the commu-

nicative relevance of the preceding utterances or constitutes the onset of a new unit of meaning or

event that subsequent utterances may add to (Passonneau and Litman 1997). In the same settings,

bhttps://github.com/alarca94/detests

chttps://github.com/franciellevargas/Deceiver

dhttps://github.com/franciellevargas/Italian-RST-News

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

704 F. Vargas et al.

discourse models as multiutterance EDUs and structural relations among them yield a discourse

tree structure (Grosz and Sidner 1986).

According to Hobbs (1979), segmental discourse structure is an artifact of coherence relations

among utterances, such as elaboration, evaluation, cause, and so on. The EDU in RST is an atomic

semantic unit, similar to a clause in a sentence, and may also be considered clause-like units that

serve as building blocks for discourse parsing (Li et al.2018). Although the types of discourse units

being coded and the relations among them vary (Passonneau and Litman 1997), to identify EDUs

is necessary to classify the prominent and complementary text spans into a sentence or document

according to the writer’s intention. For instance, we ﬁrst should identify the most prominent and

important text segment in the sentence, then, the most complementary text segment. As a result,

the most prominent text segment is classiﬁed as the nucleus, and the complementary text segment

is classiﬁed as the satellite.

3.2 Nuclearity

Nuclearity in RST consists of the identiﬁcation of prominent and complementary EDUs classiﬁed

as nucleus and satellite. Furthermore, the type of nuclearity is divided into mononuclear and

multinuclear, as shown in Figure 1. Observe that the nucleus consists of EDUs that are more

central and relevant in the text. Differently, the supporting EDUs are named satellites.

Figure 1. Nuclearity structure in mononuclear and multinuclear discourse coherence relations.

3.3 Schema

According to Mann and Thompson (1987), a schema may be deﬁned as predeﬁned patterns spec-

ifying how regions of text combine to form larger regions, up to whole texts.InFigure2,we

show ﬁve different types of schemas originally proposed by RST’s authors. Observe that a schema

based on the RST framework is characterized by a vertical line pointing to one of the text spans

that the schema covers called the nucleus. The other spans are linked to the nucleus by relations,

represented by labeled curved lines, and these spans are called satellites.

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 705

Figure 2. Types of schemes.

3.4 RST coherence relations

The RST predicts a tree of coherence relations (also known as rhetorical or discourse relations),

mainly based on the premise that the content of text units may be hierarchically organized.

Accordingly, it predicts that some units are more central (salient) to the text than others, as well

as that the other units support the text message. In this RST annotation guideline for LRLs, we

describe in detail 24 (twenty-four) coherence relations proposed originally by RST manuscript

(Thompson and Mann 1988). We highlight each of them in Sections 3.4.1–3.4.24.

3.4.1 ANTITHESIS

According to linguistics studies, antithesis is deﬁned as a ﬁgure of speech characterized by the

simultaneous use of terms, words, or sentences that are opposed to the meaning. For example, in

“truth and lies are part of everyday life,” the terms truth and lies form an antithesis construc-

tion due to opposition of meaning between the terms used to create an effect on the reader.

In general, most ﬁgures of speech are used to create particular potential effects on the reader.

According to Stede et al. (2017), deﬁne the antithesis discourse coherence relation as a relation-

ship in which the writer regards the content of the nucleus as more important. In addition, this

relation is rarely signaled by connectives, and the situation presented in the nucleus is found in

contrast with the situation presented in the satellite. The contrast may happen in only one or

a wide range of respects. Also, it is always mononuclear—it is a counteractive relation that dis-

tinguishes clearly between the nuclearity of its arguments (Carlson and Marcu 2001). Table 1

describes the deﬁnitions in terms of constraints on the nucleus, satellite, and the combination of

both for the antithesis relation. Notice this relation presents information intentionally informed

Tab le 1. Antithesis relation constraints

Type Description

Type of nucleus Mononuclear

......................................................................................................................................................................................................................................................

Type of relation Intentional/pragmatic

......................................................................................................................................................................................................................................................

Constraints on the nucleus (N) The writer judge N valid

......................................................................................................................................................................................................................................................

Constraints on the satellite (S) N/A

......................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) N and S are in contrast

......................................................................................................................................................................................................................................................

Effect on the reader The reader’s ability to comprehend N is increased

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

706 F. Vargas et al.

Figure 3. Antithesis relation in Spanish.

Figure 4. Antithesis relation in Portuguese.

Figure 5. Antithesis relation in Italian.

and it is always mononuclear. In addition, there is one constraint on the nucleus, but there are no

constraints on the satellite. In the nucleus, the writer judges the nucleus as valid. Lastly, the effect

on the reader consists of the reader accepting the nucleus better than the satellite. Figures 3,4,

and 5show examples of the antithesis relation.

Observe that the nucleus presents information of which the author is more favorable than

information presented in satellite. In addition, the nucleus and satellite are both in contrast,

and the reader accepts the nucleus better than the satellite. Finally, a ﬁnal statement is not less

important than previous statements.

3.4.2 BACKGROUND

Stede et al. (2017) deﬁne the BACKGROUND relation as a type of coherence relation that settles

a relationship between the nucleus and the satellite, in which the understanding of the satellite

makes it easier for the reader to understand the content of the nucleus. Therefore, it would be

difﬁcult to comprehend the nucleus without the “background” information provided by the satel-

lite. Furthermore, the satellite, mostly but not always, precedes the nucleus, and the satellite at

the beginning of the text often serves to introduce the topic of the text. Furthermore, Carlson

and Marcu (2001) claim that the satellite is responsible for providing the context of use or the

grounds concerning which the nucleus is to be understood. In this case, understanding the satel-

lite helps the reader understand the nucleus. Also, the satellite is not the cause/reason/motivation

of the situation presented in the nucleus, and the reader/writer’s intentions are irrelevant in deter-

mining whether such a relation holds. Finally, the background is rarely signaled by connectives

(Stede et al.2017). In Table 2, we describe the deﬁnitions for the background relation. Notice it is

a mononuclear relation and presents intentional information. The nucleus presents information

that must be understood from the information described in the satellite. There are no constraints

on the satellite. Therefore, the satellite improves the reader’s ability to understand the nucleus. On

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 707

Tab le 2. Background relation constraints

Type Description

Type of nucleus Mononuclear

......................................................................................................................................................................................................................................................

Type of relation Intentional/pragmatic

......................................................................................................................................................................................................................................................

Constraints on the nucleus (N) The reader would not understand sufﬁciently N before to read S

......................................................................................................................................................................................................................................................

Constraints on the satellite (S) N/A

......................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) S increases the reader’s ability to understand any aspect of N

......................................................................................................................................................................................................................................................

Effect on the reader The readers’ ability to understand N is increased

Figure 6. Background relation in Spanish.

Figure 7. Background relation in Portuguese.

Figure 8. Background relation in Italian.

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

708 F. Vargas et al.

the effects on the reader, the reader understands better the nucleus from the reading of satellites.

Figures 6,7,and8show examples of background relation. Observe that the nucleus presents the

information more prominent. Nevertheless, the information presented in satellites is relevant to

understanding the nucleus. In other words, without the information presented in the satellite, the

reader would not understand properly the information presented in the nucleus.

3.4.3 CIRCUMSTANCE

According to linguistic studies, the circumstantial meanings would be analyzed from different

perspectives. In Martin (1992), the authors studied the circumstantial meanings from a discourse

semantic perspective that would be related to the proposed term “setting,” which refers to mainly

locational circumstantial meanings. Nevertheless, as claimed by Dreyfus and Bennett (2017), it is

separated from the type of circumstantial meaning from the type of lexico-grammatical structure

that realizes that meaning. For instance, the sentence “I went to the union that hot Friday.”

In this sentence, it is possible to note a type of circumstantial meaning related to locational

circumstantial meanings. On the other hand, the sentence “Lunchtimes on Friday are always busy

in this cafe” provides a type of lexico-grammatical structure related to qualiﬁer circumstantial

meanings. More broadly, the circumstantial meaning is a region of ideational meaning that

is instantiated across a range of lexico-grammatical structures: from the rank of the clausal

constituent of circumstance in both directions up to clause rank and down to below or within

constituent rank (e.g., as qualiﬁer) (Dreyfus and Bennett 2017). Shortly, this last one proposes the

circumstantial meaning within the discourse semantic system of ideation (Martin 1992).

According to Mann and Thompson (1987), the circumstance relation, holds between two parts

of a text if one of the parts establishes a circumstance or situation, and the other part is interpreted

within or relative to a framework (e.g., a temporal or spatial framework). Moreover, Stede et al.

(2017) deﬁne it as a type of relationship between the nucleus and satellites, in which the satellite

supplies necessarily an event or state that occurs or has occurred; therefore, it is not a hypotheti-

cal one. Carlson and Marcu (2001) argue that the presented situation in the satellite provides the

context of which the presented situation in the nucleus should be interpreted. In addition, it is

always mononuclear, and selecting the circumstance relation over the background relation when

the events are described in the nucleus and satellite is somewhat co-temporal. Considering a com-

parative analysis between the circumstance and background, the information or the context of the

background is not always speciﬁed clearly or delimited sharply. Also, the events represented in

the nucleus and the satellite occur at distinctly different times. The events in a circumstance rela-

tion are somewhat co-temporal (Carlson and Marcu 2001). Lastly, the typical connectives found

in the circumstance relation are “as”, “when”, and “while”(Stede et al. 2017). Table 3describes the

deﬁnitions for circumstance relation. Notice it is a mononuclear relation, which presents semantic

Tab le 3. Circumstance relation constraints

Type Description

Type of nucleus Mononuclear

.........................................................................................................................................................................................................................................................................

Type of relation Intentional/pragmatic

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus (N) N/A

.........................................................................................................................................................................................................................................................................

Constraints on the satellite (S) S presents a realized situation (i.e., not a hypothetical one)

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) S provides any situation that the reader may interpret N

.........................................................................................................................................................................................................................................................................

Effect on the reader The reader recognizes that S provides a situation in which N must be

interpreted

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 709

Figure 9. Circumstance relation in Spanish.

Figure 10. Circumstance relation in Portuguese.

Figure 11. Circumstance relation in Italian.

information without constraints on the nucleus. Furthermore, the satellite must present a situa-

tion realized, and the satellite necessarily provides a situation in which the reader may interpret the

nucleus. Finally, Carlson and Marcu (2001) argue that in a circumstance relation, the satellite may

not present any cause, reason, or motivation for the situation presented by the nucleus, and the

reader/writer’s intentions are irrelevant to determine whether such a relation holds. Accordingly,

as shown in Figures 9,10,and11, the presented situation in the satellite provided the context in

which the presented situation in the nucleus must be interpreted.

3.4.4 CONCESSION

Historically, the notion of concession according to linguistic studies is associated with relations

between the utterance segments that imply a contrast. Furthermore, its deﬁnition takes into

account pragmatics and cognitive inferences. According to Kim (2001), the speaker asserts the

propositions of the two related clauses in question against the background assumption that the

two types of situations are generally incompatible. (e.g. “Even if Einstein tried to solve the math

problem, he could not solve it.”). In RST, Stede et al.(2017)deﬁnetheconcession relation, as a

type of relationship between the nucleus and satellite, in which the writer regards the content of

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

710 F. Vargas et al.

the nucleus as more important than the content presented by the satellite. Besides that, in compar-

ison to the nucleus, the writer regards the content of the satellite as less important, even though

the writer does not dispute that the satellite holds. In Carlson and Marcu (2001), the concession

is deﬁned as a type of relationship in which the situation indicated in the nucleus is contrary to

expectation in the light of the information presented in the satellite. In other words, it is always

characterized by a violated expectation. Furthermore, the typical connectives are “although”,

“but”, “and”, “despite”,“which”.Table4describes the deﬁnitions for the concession relation.

Notice it is mononuclear and presents intentional information, with constraints on the nucleus

and satellite. In addition, the writer judges the nucleus and the satellite despite them being incom-

patible. Figures 12,13,and14 shown examples of concession. Observe that the connective “but” (in

Spanish: “perfo,” in Portuguese: “mas,” and in Italian “ma”) is an example of a discourse marker,

or in other words, it is also a signal of concession relation. Accordingly, it would be agreed that

these discourse markers may be considered as “clues” to identify RST coherence relations. Finally,

as shown in Figures 12,13,and14, we also note an incompatibility between the satellite and the

nucleus raises the reader’s ability to regard the nucleus, and such property distinguishes these

discourse coherence relations from others.

Tab le 4. Concession relation constraints

Type Description

Type of nucleus Mononuclear

.........................................................................................................................................................................................................................................................................

Type of relation Intentional/pragmatic

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus (N) Writer regards the content of N as more important

.........................................................................................................................................................................................................................................................................

Constraints on the satellite (S) In comparison to N, the writer regards the content of S as less

important, but s/he does not dispute that S holds

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) The writer acknowledges a potential or apparent incompatibility

between N and S

.........................................................................................................................................................................................................................................................................

Effect on the reader The reader’s ability to positive regard for N is increased

Figure 12. Concession relation in Spanish.

Figure 13. Concession relation in Portuguese.

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 711

Figure 14. Concession relation in Italian.

3.4.5 CONDITION

Conditional sentences present two main parts: (i) the antecedent titled protasis and (ii)

the consequent denomination titled apodosis. For instance, in the following example: “If

you come closer, you’ll be able to see the parade”, the antecedent would be “if you

come closer”, which consists of protasis, and the consequent “you’ll be able to see the

parade”, would be the apodosis. Furthermore, Sweetser (1990) claims that conditional seman-

tics has three distinct domains: (i) the content domain, which is understood by relating

the content of the two clauses to each other, for example, if you drop it, it will break;

(b) the epistemic domain, which is understood as expressions of the reasoning process (e.g. “if

the streets are wet, it rained last night”); and (c) the speech act domain, which is understood

as pre-posting to a speech act (e.g. “if you leave before I see you again, have a good time”). In

RST, the condition relation may be deﬁned as a relationship between the nucleus and satellite, in

which the information associated with the nucleus must be a consequence of the achievement

of the condition in the satellite (Mann and Thompson, 1987). In addition, Carlson and Marcu

(2001) suggest that the truth of the proposition associated with the nucleus is a consequence of

the fulﬁllment of the condition in the satellite. Likewise, the satellite provides a situation that is

not realized. Also, Stede et al.(2017) claim that the satellite must present a hypothetical, future, or

in other ways, unreal situation, and the realization of the nucleus depends necessarily on the real-

ization of the satellite. Typically, this relation is signaled by connectives, such as “if ...then”, and

“in case”. Table 5describes the deﬁnitions for the condition relation. Notice it presents only one

Tab le 5. Condition relation constraints

Type Description

Type of nucleus Mononuclear

.........................................................................................................................................................................................................................................................................

Type of relation Informational/semantic

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus (N) N/A

.........................................................................................................................................................................................................................................................................

Constraints on the satellite (S) The S presents a hypothetical, future, or otherwise unrealized

situation (relative to the situational context of S)

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) The realization of N depends on the realization of S

.........................................................................................................................................................................................................................................................................

Effect on the reader The R recognizes how the realization of N depends on the

realization of S

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

712 F. Vargas et al.

Figure 15. Condition relation in Spanish.

Figure 16. Condition relation in Portuguese.

Figure 17. Condition relation in Italian.

nucleus and semantic information. There are also no constraints on the nucleus, and the satellite

presents a hypothetical situation. Figures 15,16,and17 show examples of this relation. Observe

that, in the condition relation, the nucleus presents a fact that may be performed whether the con-

dition presented in the satellite will be performed. Even though all examples presented comprise

the marker “if” (in Spanish: “si,” in Portuguese: “se,” and in Italian: “si”), according to Mann and

Thompson (1987), the condition relation does not necessarily need to be expressed with an “if”

clause.

3.4.6 CONTRAST

Natural languages comprise sophisticated mechanisms capable of connecting information. This

“connectors” are responsible for connecting sentences and establishing a coherent relation, and

they are called conjunctions. In general, when any speaker wishes to connect contrastive informa-

tion, a particular type of conjunction is employed. They are classiﬁed as adversative connectives

or contrastive connectives, besides contrastive discourse markers, which are connectors of dis-

course units whose content refers to an “opposite” or “contradictory” (Fraser 1999). In RST, parts

of texts whose relationship manifests semantically as an opposition must be classiﬁed as a contrast

relation. The contrast relation provides a type of relationship between the nucleus and nucleus

with exactly two nuclei being both of equal importance for the writer’s purposes (Mann and

Thompson, 1987). Moreover, the contents are comparable but not identical, and they differ tak-

ing into account aspects that are important to the writer. According to Carlson and Marcu (2001),

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 713

the typical CONTRAST relation includes contrastive discourse markers such as “but”, “however”,

“while”. In addition, Stede et al.(2017) claim that discourse markers for the identiﬁcation of this

relation are “on the other hand”, “yet”, and “but”. In a summarized way, the contrast is a mult-

inuclear relation, which establishes the relationship between the nucleus and nucleus to express

an opposite thought. It contrasts with the previous thought, and typically at least a nucleus is

introduced by adversative conjunctions. Table 6describes the deﬁnitions for the contrast rela-

tion. Notice it presents two or more nuclei and semantic information. Furthermore, the contrast

between text spans raises the ability’s reader to regard the contraction information present in the

nucleus. Figures 18,19,and20 show examples of this relation. Observe that the facts presented in

Tab le 6. Contrast relation constraints

Type Description

Type of nucleus Multinuclear

.........................................................................................................................................................................................................................................................................

Type of relation Informational/semantic

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus (N) Two nuclei are presented. Both are of equal importance for W’s

purposes. The contents are comparable yet not identical. They

differ in aspects that are important to write

.........................................................................................................................................................................................................................................................................

Constraints on the satellite (S) N/A

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) The characteristics of the satellite and nucleus are in

comparison

.........................................................................................................................................................................................................................................................................

Effect on the reader R recognizes the comparability and the difference(s) yielded by

the comparison is being made

Figure 18. Contrast relation in Spanish.

Figure 19. Contrast relation in Portuguese.

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

714 F. Vargas et al.

Figure 20. Contrast relation in Italian.

the nucleus are in contrast, and there are discursive markers to identify this contrast. Nevertheless,

contrastive sentences may occur without any discourse marker.

3.4.7 ELABORATION

The elaboration is a relation highly ambiguous. According to a study proposed by Marcu and

Echihabi (2002), the contrast and elaboration relations presented only 61 from 238 discourse

markers (26 percent). Moreover, Carlson and Marcu (2001) claim that this relation is extremely

common at all levels of the discourse structure, and it is especially popular to show relations across

large spans of information. In RST, in order to accurately identify the elaboration relation, the

satellite must give additional information or detail about the situation presented in the nucleus

(Carlson and Marcu 2001). Furthermore, Stede et al.(2017) deﬁne this relation as a relationship

between the nucleus and satellites, in which the satellite provides details or more information

on the state of affairs described in the nucleus. Lastly, Marcu (2000) suggests that the elabora-

tion relation may be used when none of the other relations is applied. According to Stede et al.

(2017), typical connectives are “in particular” and “for example”. Table 7describes the deﬁnitions

for the elaboration relation. Note that this relation presents only a nucleus and semantic infor-

mation beyond no constraints on the nucleus and satellite, and the satellite typically presents any

additional information about the nucleus. Also, the reader’s ability to recognize additional infor-

mation in a satellite that refers to the nucleus. Figures 21,22,and23 show examples of this relation.

Observe that the satellite typically provides a qualiﬁcation or speciﬁcation of the nucleus, besides

the adjective subordinate clauses, which are candidates to be an elaboration relation.

Tab le 7. Elaboration relation constraints

Type Description

Type of nucleus Mononuclear

.........................................................................................................................................................................................................................................................................

Type of relation Informational/semantic

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus (N) N/A

.........................................................................................................................................................................................................................................................................

Constraints on the satellite (S) The S provides details or more information on the state of affairs

described in N (but not on a single entity mentioned in N

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) The S presents additional detail about the situation or some

element of subject matter which is presented in N or inferential

accessibleinNinoneormoreofthewayslistedbelow

.........................................................................................................................................................................................................................................................................

Effect on the reader The R recognizes S as providing additional detail for N. Moreover, R

identiﬁes the element of subject matter for which detail is provided

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 715

Figure 21. Elaboration relation in Spanish.

Figure 22. Elaboration relation in Portuguese.

Figure 23. Elaboration relation in Italian.

3.4.8 ENABLEMENT

According to linguistics studies, enablement sentences present possibilities or hypotheses. For

instance, the particles “can” and “may” in general connect assumptions semantically organized to

express a possibility and pragmatically intended to raise the reader’s belief that the fact is unreal-

izable. According to Mann and Thompson (1987), the enablement relation provides information

designed to increase the reader’s desire to act. Furthermore, Stede et al. (2017) suggest that this

relation is a genre of editorial, and also it is rarely found in other textual types, hence, it is difﬁ-

cult to identify. In RST, the enablement relation is deﬁned as a nucleus responsible for presenting

information on a situation that is unrealized. In addition, the action presented in the satellite

increases the chances of the situation in the nucleus being realized (Carlson and Marcu 2001).

Table 8describes the main deﬁnitions for the enablement relation in terms of constraints on

the nucleus and satellite. Notice it presents only a nucleus; hence it is a mononuclear relation.

Additionally, there are no constraints on the satellite, and the nucleus presents an action or fact

that may be realized whether the possibility presented in the satellite is true. Finally, the reader’s

ability to act as the nucleus is increased. Figures 24,25,and26 show examples of the enablement

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

716 F. Vargas et al.

relation. Observe that, as stated by Stede et al. (2017), the nuclei present an unrealized situation,

and comprehending the satellite makes it easier for the reader to perform the action described in

the nucleus.

Tab le 8. Enablement relation constraints

Type Description

Type of nucleus Mononuclear

.........................................................................................................................................................................................................................................................................

Type of relation Intentional/pragmatic

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus (N) The N presents an action by R (including accepting an offer),

unrealized concerning the context of N

.........................................................................................................................................................................................................................................................................

Constraints on the satellite (S) N/A

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) R comprehending S increases R’s potential ability to act N

.........................................................................................................................................................................................................................................................................

Effect on the reader R’s potential ability to act N increases

Figure 24. Enablement relation in Spanish.

Figure 25. Enablement relation in Portuguese.

Figure 26. Enablement relation in Italian.

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 717

3.4.9 EVALUATION

Subjective information or subjectivity in natural language refers to aspects of language used to

express opinions, evaluations, and speculations (Wiebe et al.2004). Worldviews and points of

view are built to evaluate situations, events, and objects in the world, and this information is

analyzed at the discourse level. In RST, the evalulation relation provides subjective information.

In other words, it must comprise a point of view, appraisal, estimation, rating, interpretation, or

assessment of a situation. Therefore, each piece of subjective information is signaled by evalulation

relation.

According to Carlson and Marcu (2001), the evalulation relation is deﬁned as a relation-

ship between the nucleus and satellites, in which one span evaluates the situation presented

in the other span of the relationship on a scale of good to bad. Besides, according to Stede

et al.(2017), usually the “evaluating” segment follows the “evaluated” one, although sometimes

the order is the opposite. Therefore, the evalulation relation occurs when the satellite typi-

cally presents any subjective information. Table 9describes the deﬁnitions for the evalulation

relation. Notice it is mononuclear and presents semantic information. While there are no con-

straints on the nucleus and satellite, the satellite provides subjective content. Furthermore, the

reader recognizes the subjective value of information in the satellite. Figures 27,28,and29

show examples of this relation. Observe that the satellites provide subjective information. Hence,

toward identifying the EVALUATION discourse coherence relation, the satellite must necessarily

take account of the point of view on the topic presented in the nucleus at a scale between bad

and good.

Tab le 9. Evaluation relation constraints

Type Description

Type of nucleus Mononuclear

.........................................................................................................................................................................................................................................................................

Type of relation Informational/semantic

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus (N) Description of a state of affairs or a subjective statement (but not

from a writer’s perspective)

.........................................................................................................................................................................................................................................................................

Constraints on the satellite (S) A subjective evaluation (positive/negative, desirable/undesirable)

from the writer’s perspective

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) S relates N to the degree of the writer’s positive or negative regard

toward N

.........................................................................................................................................................................................................................................................................

Effect on the reader Reader recognizes that S assesses N and recognizes the value it

assigns

Figure 27. Evaluation relation in Spanish.

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

718 F. Vargas et al.

Figure 28. Evaluation relation in Portuguese.

Figure 29. Evaluation relation in Italian.

3.4.10 EVIDENCE

Stede et al.(2017)deﬁnetheevidence relation as a type of relationship between the nucleus and

satellites, in which the nucleus presents a subjective statement or thesis or claim, whereby the

reader might not accept or might not regard as sufﬁciently important or positive. Furthermore,

the satellite must provide a statement that the reader is likely to accept. Carlson and Marcu (2001)

argue that it is a type of discourse coherence relation in which the satellite must present a sit-

uation that is responsible for providing any evidence or justiﬁcation concerning the situation

described in the nucleus. Additionally, the evidence relation pertains to actions and situations

that are independent of the will of an animate agent. Also, the evidence is data on which judg-

ment of a conclusion may be based and is presented by the writer or an agent in the article

to convince the reader of a point. Lastly, the typical connectives are the causal connectives

(Stede et al.2017). Table 10 describes the deﬁnitions for evidence relation. Notice it is mononu-

clear and presents semantic information. Although there are no constraints on the nucleus and

satellite, the relationship between satellites is performed through the writer’s subjective content.

Table 10. Evidence relation constraints

Type Description

Type of nucleus Mononuclear

.........................................................................................................................................................................................................................................................................

Type of relation Intentional/pragmatic

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus (N) Reader might not believe N to a degree satisfactory to the writer

.........................................................................................................................................................................................................................................................................

Constraints on the satellite (S) Reader believes S or will ﬁnd it credible

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) Reader’s comprehension of S increases the reader’s belief of N

.........................................................................................................................................................................................................................................................................

Effect on the reader Reader’s belief in the N is increased

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 719

Furthermore, the reader recognizes the subjective value of information in the satellite, and also

recognizes the intention of the writer to provide credibility to the information presented in the

nucleus. Finally, according to Stede et al.(2017), this relation often provides connections between

a larger satellite segment and a shorter nucleus. For instance, in evidence whose source is a

thesis, as shown in Figures 30,31,and32. Observe that the satellite provides explicit informa-

tion that serves as contestable or incontestable evidence to support information claimed in the

nucleus. Furthermore, the EVIDENCE discourse coherence relation may also present statistical

information.

Figure 30. Evidence relation in Spanish.

Figure 31. Evidence relation in Portuguese.

Figure 32. Evidence relation in Italian.

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

720 F. Vargas et al.

3.4.11 INTERPRETATION

The interpretation relation may be complex to identify. According to Mann and Thompson (1987),

this relation is mononuclear and may be deﬁned as a type of relationship between the nucleus and

satellite, in which the satellite relates the situation presented in the nucleus to a framework of

ideas not involved in the nucleus itself and not concerned with the writer’s positive or negative

regard. Moreover, it presents the satellite evaluating the knowledge presented by the nucleus in

terms of the writer’s positive or negative regard. In the same settings, Carlson and Marcu (2001)

claim that to identify this relation one side of the relationship gives a different perspective on the

situation presented in the other side. It is subjective, presenting the personal opinion of the writer

or a third party. Additionally, Stede et al.(2017) suggests that the nucleus precedes the satellite in

the text, and the satellite shifts the content of the nucleus to a different conceptual frame. Besides,

It should be pointed out that the interpretation relation does not imply that threre is an evaluation

of the nucleus (or the evaluation is of only secondary importance). Accordingly, toward identify-

ing the interpretation relation, the reader must keep in mind that whenever the satellite primarily

provides an assessment (on the positive/negative scale) of the nucleus, the evaluation relation is

to be used, rather than interpretation. Lastly, the typical connective is “thus”. Table 11 describes

the deﬁnitions for the interpretation relation. Notice the content of the nucleus is presented to a

different conceptual frame. This does not imply an evaluation of the nucleus (or the evaluation is

of only secondary importance). Furthermore, on the report of Stede et al. (2017), the reader rec-

ognizes that the satellite relates the nucleus to a framework of ideas not involved in the knowledge

presented in the nucleus itself. Figures 33,34,and35 show examples of the interpretation rela-

tion. Observe the expression of a viewpoint of the writer or another agent in the text is observed

in this relation (Carlson and Marcu 2001). We should like to reiterate that the interpretation rela-

tion presents high ambiguity with the evaluation relation. Therefore, it is important to salient the

importance of a specialist to accurately identify this kind of discourse coherence relations.

Table 11. Interpretation relation constraints

Type Description

Type of nucleus Mononuclear

......................................................................................................................................................................................................................................................

Type of relation Informational/semantic

......................................................................................................................................................................................................................................................

Constraints on the nucleus (N) N/A

......................................................................................................................................................................................................................................................

Constraints on the satellite (S) N/A

......................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) Reader’s comprehension of S increases the reader’s belief of N

......................................................................................................................................................................................................................................................

Effect on the reader Reader’s belief in the N is increased

Figure 33. Interpretation relation in Spanish.

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 721

Figure 34. Interpretation relation in Portuguese.

Figure 35. Interpretation relation in Italian.

3.4.12 JOINT

The joint multinuclear relation holds between two segments that reﬂect different topics (Marcu

2000). This relation establishes a type of relationship between the nucleus and satellite, in which

the nucleus provides different information that are not of the same type. Hence, they are not in an

identiﬁable semantic or pragmatic relation, nor do they form an enumeration (Stede et al.2017).

Furthermore, it is also used when a multinuclear relation is needed (from the text to global per-

spective). However, it is no speciﬁc relations are applicable. Lastly, the nucleus adds information

that is not comparable and is not in temporal sequence. For instance, observe the following sen-

tence: “I bought a boot, a sneaker, and a slipper”. The terms “boot,” “sneaker,” and “slipper” are

semantically compatible, but it is not an example of the JOINT relation. Differently, in the fol-

lowing sentence: “The talented young plays the piano and constantly enjoys traveling to Asia”,

the information is distinct and not comparable; hence this is considered an example of the joint

relation. The typical connectives for the joint relation are additive connectives (Stede et al. 2017).

Table 12 describes the deﬁnitions for the joint relation. Notice it provides semantic information

and presents at least two nuclei. Although there are not any constraints on the nuclei, the reader

recognizes that the nucleus holds between two segments that denote different things. Figures 36,

37,and38 show examples of the JOINT discourse coherence relation. Observe that the nuclei

present a segment of information that while it is related denotes different aspects. Furthermore,

the additive connective “and” is a discourse marker to identify the joint relation. Nevertheless, it is

necessary to evaluate ﬁrst whether it would be classiﬁed as SEQUENCE discourse coherence rela-

tion (see Section 3.4.20). Other examples of additive connectives are “in addition”, “additionally”,

and “also”.

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

722 F. Vargas et al.

Table 12. Joint relation constraints

Type Description

Type of nucleus Multinuclear

.........................................................................................................................................................................................................................................................................

Type of relation Informational/semantic

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus (N) The nuclei provide different kinds of information, which are not of

the same type. They are not in an identiﬁable semantic or

pragmatic relation, nor do they form an enumeration. Still, there is

a coherent link, as they contribute in similar ways to the overall

text function

.........................................................................................................................................................................................................................................................................

Constraints on the satellite (S) N/A

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) N/A

.........................................................................................................................................................................................................................................................................

Effect on the reader Reader recognizes that each nucleus contributes its message,

which however serves the same overall text purpose

Figure 36. Joint relation in Spanish.

Figure 37. Joint relation in Portuguese.

Figure 38. Joint relation in Italian.

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 723

3.4.13 JUSTIFY

The “justiﬁcations” consist of linguistic structures with argumentation functions. As claimed by

Ducrot et al.(1980) and Ducrot (1987), the justiﬁcation phenomenon presents two main ele-

ments: argumentation conjunctions and argumentation indicators. Tseronis (2011) explains

that augmentation conjunctions comprise a function of connecting two or more propositions,

while argumentation indicators are expressions or words like “only” and “almost,” which change

the argumentation potential within the boundaries of a proposition and connect a sentence to

another sentence. In RST, the justify relation provides to the reader’s comprehension of the satel-

lite increases the reader’s readiness to accept the writer’s right to the effect (Mann and Thompson

1987). Moreover, Carlson and Marcu (2001) claim that the satellite acts as an argument to con-

verse with the reader on the information presented by the writer within the nucleus. Also, the

nucleus presents a subjective statement or thesis or claim, which the reader might not accept or

might not regard as sufﬁciently important or positive, besides the satellite provides a statement of

a fundamental (e.g., political, moral) attitude of the acting person (Stede et al. 2017). Lastly, the

typical connectives are explanatory coordinate conjunctions. Table 13 describes the deﬁnitions for

the justify relation. Note that there are not any constraints on the nucleus and satellite. In addi-

tion, the constraints on the nucleus and satellite consist of the understanding that the satellite will

make it easier for readers to accept the nucleus or to share the particular viewpoint of the writer.

Finally, the justify relation affects the reader’s readiness to better accept the writer’s information

presented in the nucleus. Figures 39,40,and41 show examples of the justify relation. Observe

that the nucleus presents a fact related to the writer, which is justiﬁed by information provided in

the satellite. Therefore, to identify this relation, the nucleus must present a proposition, which is

justiﬁed by the proposition presented in satellite.

Table 13. Justify relation constraints

Type Description

Type of nucleus Mononuclear

.........................................................................................................................................................................................................................................................................

Type of relation Intentional/pragmatic

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus (N) N/A

.........................................................................................................................................................................................................................................................................

Constraints on the satellite (S) N/A

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) Reader’s comprehension S increases the reader’s readiness to

accept the writer’s right to present N

.........................................................................................................................................................................................................................................................................

Effect on the reader Reader’s readiness to accept writer’s right to present N is increased

Figure 39. Justify relation in Spanish.

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

724 F. Vargas et al.

Figure 40. Justify relation in Portuguese.

Figure 41. Justify relation in Italian.

3.4.14 MOTIVATION

Mann and Thompson (1987) claim that the motivation relation is commonly found in texts evok-

ing an action on the part of the reader, although advertising text typically largely consists of

motivating material. According to Stede et al. (2017), it establishes a relationship between the

nucleus and satellite, in which the nucleus should present an action to be performed by the reader,

and the satellite presents a reason for performing the action described in the nucleus. Moreover,

typical connectives used in the motivation relation are causal connectives. Finally, the motivation

increases the reader’s desire to perform the action, in contrast with the enablement that informs

the reader how to do it. These two relations are often found together, their satellites linked to a

common nucleus (Potter 2018). In a summarized way, this relation relates to any utterance that

expresses the speaker’s desire that the hearer performs some action (the nucleus) which material

will justify the requested action (the satellites). Table 14 describes the deﬁnitions for the motivation

relation. Notice it presents only one nucleus (mononuclear) and provides pragmatic (intentional)

information. It includes constraints on the nucleus, which must present an action unrealized.

Moreover, there are not any constraints on the satellite. Also, in the motivation relation, the

nucleus must present an action in which the reader is the actor (including accepting an offer),

unrealized concerning the context of the nucleus (Mann and Thompson 1987). Figures 42,43,

and 44 show examples of the motivation relation. Observe that the nucleus provides an action to

be performed and, in the satellite, some information that fruitfully motivates the reader to per-

form that action. Lastly, as is well-known from the previous discussion, each RST relation holds

different effects on the reader. Therephore, there is a desire of the reader to act on the nucleus.

Accordingly, the motivation relation may only be used when the reader is encouraged to perform

a certain activity (nucleus), on the grounds of the satellite (Stede et al. 2017).

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 725

Table 14. Motivation relation constraints

Type Description

Type of nucleus Multinuclear

.........................................................................................................................................................................................................................................................................

Type of relation Intentional/pragmatic

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus (N) N is an action in which the reader is the actor (including accepting

an offer), unrealized concerning the context of N

.........................................................................................................................................................................................................................................................................

Constraints on the satellite (S) N/A

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) The comprehension related to S increases the reader’s desire to

act on the N

.........................................................................................................................................................................................................................................................................

Effect on the reader Reader’s desire to act on the N is increased

Figure 42. Motivation relation in Spanish.

Figure 43. Motivation relation in Portuguese.

Figure 44. Motivation relation in Italian.

3.4.15 NON-VOLITIONAL CAUSE

More broadly, the deﬁnition for “volition,” taking into consideration linguistic studies, is related

to the intentional or unintentional nature of a subject or agent to act. According to the typology of

modalities, and later revisited and expanded according to discursive-functional grammar (FDG)

(Hengeveld et al., 2008), the volitional modality, as it is denominated by authors, the volitional

concept, is understood, concerning the domain semantic, as a type of modularization relating to

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

726 F. Vargas et al.

what is (un)desirable, being situated, hence, on the axis of volition, and presenting three types

of modal orientation: participant,event, and proposition (Hengeveld 2004). In RST, the non-

volitional cause relation is deﬁned as a type of relationship between the nucleus and satellite of

which the nucleus comprises a volitional action unrealized, and on the satellite, there are no con-

straints (Mann and Thompson 1987). Table 15 describes the deﬁnitions in terms of constraints

on the nucleus and satellite, as well as the combination of both for the non-volitional cause rela-

tion. Notice it presents only a nucleus and semantic information. Besides that, the nucleus must

hold at least one non-volitional action. Otherwise, there are no constraints on the satellite. As

previously stated, the RST relations carry an effect on the reader. Therefore, in this relation, the

reader recognizes the satellite’s direct cause of the nucleus. Figures 45,46,and47 show exam-

ples of the non-volitional cause relation. Observe that the satellite comprises a situation that, by

means, other than motivating a volitional action, caused the situation presented in the nucleus,

without the presence of the satellite (Mann and Thompson 1987). Finally, the nucleus is caused

by the non-volitional fact provided in the satellite.

Table 15. Non-volitional cause relation constraints

Type Description

Type of nucleus Mononuclear

.........................................................................................................................................................................................................................................................................

Type of relation Informational/semantic

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus (N) N is not a volitional action

.........................................................................................................................................................................................................................................................................

Constraints on the satellite (S) N/A

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) S, by means other than motivating a volitional action, caused N;

without the presentation of S, the reader might not know the

particular cause of the situation; a presentation of N is more

central than S to the writer’s purposes in putting forth the N-S

combination

.........................................................................................................................................................................................................................................................................

Effect on the reader Reader recognizes S as a cause of N

Figure 45. Non-volitional cause relation in Spanish.

Figure 46. Non-volitional cause relation in Portuguese.

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 727

Figure 47. Non-volitional cause relation in Italian.

3.4.16 NON-VOLITIONAL RESULT

The point of departure to understand the non-volitional result relation consists of the fact that it

presents the same set of proprieties from non-volitional cause,exceptforthefactthattheposi-

tion of the nucleus and satellite is inversely proportional. As discussed previously, the deﬁnition

of “volition” refers to the intentional or unintentional nature of a subject or agent to act. In RST,

the non-volitional result relation is deﬁned as a type of relationship between the nucleus and satel-

lite, in which the satellite comprises necessarily a non-volitional action, even though there are no

constraints on the nucleus (Mann and Thompson 1987). Once again, the relationship between the

nucleus and satellite is inversely proportional to the non-volitional cause.Table16 describes the

deﬁnitions for the non-volitional result relation. Notice it is mononuclear and provides semantic

information. Furthermore, there are no constraints on the nucleus; however, the satellite must

supply a non-volitional action. Lastly, in the non-volitional result relation, the reader recognizes

that the nucleus may have caused the information provided in the satellite. Figures 48,49,and

50 show examples of this relation. Observe that the nucleus comprises information more relevant

than the situation presented in the satellite. In addition, the nucleus provides relevant informa-

tion to understand the context, and the information presented in the nucleus is caused by the

situation provided in the satellite.

Table 16. Non-volitional result relation constraints

Type Description

Type of nucleus Mononuclear

....................................................................................................................................................................................................................

Type of relation Informational/semantic

....................................................................................................................................................................................................................

Constraints on the nucleus (N) N/A

....................................................................................................................................................................................................................

Constraints on the satellite (S) S is not a volitional action

....................................................................................................................................................................................................................

Constraints on the nucleus

and satellite (N +S)

N caused S; presentation of N is more central to the writer’s

purposes in putting forth the N-S combination than is the

presentation of S

....................................................................................................................................................................................................................

Effect on the reader R recognizes that N could have caused the situation in S

Figure 48. Non-volitional result relation in Spanish.

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

728 F. Vargas et al.

Figure 49. Non-volitional result relation in Portuguese.

Figure 50. Non-volitional result relation in Italian.

3.4.17 OTHERWISE

In RST, the otherwise relation is deﬁned as a type of mutually exclusive relation between two

elements of equal importance, in which the situations presented by both the satellite and the

nucleus are unrealized (Carlson and Marcu, 2001). In addition, the realization of the situation

associated with the nucleus will prevent the realization of the consequences associated with the

satellite. In the same settings, Stede et al.(2017) argue that the this relation comprises a type

of relationship between the nucleus and satellite, in which the nucleus presents a hypothetical,

future, or in other words, unreal situation, and the satellite presents a hypothetical, future, or in

other ways unreal situation. Finally, the realization of the nucleus impedes the realization of the

satellite (Mann and Thompson 1987). Table 17 describes the deﬁnitions for the otherwise rela-

tion. Notice it is mononuclear and provides semantic information. Furthermore, the satellite, as

well as the nucleus, must hold an unrealized situation, and the reader recognizes the dependency

relation of prevention between the realization of the nucleus and the realization of the satellite

(Stede et al., 2017). Figure 51,52,and53 show examples. Observe that the nucleus provides a

hypothetical or future situation. Likewise, the satellite also provides a hypothetical or future sit-

uation. Accordingly, in the otherwise relation, the nucleus and satellite should accommodate an

unrealized situation.

Table 17. Otherwise relation constraints

Type Description

Type of nucleus Mononuclear

.........................................................................................................................................................................................................................................................................

Type of relation Informational/semantic

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus (N) N is an unrealized situation

.........................................................................................................................................................................................................................................................................

Constraints on the satellite (S) S is an unrealized situation

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) The realization of N impedes the realization of S

.........................................................................................................................................................................................................................................................................

Effect on the reader Reader recognizes the dependency relation of prevention between

the realization of N and the realization of S

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 729

Figure 51. Otherwise relation in Spanish.

Figure 52. Otherwise relation in Portuguese.

Figure 53. Otherwise relation in Italian.

3.4.18 PURPOSE

Carlson and Marcu (2001) claim that in the purpose relation, the situation presented in the satel-

lite is only putative; that is, it is yet to be achieved. Furthermore, this relation may be paraphrased

as “nucleus to satellite.” In the same settings, Stede et al.(2017)deﬁnethepurpose relation as a

causal relationship in a wide sense. In other words, the difference between the relations volitional

and non-volitional cause and result is that within the purpose relation, the satellite is signaled

as hypothetical/unrealized and represents the intention or goal of the acting person. The typical

connective in purpose relation is “to”. In a summarized way, this relation expresses purpose, objec-

tive, or end, and it is typically introduced by the ﬁnal adverbial subordinate conjunctions (e.g., to,

that, in order to); hence clauses classiﬁed as ﬁnal adverbial subordinate may be candidates for this

relations (e.g. “it is necessary for us to ﬁght so that we can triumph.”). Table 18 describes the deﬁni-

tions in terms of constraints on the nucleus, satellite, and the combination of both for the purpose

relation. Notice it presents only a nucleus and semantic information. Moreover, the nucleus com-

prises an action, and the satellite provides a situation that is unrealized. Regarding the effect on

the reader, the piece of information provided in the nucleus is recognized as a starting activity to

realize the satellite. Figures 54,55,and56 show examples. Observe that the nucleus supplies an

activity or action, and in the satellite, a hypothetical or unrealized situation is presented. In addi-

tion, on the report of Carlson and Marcu (2001) in this relation, the particle “to” to indicate the

inﬁnitive verbs must not be confused with a post-nominal modiﬁer.

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

730 F. Vargas et al.

Table 18. Purpose relation constraints

Type Description

Type of nucleus Mononuclear

.........................................................................................................................................................................................................................................................................

Type of relation Informational/semantic

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus (N) N is an activity

.........................................................................................................................................................................................................................................................................

Constraints on the satellite (S) S is a situation that is unrealized

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) S is to be realized through the activity in N

.........................................................................................................................................................................................................................................................................

Effect on the reader Reader recognizes that the activity in N is initiated to realize S

Figure 54. Purpose relation in Spanish.

Figure 55. Purpose relation in Portuguese.

Figure 56. Purpose relation in Italian.

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 731

3.4.19 RESTATEMENT

Mann and Thompson (1987)deﬁnetherestatement as a type of relation that establishes a rela-

tionship between the nucleus and satellite, in which the satellite restates the nucleus, where the

satellite and the nucleus are of comparable bulk. Furthermore, Carlson and Marcu (2001) claim

that it is always mononuclear, where the satellite and nucleus are of (roughly) comparable size.

Besides, the satellite reiterates the information presented in the nucleus, typically with slightly

different wording; nevertheless, it does not add to or interpret the information. In the same set-

tings, Stede et al.(2017) suggest that it provides a nucleus that precedes the satellite in the text,

and the satellite repeats the information given in the nucleus using different wording. Hence, the

nucleus and satellite are of roughly equal size, and the reader recognizes the satellite as a restate-

ment of the nucleus. The typical connective is in “other words”. Lastly, the restatement establishes

a relationship between the nucleus and the satellite, in which the satellite presents a reformation

of the information from the nucleus, therephore, paraphraseseare candidates for the restatement

relation. Table 19 describes the deﬁnitions for the restatement relation. Notice it is a mononu-

clear relation and provides semantic information. Furthermore, according to Stede et al.(2017),

the nucleus precedes the satellite in the text, and the satellite repeats the information given in the

nucleus using different wording. Also, there are not constraints on the nucleus and satellite, and

the nucleus and satellite are of roughly equal size. Figures 57,58,and59 show examples of this

relation. Observe that the nucleus presents information, which was reformulated by a piece of

information given by the satellite.

Table 19. Restatement relation constraints

Type Description

Type of nucleus Mononuclear

.........................................................................................................................................................................................................................................................................

Type of relation Informational/semantic

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus (N) N/A

.........................................................................................................................................................................................................................................................................

Constraints on the satellite (S) N/A

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) S restates N, where S and N are of comparable bulk; N is more

central to writer’s purposes than S is.

.........................................................................................................................................................................................................................................................................

Effect on the reader Reader recognizes S as a restatement of N

Figure 57. Restatement relation in Spanish.

eParaphrase consists of a linguistic phenomenon in which to repeat something written or spoken using different words,

often in a humorous form or in a simpler and shorter form that makes the original meaning clearer (Cambridge Dictionary).

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

732 F. Vargas et al.

Figure 58. Restatement relation in Portuguese.

Figure 59. Restatement relation in Italian.

3.4.20 SEQUENCE

The sequence relation comprises information sequentially chained. Nevertheless, temporal succes-

sion is not the only type of succession for which this relation might be appropriate. Others could

include descriptions of a group of cars according to size or cost, colors of the rainbow, who lives

in a row of apartments, etc. (Mann and Thompson 1987). Moreover, Carlson and Marcu (2001)

deﬁne it as a type of relation multinuclear that establishes a relationship between nuclei in which

a list of events is presented according to chronological order or inverted chronological order.In

the same settings, Stede et al.(2017) suggest that it consists of a relationship between two or vari-

ous nuclei, in which the nuclei describe states of affairs that occur in a particular temporal order,

and the reader recognizes the succession relationships among the nuclei. The typical connectives

are “then”, “before”, and “afterward”. Table 20 describes the deﬁnitions for the sequence relation.

Notice that the relevant property of this relation consists of the fact that the nuclei comprise

situations in a temporal sequence. In addition, the main difference between the sequence relation

Table 20. Sequence relation constraints

Type Description

Type of nucleus Multinuclear

......................................................................................................................................................................................................................................................

Type of relation Informational/semantic

......................................................................................................................................................................................................................................................

Constraints on the nucleus (N) N/A

......................................................................................................................................................................................................................................................

Constraints on the satellite (S) N/A

......................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) There is a succession relationship between the situations in the

nuclei

......................................................................................................................................................................................................................................................

Effect on the reader Reader recognizes the succession relationships among the

nuclei

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 733

Figure 60. Sequence relation in Spanish.

Figure 61. Sequence relation in Portuguese.

Figure 62. Sequence relation in Italian.

and other multinuclear relations such as the joint is the secession temporal relationships among

nuclei.Figures60,61,and62 show examples of this relation. Observe that the information pre-

sented in the nuclei provides a type of semantic information in sequence. Hence, the reader must

recognize the presence of a temporal sequence among the nucleus.

3.4.21 SOLUTIONHOOD

The solutionhood relation establishes a relationship between the nucleus and satellite, in which the

reader recognizes that the body of the text presents a solution to the problem of having to clean

ﬂoppy disk heads too often (Stede et al.2017). Furthermore, the terms “problem” and “solution”

are broader than one might expect. In Carlson and Marcu (2001), the authors titled problem-

solution this relation. Shortly, in order to identiy this relation, one textual span must comprise

a problem, and the other text span must comprise a solution. In addiation, it should be pointed

out that this relation can be mononuclear or multinuclear, depending on the context. When the

problem is perceived as more important than the solution, the problem is assigned the role of the

nucleus, and the solution is the satellite. In that case, the relation would be monocluear. Otherwise,

it would be multinuclear. Also, Stede et al.(2017) claim that the nucleus precedes the satellite in

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

734 F. Vargas et al.

the text, and the reader recognizes the nucleus as a solution to the problem presented in the satel-

lite. Lastly, this relation rarely is signaled by connectives (Stede et al.2017). Table 21 describes the

deﬁnitions for the solutionhood relation. Note that the content of the satellite may be regarded

as a problem. Otherwise, the nucleus presents a solution to the problem presented in the satel-

lite. Moreover, the reader must recognize the nucleus as a solution to the problem presented in

the satellite. Figures 63,64,and65 show examples of this relation. Observe that the nucleus and

satellites do not present any discourse marker or connective. In addition, the satellite explicitly

provides a solution proposal for the event presented in the nucleus.

Table 21. Solutionhood relation constraints

Type Description

Type of nucleus Mononuclear

.........................................................................................................................................................................................................................................................................

Type of relation Informational/semantic

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus (N) N is a solution to the problem presented in S

.........................................................................................................................................................................................................................................................................

Constraints on the satellite (S) S presents a problem

.........................................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) N is a solution to the problem presented in S

.........................................................................................................................................................................................................................................................................

Effect on the reader R recognizes N as a solution to the problem presented in S

Figure 63. Solutionhood relation in Spanish.

Figure 64. Solutionhood relation in Portuguese.

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 735

Figure 65. Solutionhood relation in Italian.

3.4.22 SUMMARY

Here, a relevant point of departure consists of the fact that the size of the summary presented in

the satellite is shorter than the size of the nucleus. Therefore, in the summary relation, the satel-

lite summarizes the information presented in the nucleus, and the emphasis is on the situation

presented in the nucleus. According to Stede et al.(2017), it is deﬁned as mononuclear that estab-

lishes a type of relationship between the nucleus and satellite, in which the satellite succeeds the

nucleus in the text and repeats the information given in the nucleus, however, in a shorter form.

Furthermore, the typical connectives are “short”, “shortly”, and “in a summarized way”. Table 22

describes the deﬁnitions for the summary relation. Note that the nucleus presents more than one

EDU. Although there are no constraints on the satellite, the nucleus position is succeeded by the

satellite position. Lastly, the satellites present the same information presented in the nucleus, how-

ever, shortly. Figures 66,67,and68 show examples of this relation. Observe that it establishes a

relationship between the nucleus and the satellite, in which the satellite presents a restatement

of the content of the nucleus, which is shorter in the bulk. Hence, the satellite summarizes the

information presented in the nucleus.

Table 22. Summary relation constraints

Type Description

Type of nucleus Mononuclear

.......................................................................................................................................................................................................................................................................

Type of relation Informational/semantic

.......................................................................................................................................................................................................................................................................

Constraints on the nucleus (N) N must be more than one unit

.......................................................................................................................................................................................................................................................................

Constraints on the satellite (S) N/A

.......................................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) The S presents a restatement of the content of N, which is shorter

in bulk

.......................................................................................................................................................................................................................................................................

Effect on the reader R recognizes S as a shorter restatement of N

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

736 F. Vargas et al.

Figure 66. Summary relation in Spanish.

Figure 67. Summary relation in Portuguese.

Figure 68. Summary relation in Italian.

3.4.23 VOLITIONAL CAUSE

According to Mann and Thompson (1987), the volitional cause relation is deﬁned as a type of

relationship between the nucleus and satellite in which the nucleus provides the cause of the voli-

tionalfaction presented in the satellite. In the same settings, Stede et al.(2017) claim that the

fVolitional is acting as a result of a decision or choice; done because someone has decided or chosen to fundamental

processes over which we have volitional control. The action is not mechanical; it is volitional (Cambridge Dictionary).

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 737

nucleus and satellite must provide a state or event in the world. In order words, the state/event

in the nucleus must be caused by the state/event in the satellite, and the reader must recognize

the satellite as a cause of the content provided in the nucleus. Furthermore, the volitional cause

relation is frequently confused with the motivation relation (see Section 3.4.14). Nonetheless, the

main element that distinguishes them consists of the intended effect of the motivation relation to

make the reader want to perform an action evoked in the text. In Table 23, we describe the deﬁni-

tions in terms of constraints on the nucleus and satellite along with the combination of both for

the volitional cause relation. Notice it is a mononuclear relation and provides semantic informa-

tion. Moreover, it is deﬁned as a type of relationship between the nucleus and satellites without

any constraints on the satellite, while the nucleus must provide information on a volitional action.

Besides, there are constraints on the nucleus and satellite together. Finally, the typical connectives

are “because”, “since”, and “therefore”. Figures 69,70,and71 show examples of the volitional cause

relation. Observe that the nucleus presents necessarily a volitional information. As it is known

Table 23. Volitional cause relation constraints

Type Description

Type of nucleus Mononuclear

............................................................................................................................................................................................................................................................

Type of relation Informational/semantic

............................................................................................................................................................................................................................................................

Constraints on the nucleus (N) N is a volitional action or else a situation that could have arisen

from a volitional action

............................................................................................................................................................................................................................................................

Constraints on the satellite (S) N/A

............................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) S could have caused the agent of the volitional action in N to

perform that action; without the presentation of S, R might not

regard the action as motivated or know the particular motivation;

N is more central to W’s purposes in putting forth the N-S

combination than S is

............................................................................................................................................................................................................................................................

Effect on the reader R recognizes S as a cause for the volitional action in N

Figure 69. Volitional cause relation in Spanish.

Figure 70. Volitional cause relation in Portuguese.

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

738 F. Vargas et al.

Figure 71. Volitional cause relation in Italian.

from previous discussions, the conception of “volition” according to linguistics studies is related

to the intentional or unintentional nature of a subject or agent to act.

3.4.24 VOLITIONAL RESULT

According to Mann and Thompson (1987)thevolitional result relation comprises a volitional

action or situation that may appear from a volitional action. In Stede et al.(2017), the volitional

result relation is titled “result” relation. In the same settings, they deﬁne this relation as type of

relationship between nucleus and satellite, in which the information presented in the satellite is

the cause of the situation presented in the nucleus. Furthermore, the result, which is the nucleus,

is the most important part of this relation. Therefore, without accessing the satellite, the reader

may not know what caused the result in the nucleus. Lastly, the typical connectives are “because”,

“since”, and “therefore”. Table 24 describes the deﬁnitions in terms of constraints on the nucleus

and satellite and the combination of both. Notice it is a mononuclear relation that comprises

semantic information and the relationship between the nucleus and satellite. In this relation, the

very important point that we should keep in mind consists of the fact that the constraints on the

nucleus and satellite present information that are inversely proportional to the volitional cause

relation. Furthermore, in this relation, there are no contractions on the nucleus, and the satellite

provides a volitional action. Whether it is compared with the volitional cause, the nucleus pro-

vides a volitional action, and no contractions are on the satellite. Additionally, in the volitional

result, the reader recognizes that the nucleus probably has caused the action supply in the satel-

lite. Figures 72,73,and74 show examples of the volitional result relation. Observe that the satellite

comprises necessarily the presence of volitional information, and the nucleus provides the result

of volitional action presented by satellite.

Table 24. Volitional result relation constraints

Type Description

Type of nucleus Mononuclear

............................................................................................................................................................................................................................................................

Type of relation Informational/semantic

............................................................................................................................................................................................................................................................

Constraints on the nucleus (N) N/A

............................................................................................................................................................................................................................................................

Constraints on the satellite (S) S is a volitional action or a situation that could have arisen from a

volitional action

............................................................................................................................................................................................................................................................

Constraints on the nucleus and satellite (N +S) N could have caused S; the presentation of N is more central to the

writer’s purposes than is the presentation of S

............................................................................................................................................................................................................................................................

Effect on the reader Reader recognizes that N could be a cause for the action or

situation in S

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 739

Figure 72. Volitional result relation in Spanish.

Figure 73. Volitional result in Portuguese.

Figure 74. Volitional result in Italian.

4. Annotation bias

Annotation bias or annotator bias consists of differences between annotator preferences for sub-

jective reasons (Amidei, Piwek, and Willis 2020). According to Sampson and Babarczy (2008)

and Amidei et al. (2018), annotators most diverge in language annotation tasks due to a range of

ineliminable factors such as background knowledge, preconceptions about language, geographic

factors (origin) and general educational level. In this survey, we aim to provide a well-deﬁned,

structured and accurate discourse annotation guideline focused on LRLs, which is easy to follow

and rich in examples of discourse coherence relations, as a strategy to mitigate potential anno-

tator bias. We also recommend two strategies to mitigate bias: (i) selection of expert annotators

and (ii) diverse proﬁle of annotators (gender, race, political orientation, etc.).

5. Evaluation of RST trees

Evaluation metrics of discourse coherence are important to distinguish coherent texts from inco-

herent texts. The evaluation of RST trees has focused on (i) human-based evaluation,which

uses expert human raters, and (ii) automatic-based evaluation, which uses classical and neu-

ral machine learning (ML) algorithms. Guz et al. (2020) propose to evaluated the RST tree using

neural networks leveraging its representations as features to evaluate coherent texts. Wan et al.

(2019) compared automatically RST trees. Naismith et al. (2023) produced ratings by training

GPT-4 contrasting with expert human raters to assess discourse coherence.

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

740 F. Vargas et al.

6. Main challenges and opportunities

While the RST has been applied to a wide variety of successful applications, we should not simply

see it without any criticism. For instance, discourse coherence relations may present a relevant

level of ambiguity. In addition, the discourse segmentation process was not accurately deﬁned by

RST’s authors. Hence, the identiﬁcation automatic of an EDU is a complex task. Nevertheless,

discourse-aware computational resources have proved to be useful and efﬁcient in different NLP

applications. Lei et al. (2022) showed that embedded discourse structure for sentence-level media

bias effectively increases the recall by 8.27 percent–8.62 percent and precision by 2.82 percent–

3.48 percent. Devatine et al. (2022) predicted the political orientation (left, center, right) of news

articles using a discourse framework. Aldogan and Yaslan (2015) and Appel et al. (2016) evaluated

different features including discourse structures for sentiment analysis. Alós (2015) used discourse

structure for automatic translation. Huang (2013) applied RST to building clinical question-

answering systems. Hewett (2023) applied RST for the text simpliﬁcation task, and Vargas et al.

(2021) used RST for multilingual fake news detection. Xu et al. (2020) proposed a discourse-aware

neural summarization model, which extracts sub-sentential discourse units (instead of sentences)

as candidates based on RST trees and coreference mentions. And, Li et al. (2020) showed that an

EDU is a more appropriate textual unit of content selection than the sentence unit for abstractive

summarization. RST is also used for natural language generation (Mann 1984;Hovy1990;Isard

2016; Adewoyin, Dutta, and He 2022) and for building of discourse parsing (Li, Li, and Hovy 2014;

Mabona et al. 2019).

7. Final remarks

We provide the ﬁrst discourse annotation guideline using the RST for LRLs. Speciﬁcally, we

accurately described 24 (twenty-four) discourse coherence relations in three romance languages:

Italian, Portuguese, and Spanish. We also present a comprehensive survey addressing discourse

analysis in AI, hence offering an accessible resource to new researchers and annotators. Finally,

we hope to contribute to the advancement of research and NLP systems development focused on

discourse-level language understating and generation for LRLs.

Acknowledgments. The authors are grateful to Marcello Gecchele for providing reliable and expert RST-annotated

examples in Italian.

Competing interests. The author(s) declare none.

References

Adewoyin R.,Dutta R. and He Y. (2022). RSTGen: Imbuing ﬁne-grained interpretable control into long-FormText genera-

tors. In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics:

Human Language Technologies, Seattle, USA, pp. 1822–1835.

Aldogan D. and Yaslan Y. (2015). A Comparison Study On Ensemble Strategies and Feature Sets for Sentiment Analysis.In

Proceedings of the 30th International Symposium on Computer and Information Sciences, London, UK, pp. 359–370.

Alós J. (2015). Discourse relation recognition in translation: A relevance-theoretic perspective. Perspectives,24(2), 201–217.

Amidei J.,Piwek P. and Willis A. (2018). Rethinking the agreement in human evaluation tasks. In Proceedings of the 27th

International Conference on Computational Linguistics, New Mexico, USA, pp. 3318–3329.

Amidei J.,Piwek P. and Willis A. (2020). Identifying annotator bias: A new IRT-based method for bias identiﬁcation. In

Proceedings of the 28th International Conference on Computational Linguistics, Held Online, pp. 4787–4797.

Androutsopoulos I.,Lampouras G. and Galanis D. (2013). Generating natural language descriptions from owl ontologies:

The naturalowl system. Journal of Artiﬁcial Intelligence Research 48, 671–715.

Appel O.,Chiclana F.,Carter J. and Fujita H. (2016). A hybrid approach to the sentiment analysis problem at the sentence

level. Knowledge-Based Systems 108, 110–124.

Ariza-Casabona A.,Schmeisser-Nieto W. S.,Nofre M.,Taul é M . ,Amigó E.,Chulvi B. and Rosso P. (2022). Over view

of DETESTS at IberLEF 2022: DETEction and classiﬁcation of racial STereotypes in Spanish. Procesamiento del lenguaje

natural. Number 69. pp. 217–228. http://journal.sepln.org/sepln/ojs/ojs/index.php/pln/article/view/6442

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 741

Asher N. and Lascarides A. (2003). Logics of Conversation. Studies in Natural Language Processing. Cambridge University

Press. 526 pp.

Braud C.,Hardmeier C.,Li J.J.,Loaiciga M. and Zeldes A. (eds.) (2022). Proceedings of the 3rd Workshop on Computational

Approaches to Discourse, Gyeongju, Republic of Korea.

Braud C.,Hardmeier C.,Li J. J.,Louis A. and M. Strube (eds.) (2020). Proceedings of the 1st Workshop on Computational

Approaches to Discourse, Held Online.

Braud C.,Hardmeier C.,Li J. J.,Louis A.,Strube M. and Zeldes A. (eds.) (2021). Proceedings of the 2nd Workshop on

Computational Approaches to Discourse, Punta Cana, Dominican Republic.

Bussmann H. (1998). Routledge Dictionary of Language and Linguistics. Translated and edited by Gregory Trauth and Kerstin

Kazzazi, London: Routledge.

Carlson L. and Marcu D. (2001). Discourse tagging manual. Tech.rep.ISI-TR-545, 01–87. https://tinyurl.com/v8y2ju76

Castagnola L. (2002). Anaphora resolution for question answering, Master’s thesis. Massachusetts Institute of Technology,

Massachusetts, United States.

Cieri C.,Maxwell M.,Strassel S. and Tracey J. (2016). Selection criteria for low resource language programs. In Proceedings

of the 10th International Conference on Language Resources and Evaluation, Portorož, Slovenia, pp. 4543–4549.

Devatine N.,Muller P. and Braud C. (2022). Predicting political orientation in news with latent discourse structure to

improve bias understanding. In Proceedings of the 3rd Workshop on Computational Approaches to Discourse, Gyeongju,

Republic of Korea and Online, pp. 77–85.

Dreyfus S. and Bennett I. (2017). Circumstantiation: Taking a broader look at circumstantial meanings. Functional

Linguistics 1(4-5), 1–31.

DuBois J. W. (2003). Discourse and grammar. In Tomasello M., (ed), The New Psychology of Language: Cognitive and

Functional Approaches to Language Structure,vol.2, Lawrence Erlbaum Associates Publishers, pp. 47–87.

Ducrot O. (1987). O Dizer e o dito. Pontes, Campinas: 222 pp.

Ducrot O.,Bruxelles S. and Bourcier D. (1980). Les mots du discours. les editions de minuit ed. France.

Fairclough N. (2003). Analysing Discourse: Textual Analysis for Social Research. London and New York: Routledge Taylor &

Francis Group.

Fawcett R. P. and Davies B. L. (1992). Monologue as a turn in dialogue: Towards an integration of Exchange Structure and

Rhetorical Structure Theory.InProceedings of the 6th International Workshop on Natural Language Generation, Trento,

Italy, pp. 151–166.

Fraser B. (1999). What are discourse markers? Journal of Pragmatics 31(7), 931–952.

Grice H. P. (1975). Logic and conversation. In Syntax and Semantics: Vol. 3: Speech Acts, New York, Speech Acts.

Grosz B. J. (1987). Whither discourse and speech acts?. In Wilks Y., (ed), Theoretical Issues in Natural Language Processing,

vol. 3.

Grosz B. J. and Sidner C. L. (1986). Attention, intentions, and the structure of discourse. Computational Linguistics 12(3),

175–204.

Guz G.,Bateni P.,Muglich D. and Carenini G. (2020). Neural RST-based evaluation of discourse coherence. In Proceedings

of the 1st Conference of the Asia-Paciﬁc Chapter of the Association for Computational Linguistics and the 10th International

Joint Conference on Natural Language Processing, Suzhou, China, pp. 664–671.

Halliday M. (1995). An Introduction to Functional Grammar,1sted.Arnold,London.

Hengeveld Kees and Mackenzie J. L. (2008). Functional Discourse Grammar: A Typologically Based Theory of Language

Structure. Oxford Linguistics, Oxford..

Hengeveld K. (2004). Illocution, mood, and modality. In Booij B., Lehmann C. and Mugdan J., (eds), Morphology:

A Handbook On Inﬂection and Word Formation. 2nd ed. Berlin: Mouton de Gruyter, pp. 1190–1201.

Hewett F. (2023). APA-RST: A text simpliﬁcation corpus with RST annotations. In Proceedings of the 4th Workshop on

Computational Approaches to Discourse, Toronto, Canada, pp. 173–179.

Hobbs J. R. (1979). Coherence and coreference. Cognitive Science 3(1), 67–90.

Hou S.,Zhang S. and Fei C. (2020). Rhetorical structure theory: A comprehensive review of theory, parsing methods and

applications. Expert Systems with Applications 157, 113421.

Hovy E. (1992). A new level of language generation technology - capabilities and possibilities. IEEE Expert-Intelligent Systems

& Their Applications 7(2), 12–17.

Hovy E. (1993a). Automated discourse generation using discourse structure relations. Artiﬁcial Intelligence 63(1-2), 341–385.

Hovy E. (1993b). In defense of syntax: Informational, intentional, and rhetorical structures in discourse. In Intentionality and

Structure in Discourse Relations, pp. 35–39. https://aclanthology.org/W93-0210.pdf

Hovy E. H. (1990). Parsimonious and proﬂigate approaches to the question of discourse structure relations. In Proceedings of

the 5th International Workshop on Natural Language Generation, Pennsylvania, USA, pp. 128–136.

Huang X. (2013). Applying a generic function-based topical relevance typology to structure clinical questions and answers.

Journal of the American Society for Information Science and Technology 64(1), 65–85.

Isard A. (2016). The methodius corpus of rhetorical discourse structures and generated texts. In Proceedings of the 10th

International Conference on Language Resources and Evaluation, Portorož, Slovenia, pp. 1732–1736.

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

742 F. Vargas et al.

Jurafsky D. (2020). Discourse coherence. In Speech and Language Processing, Stanford University, pp. 01–25.

https://web.stanford.edu/~jurafsky/slp3/old_jan23/27.pdf

Jurafsky D. and Martin J. H. (2009). Speech and Language Processing: An Introduction to Natural Language Processing,

Computational Linguistics, and Speech Recognition, Prentice Hall Series in Artiﬁcial Intelligence, 2nd ed. Pearson Education

International, Prentice Hall, NJ.

Khan M.,Ullah K.,Alharbi Y.,Alferaidi A.,Alharbi T. S.,Yad a v K. ,Alsharabi N. and Ahmad A. (2023). Understanding

the research challenges in low-resource language and linking bilingual news articles in multilingual news archive. Applied

Sciences 13(15), 8566.

Kim Y.-B. (2001). Concession and linguistic inference. In Proceedings of the 16th Paciﬁc Asia Conference on Language,

Information and Computation, Jeju, Korea, pp. 187–194.

Lei Y.,Huang R.,Wang L. and Beauchamp N. (2022). Sentence-level media bias analysis informed by discourse structures.

In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, Abu Dhabi, United Arab

Emirates, pp. 10040–10050.

Li J.,Li R. and Hovy E. (2014). Recursive deep models for discourse parsing. In Proceedings of the 2014 Conference on

EmpiricalMethodsinNaturalLanguageProcessing, Doha, Qatar, pp. 2061–2069.

Li J.,Sun A. and Joty S. (2018). Segbot: A generic neural text segmentation model with pointer network. In Proceedings of

the 27th International Joint Conference on Artiﬁcial Intelligence, pp. 4166–4172.

Li Z.,Wu W. and Li S. (2020). Composing elementary discourse units in abstractive summarization. In Proceedings of the

58th Annual Meeting of the Association for Computational Linguistics, Online, pp. 6191–6196.

Mabona A.,Rimell L.,Clark S. and Vlachos A. (2019). Neural generative rhetorical structure parsing. In Proceedings of

the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on

Natural Language Processing, Hong Kong, China, pp. 2284–2295.

Mann W. C. (1984). Discourse structures for text generation. In 10th International Conference on Computational Linguistics

and 22nd Annual Meeting of the Association for Computational Linguistics, Stanford, California, pp. 367–375.

Mann W. C.,Matthiessen C. M. I. M. and Thompson S. A. (1992). Rhetorical structure theory and text analysis. In

Discourse Description: Diverse Linguistic Analyses of a Fund-Raising Text. Amsterdam and Philadelphia: John Benjamins,

pp. 39–78.

Mann W. C. and Thompson S. A (1987). Rhetorical Structure Theory: A Theory of Text Organization. Technical Report.

RS-87-190, Information Sciences Institute. University of Southern California, Los Angeles, USA. pp. 1–82.

Marcu D. (2000). The rhetorical parsing of unrestricted texts: a surface-based approach. Computational Linguistics 26(3),

395–448.

Marcu D. and Echihabi A. (2002). An unsupervised approach to recognizing discourse relations. In Proceedings of the 40th

Annual Meeting of the Association for Computational Linguistics, Philadelphia, USA, pp. 368–375.

Martin J. R. (1992). English Text: System and structure. John Benjamins, Amsterdam.

Megerdoomian K. and Parvaz D. (2008). Low-density language bootstrapping: the case of Tajiki Persian. In Proceedings of

the 6th International Conference on Language Resources and Evaluation, Marrakech, Morocco, pp. 3293–3298.

Moore J. D. and Wiemer-Hastings P. (2003). Discourse in Computational Linguistics and Artiﬁcial Intelligence. In Handbook

of Discourse Processes, 1st ed., University of Edinburgh, West.

Mukherjee S. and Joshi S. (2013). Sentiment aggregation using ConceptNet ontology. In 6th International Joint Conference

on Natural Language Processing, Nagoya, Japan, pp. 570–578.

Naismith B.,Mulcaire P. and Burstein J. (2023). Automated evaluation of written discourse coherence using GPT-4. In

Proceedings of the 18th Workshop on Innovative Use of NLP for Building Educational Applications, Toronto, Canada,

pp. 394–403.

Nunan D. (1993). Introducing Discourse Analysis. London: Penguin English.

Passonneau R. J. and Litman D. J. (1997). Discourse segmentation by human and automated means. Computational

Linguistics 23(1), 103–139.

Potter A. (2018). Reasoning between the lines: A logic of relational propositions. Dialogue & Discourse 9(2).

https://journals.uic.edu/ojs/index.php/dad/article/view/10690

Prasad R.,Dinesh N.,Lee A.,Miltsakaki E.,Robaldo L.,Joshi A. and Webber B. (2008). The Penn Discourse TreeBank 2.0,

Proceedings of the 6th International Conference on Language Resources and Evaluation, Marrakech, Morocco, pp. 2961–

2968.

Prevot L.,Hunter J. and Muller P. (2023). Comparing methods for segmenting elementary discourse units in a French

conversational corpus. In Alumäe T. and Fishel M. (eds), Proceedings of the 24th Nordic Conference on Computational

Linguistics (NoDaLiDa), Tórshavn, Faroe Islands, University of Tartu Library, pp. 436–446.

Ramsay A. (2005). Discourse. In Mitkov R., (ed), The Oxford Handbook of Computational Linguistics,vol.1, Oxford

University Presss, Inc, pp. 112–135.

Rohde H.,Johnson A.,Schneider N. and Webber B. (2018). Discourse coherence: Concurrent explicit and implicit relations,

Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, Melbourne, Australia, pp. 2257–

2267.

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Natural Language Processing 743

Sampson G. and Babarczy A. (2008). Deﬁnitional and human constraints on structural annotation of English. Natural

Language Engineering 14(4), 471–494.

Stede M.,Tab o a d a M . a n d D a s D . (2017). Annotation Guidelines for Rhetorical Structure. Linguistics Department at The

University of Potsdam, pp. 1–31. https://www.sfu.ca/~mtaboada/docs/research/RST_Annotation_Guidelines.pdf

Strube M.,Braud C.,Hardmeier C.,Li J. J.,Loaiciga S.,and Zeldes A., (eds.). Proceedings of the 4th Workshop on

Computational Approaches to Discourse, Toronto, Canada.

Sweetser E. (1990). From Etymology to Pragmatics: Metaphorical and Cultural Aspects of Semantic Structure. Cambridge

Studies in Linguistics. Cambridge University Press.

Thompson S. A. and Mann W. C. (1988). Rhetorical structure theory: A framework for the analysis of texts. IPRA Papers in

Pragmatics 1, 79–105.

Trnavac R.,Das D. and Taboada M. (2016). Discourse relations and evaluation. Corpora 11(2), 169–190.

Tser o n i s A . (2011). From connectives to argumentative markers: A quest for markers of argumentative moves and of related

aspects of argumentative discourse. Argumentation: an International Journal on Reasoning 25(4), 427–447.

Vargas F.,Benevenuto F. and Pardo T. (2021). Toward discourse-aware models for multilingual fake news detection. In

Proceedings of the International Conference Recent Advances in Natural Language Processing - Student Research Workshop,

Held Online, pp. 210–218.

Vargas F.,D’Alessandro J.,Rabinovich Z.,Benevenuto F. and Pardo T. (2022). Rhetorical structure approach for online

deception detection: A survey. In Calzolari N., Béchet F., Blache P., Choukri K., Cieri C., Declerck T., Goggi S., Isahara

H., Maegaard B., Mariani J., Mazo H., Odijk J. and Piperidis S. (eds), Proceedings of the Thirteenth Language Resources and

Evaluation Conference, Marseille, France, European Language Resources Association, pp. 5906–5915.

Wan S.,Kutschbach T.,Lüdeling A. and Stede M. (2019). RST-tace a tool for automatic comparison and evaluation

of RST trees. In Proceedings of the Workshop on Discourse Relation Parsing and Treebanking 2019, Minneapolis, USA,

pp. 88–96.

Wiebe J.,Wilson T.,Bruce R.,Bell M. and Martin M. (2004). Learning subjective language. Computational Linguistics 30(3),

277–308.

Wiemerslage A.,Silfverberg M.,Yang C . ,McC arthy A.,Nicolai G.,Colunga E. and Kann K. (2022). Morphological process-

ing of low-resource languages: Where we are and what’s next. In Findings of the Association for Computational Linguistics.

Dublin, Ireland, pp. 988–1007.

Xu J.,Gan Z.,Cheng Y. and Liu J. (2020). Discourse-aware neural extractive text summarization. In Proceedings of the 58th

Annual Meeting of the Association for Computational Linguistics, Held Online, pp. 5021–5031.

Cite this article: Vargas F, Schmeisser-Nieto W, Rabinovich Z, Pardo TAS and Benevenuto F (2025). Discourse annotation

guideline for low-resource languages. Natural Language Processing 31, 700–743. https://doi.org/10.1017/nlp.2024.19

https://doi.org/10.1017/nlp.2024.19 Published online by Cambridge University Press

Discourse annotation guideline for low-resource languages

Abstract

Recommended publications

Submit your application to win an all-inclusive 11-days at Sao Paulo School of Advanced Sciences on...

Rhetorical Structure Approach for Online Deception Detection: A Survey

Developing a Rhetorical Structure Theory Treebank for Czech

Predicting Sentence-Level Factuality of News and Bias of Media Outlets

NoHateBrazil: A Brazilian Portuguese Text Offensiveness Analysis System

Context-aware and expert data resources for Brazilian Portuguese hate speech detection