DiscoExplorer: An Open Interface for the Study of Multilingual Discourse Relations

# An Open Interface for the Study of Multilingual Discourse Relations
Source: [https://arxiv.org/html/2605.15304](https://arxiv.org/html/2605.15304)
###### Abstract

The relations connecting propositions in discourse such ascause\(A because B\) orconcession\(A although B\) are a subject of intense interest in Computational Linguistics and Pragmatics, but challenging to study and compare across languages\. Recent progress in standardizing discourse relation inventories across datasets offers the potential to facilitate such studies, but is hindered by the complexity of relevant data and the lack of easily accessible interfaces to analyze it\. In this paper we present DiscoExplorer, a new open source web interface, capable of running on local computers, which we use to make datasets from the DISRPT Shared Task on discourse relation classification publicly available, covering 16 different languages\. We present the query language, search and visualization facilities for relations and signaling devices such as connectives, as well as some example studies\.

DiscoExplorer: An Open Interface for the Study of Multilingual Discourse Relations

Amir ZeldesGeorgetown Universityamir\.zeldes@georgetown\.edu

## 1Introduction

Discourse relations are the implicit and explicit semantic/pragmatic connections that arise when multiple propositions are juxtaposed in a text or conversation\. For example, in[1](https://arxiv.org/html/2605.15304#S1), the explicit connective ‘when’ indicates atemporalrelation between the two arguments 1 and 2, while thecausalrelation between 1 and 3 is understood implicitly \(Jin is upsetbecauseKim left\)\.

\\ex

\.\[\[Kim left\]1\]\_\{1\}\[\[when Jin arrived\.\]2\]\_\{2\}\[\[Jin is upset now\.\]3\]\_\{3\}

A variety of theories have attempted to describe discourse relations and construct datasets for their study, including Rhetorical Structure Theory \(RST,Mann and Thompson[1988](https://arxiv.org/html/2605.15304#bib.bib1)\), Segmented Discourse Representation Theory \(SDRT,Asher and Lascarides[2003](https://arxiv.org/html/2605.15304#bib.bib2)\), the Penn Discourse Treebank \(PDTB,Prasadet al\.[2014](https://arxiv.org/html/2605.15304#bib.bib3)\) and discourse dependenciesMoreyet al\.\([2018](https://arxiv.org/html/2605.15304#bib.bib4)\)\. However because each theory and dataset has tended to use distinct relation inventories and data structures \(for example hierarchical trees, graphs, or pairs of text spans\), comparisons across languages or even datasets in the same language have been challenging\.

More recently, the DISRPT shared taskBraudet al\.\([2024](https://arxiv.org/html/2605.15304#bib.bib7)\)has made progress in unifying data from such formalisms by focusing on what they have in common: the postulation of relations between parts of a text, and optional inclusion of information about signaling devices, for example the distinction between implicit and explicit relations above\. In its most recent edition the shared task also unified relation labels across 38 datasets in 16 different languagesBraudet al\.\([2025](https://arxiv.org/html/2605.15304#bib.bib8)\), facilitating cross\-linguistic comparisons for the first time, similarly to initiatives to consolidate labels for describing multilingual syntactic functions in projects such as Universal Dependencies \(UD,de Marneffeet al\.[2021](https://arxiv.org/html/2605.15304#bib.bib9)\)\. However what has been missing compared to projects like UD is an easily accessible interface to search and compare data, identify errors, and visualize patterns in datasets\. The main contributions of this short paper aim to fill this gap:

- •We provide a high performance, open source, client\-side interface in pure JavaScript that can be run on any PC
- •We make the datasets from the DISRPT shared task searchable online for the public
- •We propose a simple, flexible query language to facilitate access for new users

## 2Related work

While many local graph search tools exist for linguistic data, such as Semgrex, Ssurgeon or Semgrex\-PlusTamburini \([2017](https://arxiv.org/html/2605.15304#bib.bib23)\); Baueret al\.\([2023](https://arxiv.org/html/2605.15304#bib.bib22)\), almost all are limited to searching within sentence boundaries, and are therefore not capable of representing relations across entire texts\. Several online interfaces have facilitated search in syntactically and even semantically annotated treebanksGuibonet al\.\([2020](https://arxiv.org/html/2605.15304#bib.bib5)\); Amblardet al\.\([2022](https://arxiv.org/html/2605.15304#bib.bib6)\), but dedicated interfaces for discourse relations are rare, and have generally been fitted to a single resource and theory, such as the Spanishda Cunhaet al\.\([2011](https://arxiv.org/html/2605.15304#bib.bib10)\)and BasqueIruskietaet al\.\([2013](https://arxiv.org/html/2605.15304#bib.bib11)\)RST treebank interfaces\. Converters for RST data exist to enable searching through data using ANNISKrause and Zeldes \([2016](https://arxiv.org/html/2605.15304#bib.bib12)\), a generic multilayer corpus search interface\. However the system is considerably heavier, slower and has a complex query language which is not tailored to discourse relations, and currently cannot import data from other discourse formalisms or the DISRPT format\.

Our work takes its primary inspiration from the Grew Match search interface for UD treebanksGuibonet al\.\([2020](https://arxiv.org/html/2605.15304#bib.bib5)\), which leverages the consistent format and label inventory of the UD project to allow access to treebanks using a consistent query language and architecture\.

## 3DiscoExplorer

### 3\.1Architecture

Our architecture is designed with three goals in mind: 1\. minimizing compute costs to prevent needing a dedicated \(and expensive\) server; 2\. making it possible to run the interface locally for users with proprietary data that cannot be exposed online; and 3\. running a fast and responsive search with minimal dependencies\. To achieve these goals, we implemented a client\-side solution in JavaScript using React, without a database backend, no dedicated indexing \(e\.g\. Meilisearch\) or visualization libraries \(e\.g\. D3\.js\)\. Instead, we focus on using pure JavaScript, HTML and CSS wherever possible to ensure stability and longevity of the software\.

Our data model focuses on discourse relations as the instance to be searched over, where relations are aligned to token positions in documents and span over two possibly discontinuous argument spans \(e\.g\. the cause and effect forcausalrelations\)\. Relations that do not cover entire sentences are also associated with context spans indicating words before, after or between the arguments within the same sentences, ensuring that full sentence context is provided with each match\. Finally, relations carry labels, a direction \(1\>2 or 1<2\) and possibly a list of typed and subtyped signal tokens, for datasets marking connectives or other signal types\.

### 3\.2Basic interface

The web interface is arranged around two areas: the query form at the top of Figure[1](https://arxiv.org/html/2605.15304#S3.F1), and the results area at the bottom, which can display concordances for qualitative searches, or switch to a ‘frequencies’ tab for quantitative analysis\. The interface was initially tested with students in a seminar on computational models of discourse at Georgetown University \(LING\-8415\), and based on student feedback, an additional tab was added to perform comparisons between datasets\. We are also planning to collect feedback from CODI attendees and the DISRPT community to develop additional features\.

![Refer to caption](https://arxiv.org/html/2605.15304v1/latex/search-interface.png)Figure 1:DiscoExplorer search interface: Users can input a query and select filters\. Underlines show query matches and signals are highlighted \(e\.g\. red for discourse markers, yellow for lexical signals\)\.The basic query form is meant to be user friendly by exposing the available datasets and labels in each dataset as drop down filters\. Negation of a filter is realized as a simple checkbox – for example, selecting the labelconditionand the negation box for any signal type ineng\.erst\.gum\(data from the GUM corpus,Zeldes[2017](https://arxiv.org/html/2605.15304#bib.bib13)\) yields examples of implicit conditionals, as in[1](https://arxiv.org/html/2605.15304#S3.F1)\. More complex queries must utilize the DiscoExplorer query language \(DEQL\) described in[3\.3](https://arxiv.org/html/2605.15304#S3.SS3)\.

\\ex

\.\[\[you take this painting,\]1\]\_\{1\}\[\[I want that recorder\]2\]\_\{2\}\(=if you take this painting\)

### 3\.3Query language – DEQL

Our query language aims to be simple but powerful, meaning on the one hand, it should respond as expected to simply typing words in the search box, while on the other hand allowing users to do exact sequence or flexible match queries, queries restricted to the first/second or source/target spans of the relation, as well as leveraging token annotations\. Since DISRPT data is released with accompanying UD annotations, we expose the UD POS tags, dependency labels and lemmas directly for querying\. All queries can be restricted by the UI to a specific relation label chosen from a drop down list \(either the universal DISRPT label, or each dataset’s original labels, or both\), specific signal types or subtypes if available in the data \(e\.g\. explicit connectives\), and relation directions\. The exact query can be saved and reproduced via a shareable link\.

As an example of simple text based queries and their interactions with argument spans, consider the differences between the following, all executed with the ‘exact sequence’ match turned off and theconditionlabel selected:111UI filters are indicated in red and are not part of the query string, but are stored in reproducible shareable query links\.

\\ex

\.conditionif then\(findsconditionrelations with ‘if’ and ‘then’ anywhere\)

\\ex

\.conditionif \|\| then\(same, but ensures ‘if’ and ‘then’ are in arg1 and arg2\)

\\ex

\.conditionif \-\|\|\> then\(same, but ‘if’ must be in the relation source and ‘then’ in the target, regardless of text order\)

While in[3\.3](https://arxiv.org/html/2605.15304#S3.SS3)we only guarantee that ‘if’ and ‘then’ appear somewhere, in[3\.3](https://arxiv.org/html/2605.15304#S3.SS3)we require that they appear in that text order, one in each argument\. By contrast,[3\.3](https://arxiv.org/html/2605.15304#S3.SS3)requires that ‘if’ appears in the source of the relation \(the protasis\) and ‘then’ in the target \(the apodosis\), regardless of their text order\.

More experienced users who are familiar with UD annotations may also want to use token annotations to restrict queries\. To enable this we use the formatword\|lemma\|pos\|deprel, where each of these elements may be lacking\. If less than three annotations are specified, the system uses the search values to implicitly identify the key, since POS and deprel have closed vocabularies\. Thus the following searches find:

\\ex

\.purposeexactto\|PART \|VERB\|advcl \-\|\|\>\(purposerelation with a to\-infinitive\)

\\ex

\.temporalexactwhen \|ADJ\|advcl \-\|\|\>\(temporalwith ‘when’ followed by a reduced adjectival adverbial clause\)

The example in[3\.3](https://arxiv.org/html/2605.15304#S3.SS3)will find VERBs heading an adverbial clause \(UDadvcl\) immediately preceded by the word ‘to’ tagged as PART\. The interface automatically detects that VERB is a POS tag value andadvclis a dependency relation\. In[3\.3](https://arxiv.org/html/2605.15304#S3.SS3)we find reduced temporal clauses of the type ‘when possible’, since the word ‘when’ must be followed immediately by an adjective heading an adverbial clause\. The final operator\-\|\|\>ensures that both searches only consider the source span of the relation, regardless of text order\.

### 3\.4Frequencies interface

The frequencies tab gives raw counts, percentages and plots of a category or numerical variable selected by the user from the Breakdown drop down \(see Figure[2](https://arxiv.org/html/2605.15304#S3.F2)\)\. Categorical variables include DISRPT labels, original labels, relation direction, signal type/subtype and any available metadata \(for example genre, if known\)\. If filters are selected for any of these in the query, a binary yes/no breakdown of the selected feature is also available\. Updating the query instantly updates matches, numbers and plots, and raw results are also downloadable as a \.tsv file\.

![Refer to caption](https://arxiv.org/html/2605.15304v1/latex/freqs.png)Figure 2:Frequency breakdown of DISRPT labels\.A second drop down called ‘Cross\-tabulate’ allows users to select a second dimension from the same options and generate a cross table, coupled with a chi\-squared residual plot indicating combinations that appear more or less than expected, as well as displaying significance codes\. For example, Figure[3](https://arxiv.org/html/2605.15304#S3.F3)shows an association plot of explicit connective signals vs\. DISRPT label in the English PDTB corpus, showing that whileconcession,conditionandconjunctionare mostly explicit,causalrelations are more often implicit, whilecontrastrelations are more balanced\.

![Refer to caption](https://arxiv.org/html/2605.15304v1/latex/crosstab.png)Figure 3:Association of explicitness vs\. label in PDTB\.If a numerical variable is chosen for breakdown, the interface will plot a boxplot \(for a single variable\) or scatterplot \(two cross\-tabulated numerical variables\) or multiple boxplots \(numerical cross\-tabulated with categorical\)\. Available numerical variables are currently argument length in tokens and percentile position in document \(for argument 1 or 2 in text order\), the same for the source or target argument \(regardless of text order\), distance in tokens between arguments, and the number of signals for the relation \(if available\)\.

### 3\.5Comparison interface

Based on student feedback, comparing datasets is a desirable capability, and we implement this in a similar way to cross\-tabulation, where, instead of using a categorical variable, we use dataset identity\. However, since each dataset has its own distribution for each variable, we display results for each value side\-by\-side, with the primary selected dataset in blue and the comparison in orange with pairwise plots, as shown for a categorical variable \(label type\) with barplots in Figure[4](https://arxiv.org/html/2605.15304#S3.F4)for a comparison between the eRST GUM corpus and the eRST GENTLE corpus \(Genre Tests for Linguistic Evaluation,Aoyamaet al\.[2023](https://arxiv.org/html/2605.15304#bib.bib20)\), which follows the same annotation scheme but includes 8 challenging genres such as medical texts, poetry and even course syllabuses\.

![Refer to caption](https://arxiv.org/html/2605.15304v1/latex/comparison.png)Figure 4:Relation labels in GUM vs\. GENTLE\.The figure shows thatconjunctionis more common in GENTLE \(in orange\), which is primarily due to genres containing many lists, such as medical notes and syllabuses\. Theelaborationlabel, but contrast, is very similar in prevalence\.

As with frequencies, numerical variables receive side\-by\-side boxplots\. Figure[5](https://arxiv.org/html/2605.15304#S3.F5)shows the number of signals per relation, this time filtered to show justmoderelations \(manner and means\)\. These have significantly fewer signals in GENTLE, largely owing to data from the poetry and medical genres\.

![Refer to caption](https://arxiv.org/html/2605.15304v1/latex/comparison2.png)Figure 5:Signals permoderelation compared\.

## 4Evaluation

#### Data

We import the 38/39 DISRPT 2025 datasets which contain discourse relations \(the remaining dataset contains only discourse unit segmentation information, without labels\)\. The datasets come from five different frameworks: RST, PDTB, SDRT, eRSTZeldeset al\.\([2025](https://arxiv.org/html/2605.15304#bib.bib14)\)and discourse dependencies\. In total, these cover over 300,000 relations across 5 million tokens in almost 10,000 documents \(see Table[2](https://arxiv.org/html/2605.15304#A1.T2)in Appendix[A](https://arxiv.org/html/2605.15304#A1)for full details\)\. The largest dataset is English PDTBPrasadet al\.\([2014](https://arxiv.org/html/2605.15304#bib.bib3)\)with over 47K relations and 1\.1M tokens, as well as five signal types \(explicit or implicit connectives, alternative lexicalizations and constructions, and a special hypohphora type for questions\)\. Datasets in the eRST framework also distinguish signal subtypes, in a taxonomy of 8 major types and over 40 subtypes\.

#### Performance

While it is difficult to benchmark our system due to a lack of directly comparable alternatives, we conduct a simple timing experiment using the GUM data in comparison to its publicly available version in ANNIS\. We note that ANNIS can perform much more elaborate searches than DiscoExplorer, such as dependency graph queries between tokens \(e\.g\. checking that a token is the subject of a specific verb\), as well as querying other annotation layers, such as entity annotations; here we limit comparison to simple searches for tokens and discourse relations on a consumer laptop\. Since our data is loaded into main memory, query response times are close to instantaneous, with the only added latency of loading the dataset once, which ANNIS does not have \(Table[1](https://arxiv.org/html/2605.15304#S4.T1)\)\.

Table 1:Query latency compared with ANNIS\.

## 5Discussion and Conclusion

This short paper presented DiscoExplorer, a new browser based search interface for multilingual discourse relation datasets based on the DISRPT benchmark\. The interface offers a simple user\-friendly way to search for discourse relations and examine their distributions using filters, as well as a more complex query language to restrict matches by tokens and their annotations\. Quantitative results can be tabulated, plotted and downloaded\.

A comparison of query run times with ANNIS showed that although the interface requires an initial load time for each dataset, the approach using main memory search in JavaScript is very fast\. While some of the complex searches a system like ANNIS would allow are not supported, tailoring the interface to the DISRPT data model, centered around discourse relation instances, allows for a simple query language and data structure, and also means we do not need a backend or any compute resources to offer the system to the public\.

A further advantage of relying on the DISRPT data model is the abundance of data already available in the shared task format \(currently 38 datasets\), and the likely release of further public data in the format of the shared task, which has been running for four iterations as of 2026\.

As a result of submissions to the task, and especially the transition to multilingual models trained on all datasets, it is also increasingly possible to generate predicted datasets following the DISRPT label scheme in a variety of languages\. The current best performing system for predicting relation labels, DeDisCoJuet al\.\([2025](https://arxiv.org/html/2605.15304#bib.bib16)\), achieves 76\.13% accuracy on the DISRPT test set across languages, and 75\.55% on Chinese RSTPenget al\.\([2022](https://arxiv.org/html/2605.15304#bib.bib17)\), 79\.17% on PortugueseMendes and Lejeune \([2022](https://arxiv.org/html/2605.15304#bib.bib18)\), or 71\.39% on the PDTB framework Georgetown Discourse Treebank \(GDTB,Liuet al\.[2024](https://arxiv.org/html/2605.15304#bib.bib21)\), suggesting that automatically tagged corpora that are useful for research on discourse relations across languages may not be far and could easily be made searchable using this system\.

## Limitations

This paper only presents and evaluates a search interface in terms of responsiveness and does not conduct a user study, though we hope to gather some feedback on the system from participants at the CODI workshop and the community using DISRPT data\. The data used by the system is provided by DISRPT as\-is, and we make no claims regarding the accuracy of particular annotations within those datasets\. AI was not used in any way to write this paper, though AI coding assistants were used in the creation and debugging of the system itself\.

## References

- S\. Afantenos, N\. Asher, F\. Benamara, M\. Bras, C\. Fabre, M\. Ho\-dac, A\. L\. Draoulec, P\. Muller, M\. Péry\-Woodley, L\. Prévot, J\. Rebeyrolles, L\. Tanguy, M\. Vergez\-Couret, and L\. Vieu \(2012\)An empirical resource for discovering cognitive principles of discourse organisation: The ANNODIS corpus\.InProceedings of the Eighth International Conference on Language Resources and Evaluation \(LREC’12\),N\. Calzolari, K\. Choukri, T\. Declerck, M\. U\. Doğan, B\. Maegaard, J\. Mariani, A\. Moreno, J\. Odijk, and S\. Piperidis \(Eds\.\),Istanbul, Turkey,pp\. 2727–2734\.External Links:[Link](https://aclanthology.org/L12-1498/)Cited by:[12nd item](https://arxiv.org/html/2605.15304#A4.I1.i12.p1.1)\.
- M\. Amblard, B\. Guillaume, S\. Pavlova, and G\. Perrier \(2022\)Graph querying for semantic annotations\.InProceedings of the 18th Joint ACL \- ISO Workshop on Interoperable Semantic Annotation within LREC2022,H\. Bunt \(Ed\.\),Marseille, France,pp\. 95–101\.External Links:[Link](https://aclanthology.org/2022.isa-1.13/)Cited by:[§2](https://arxiv.org/html/2605.15304#S2.p1.1)\.
- T\. Aoyama, S\. Behzad, L\. Gessler, L\. Levine, J\. Lin, Y\. J\. Liu, S\. Peng, Y\. Zhu, and A\. Zeldes \(2023\)GENTLE: a genre\-diverse multilayer challenge set for English NLP and linguistic evaluation\.InProceedings of the 17th Linguistic Annotation Workshop \(LAW\-XVII\),J\. Prange and A\. Friedrich \(Eds\.\),Toronto, Canada,pp\. 166–178\.External Links:[Link](https://aclanthology.org/2023.law-1.17/),[Document](https://dx.doi.org/10.18653/v1/2023.law-1.17)Cited by:[§3\.5](https://arxiv.org/html/2605.15304#S3.SS5.p1.1)\.
- N\. Asher, J\. Hunter, M\. Morey, B\. Farah, and S\. Afantenos \(2016\)Discourse structure and dialogue acts in multiparty dialogue: the STAC corpus\.InProceedings of the Tenth International Conference on Language Resources and Evaluation \(LREC’16\),N\. Calzolari, K\. Choukri, T\. Declerck, S\. Goggi, M\. Grobelnik, B\. Maegaard, J\. Mariani, H\. Mazo, A\. Moreno, J\. Odijk, and S\. Piperidis \(Eds\.\),Portorož, Slovenia,pp\. 2721–2727\.External Links:[Link](https://aclanthology.org/L16-1432/)Cited by:[9th item](https://arxiv.org/html/2605.15304#A4.I1.i9.p1.1)\.
- N\. Asher and A\. Lascarides \(2003\)Logics of conversation\.Studies in Natural Language Processing,Cambridge University Press,Cambridge\.Cited by:[§1](https://arxiv.org/html/2605.15304#S1.p3.1)\.
- J\. Bauer, C\. Kiddon, E\. Yeh, A\. Shan, and C\. D\. Manning \(2023\)Semgrex and ssurgeon, searching and manipulating dependency graphs\.InProceedings of the 21st International Workshop on Treebanks and Linguistic Theories \(TLT, GURT/SyntaxFest 2023\),D\. Dakota, K\. Evang, S\. Kübler, and L\. Levin \(Eds\.\),Washington, D\.C\.,pp\. 67–73\.External Links:[Link](https://aclanthology.org/2023.tlt-1.7/)Cited by:[§2](https://arxiv.org/html/2605.15304#S2.p1.1)\.
- S\. Behzad and A\. Zeldes \(2020\)A cross\-genre ensemble approach to robust Reddit part of speech tagging\.InProceedings of the 12th Web as Corpus Workshop,A\. Barbaresi, F\. Bildhauer, R\. Schäfer, and E\. Stemle \(Eds\.\),Marseille, France,pp\. 50–56\(eng\)\.External Links:[Link](https://aclanthology.org/2020.wac-1.7/),ISBN 979\-10\-95546\-68\-9Cited by:[Appendix A](https://arxiv.org/html/2605.15304#A1.p1.1)\.
- C\. Braud, A\. Zeldes, C\. Li, Y\. J\. Liu, and P\. Muller \(2025\)The DISRPT 2025 shared task on elementary discourse unit segmentation, connective detection, and relation classification\.InProceedings of the 4th Shared Task on Discourse Relation Parsing and Treebanking \(DISRPT 2025\),C\. Braud, Y\. J\. Liu, P\. Muller, A\. Zeldes, and C\. Li \(Eds\.\),Suzhou, China,pp\. 1–20\.External Links:[Link](https://aclanthology.org/2025.disrpt-1.1/),[Document](https://dx.doi.org/10.18653/v1/2025.disrpt-1.1),ISBN 979\-8\-89176\-344\-9Cited by:[§1](https://arxiv.org/html/2605.15304#S1.p4.1)\.
- C\. Braud, A\. Zeldes, L\. Rivière, Y\. J\. Liu, P\. Muller, D\. Sileo, and T\. Aoyama \(2024\)DISRPT: a multilingual, multi\-domain, cross\-framework benchmark for discourse processing\.InProceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation \(LREC\-COLING 2024\),N\. Calzolari, M\. Kan, V\. Hoste, A\. Lenci, S\. Sakti, and N\. Xue \(Eds\.\),Torino, Italia,pp\. 4990–5005\.External Links:[Link](https://aclanthology.org/2024.lrec-main.447/)Cited by:[§1](https://arxiv.org/html/2605.15304#S1.p4.1)\.
- S\. Cao, I\. da Cunha, and M\. Iruskieta \(2018\)The RST Spanish\-Chinese treebank\.InProceedings of the Joint Workshop on Linguistic Annotation, Multiword Expressions and Constructions \(LAW\-MWE\-CxG\-2018\),A\. Savary, C\. Ramisch, J\. D\. Hwang, N\. Schneider, M\. Andresen, S\. Pradhan, and M\. R\. L\. Petruck \(Eds\.\),Santa Fe, New Mexico, USA,pp\. 156–166\.External Links:[Link](https://aclanthology.org/W18-4917/)Cited by:[21st item](https://arxiv.org/html/2605.15304#A4.I1.i21.p1.1)\.
- P\. C\. F\. Cardoso, E\. G\. Maziero, M\. L\. R\. C\. Jorge, E\. M\. R\. Seno, A\. D\. Felippo, L\. H\. M\. Rino, M\. das Graças V\. Nunes, and T\. A\. S\. Pardo \(2011\)CSTNews – a discourse\-annotated corpus for single and multi\-document summarization of news texts in brazilian portuguese\.InAnais do III Workshop “A RST e os Estudos do Texto”,Cuiabá, MT, Brasil,pp\. 88–105\.Cited by:[17th item](https://arxiv.org/html/2605.15304#A4.I1.i17.p1.1)\.
- L\. Carlson, D\. Marcu, and M\. E\. Okurowski \(2003\)Building a discourse\-tagged corpus in the framework of rhetorical structure theory\.InCurrent and New Directions in Discourse and Dialogue,Text, Speech and Language Technology 22,pp\. 85–112\.Cited by:[6th item](https://arxiv.org/html/2605.15304#A4.I1.i6.p1.1)\.
- I\. da Cunha, J\. Torres\-Moreno, G\. Sierra, L\. A\. Cabrera\-Diego, B\. G\. C\. Rolón, and J\. M\. R\. Bartilotti \(2011\)The RST spanish treebank on\-line interface\.InRecent Advances in Natural Language Processing, RANLP 2011, 12\-14 September, 2011, Hissar, Bulgaria,G\. Angelova, K\. Bontcheva, R\. Mitkov, and N\. Nicolov \(Eds\.\),pp\. 698–703\.External Links:[Link](https://aclanthology.org/R11-1101/)Cited by:[§2](https://arxiv.org/html/2605.15304#S2.p1.1)\.
- M\. de Marneffe, C\. D\. Manning, J\. Nivre, and D\. Zeman \(2021\)Universal Dependencies\.Computational Linguistics47\(2\),pp\. 255–308\.External Links:[Link](https://aclanthology.org/2021.cl-2.11/),[Document](https://dx.doi.org/10.1162/coli%5Fa%5F00402)Cited by:[§1](https://arxiv.org/html/2605.15304#S1.p4.1)\.
- L\. Gessler, Y\. Liu, and A\. Zeldes \(2019\)A discourse signal annotation system for RST trees\.InProceedings of the Workshop on Discourse Relation Parsing and Treebanking 2019,A\. Zeldes, D\. Das, E\. M\. Galani, J\. D\. Antonio, and M\. Iruskieta \(Eds\.\),Minneapolis, MN,pp\. 56–61\.External Links:[Link](https://aclanthology.org/W19-2708/),[Document](https://dx.doi.org/10.18653/v1/W19-2708)Cited by:[Figure 6](https://arxiv.org/html/2605.15304#A1.F6)\.
- G\. Guibon, M\. Courtin, K\. Gerdes, and B\. Guillaume \(2020\)When collaborative treebank curation meets graph grammars\.InProceedings of the Twelfth Language Resources and Evaluation Conference,N\. Calzolari, F\. Béchet, P\. Blache, K\. Choukri, C\. Cieri, T\. Declerck, S\. Goggi, H\. Isahara, B\. Maegaard, J\. Mariani, H\. Mazo, A\. Moreno, J\. Odijk, and S\. Piperidis \(Eds\.\),Marseille, France,pp\. 5291–5300\(eng\)\.External Links:[Link](https://aclanthology.org/2020.lrec-1.651/),ISBN 979\-10\-95546\-34\-4Cited by:[§2](https://arxiv.org/html/2605.15304#S2.p1.1),[§2](https://arxiv.org/html/2605.15304#S2.p2.1)\.
- M\. Iruskieta, M\. J\. Aranzabe, A\. D\. de Ilarraza, I\. Gonzalez\-Dios, M\. Lersundi, and O\. L\. de Lacalle \(2012\)The RST Basque TreeBank\.self\(basque\)\.External Links:[Link](http://ixa2.si.ehu.eus/diskurtsoa/en/)Cited by:[10th item](https://arxiv.org/html/2605.15304#A4.I1.i10.p1.1)\.
- M\. Iruskieta, M\. J\. Aranzabe, A\. Diaz de Ilarraza, I\. Gonzalez\-Dios, M\. Lersundi, and O\. Lopez de Lacalle \(2013\)The RST Basque TreeBank: An online search interface to check rhetorical relations\.InAnais do IV Workshop “A RST e os Estudos do Texto”,Fortaleza, CE, Brasil,pp\. 40–49\.Note:21–23 October 2013Cited by:[§2](https://arxiv.org/html/2605.15304#S2.p1.1)\.
- Z\. Ju, J\. Wu, A\. Purushothama, and A\. Zeldes \(2025\)DeDisCo at the DISRPT 2025 shared task: a system for discourse relation classification\.InProceedings of the 4th Shared Task on Discourse Relation Parsing and Treebanking \(DISRPT 2025\),C\. Braud, Y\. J\. Liu, P\. Muller, A\. Zeldes, and C\. Li \(Eds\.\),Suzhou, China,pp\. 48–62\.External Links:[Link](https://aclanthology.org/2025.disrpt-1.4/),[Document](https://dx.doi.org/10.18653/v1/2025.disrpt-1.4),ISBN 979\-8\-89176\-344\-9Cited by:[§5](https://arxiv.org/html/2605.15304#S5.p4.1)\.
- T\. Krause and A\. Zeldes \(2016\)ANNIS3: A new architecture for generic corpus query and visualization\.Digital Scholarship in the Humanities31\(1\),pp\. 118–139\.Cited by:[§2](https://arxiv.org/html/2605.15304#S2.p1.1)\.
- Y\. J\. Liu, T\. Aoyama, W\. Scivetti, Y\. Zhu, S\. Behzad, L\. E\. Levine, J\. Lin, D\. Tiwari, and A\. Zeldes \(2024\)GDTB: genre diverse data for English shallow discourse parsing across modalities, text types, and domains\.InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing,Y\. Al\-Onaizan, M\. Bansal, and Y\. Chen \(Eds\.\),Miami, Florida, USA,pp\. 12287–12303\.External Links:[Link](https://aclanthology.org/2024.emnlp-main.684/),[Document](https://dx.doi.org/10.18653/v1/2024.emnlp-main.684)Cited by:[§5](https://arxiv.org/html/2605.15304#S5.p4.1)\.
- W\. C\. Mann and S\. A\. Thompson \(1988\)Rhetorical Structure Theory: toward a functional theory of text organization\.Text8\(3\),pp\. 243–281\.Cited by:[§1](https://arxiv.org/html/2605.15304#S1.p3.1)\.
- A\. Mendes and P\. Lejeune \(2022\)CRPC\-DB a discourse bank for Portuguese\.InComputational Processing of the Portuguese Language: 15th International Conference, PROPOR 2022, Fortaleza, Brazil, March 21–23, 2022, Proceedings,Berlin, Heidelberg,pp\. 79–89\.External Links:ISBN 978\-3\-030\-98304\-8,[Link](https://doi.org/10.1007/978-3-030-98305-5_8),[Document](https://dx.doi.org/10.1007/978-3-030-98305-5%5F8)Cited by:[§5](https://arxiv.org/html/2605.15304#S5.p4.1)\.
- M\. Morey, P\. Muller, and N\. Asher \(2018\)A dependency perspective on RST discourse parsing and evaluation\.Computational Linguistics44\(2\),pp\. 197–235\.External Links:[Link](https://aclanthology.org/J18-2001),[Document](https://dx.doi.org/10.1162/COLI%5Fa%5F00314)Cited by:[§1](https://arxiv.org/html/2605.15304#S1.p3.1)\.
- N\. Nishida and Y\. Matsumoto \(2022\)Out\-of\-domain discourse dependency parsing via bootstrapping: an empirical analysis on its effectiveness and limitation\.Transactions of the Association for Computational Linguistics10,pp\. 127–144\.External Links:[Link](https://aclanthology.org/2022.tacl-1.8/),[Document](https://dx.doi.org/10.1162/tacl%5Fa%5F00451)Cited by:[2nd item](https://arxiv.org/html/2605.15304#A4.I1.i2.p1.1)\.
- M\. Ogrodniczuk, A\. Tomaszewska, D\. Ziembicki, S\. Żurowski, R\. Tuora, and A\. Zwierzchowska \(2024\)Polish discourse corpus \(PDC\): corpus design, ISO\-compliant annotation, data highlights, and parser development\.InProceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation \(LREC\-COLING 2024\),N\. Calzolari, M\. Kan, V\. Hoste, A\. Lenci, S\. Sakti, and N\. Xue \(Eds\.\),Torino, Italia,pp\. 12829–12835\.External Links:[Link](https://aclanthology.org/2024.lrec-main.1123/)Cited by:[16th item](https://arxiv.org/html/2605.15304#A4.I1.i16.p1.1)\.
- S\. Peng, Y\. J\. Liu, and A\. Zeldes \(2022\)GCDT: a Chinese RST treebank for multigenre and multilingual discourse parsing\.InProceedings of the 2nd Conference of the Asia\-Pacific Chapter of the Association for Computational Linguistics and the 12th International Joint Conference on Natural Language Processing \(Volume 2: Short Papers\),Y\. He, H\. Ji, S\. Li, Y\. Liu, and C\. Chang \(Eds\.\),Online only,pp\. 382–391\.External Links:[Link](https://aclanthology.org/2022.aacl-short.47/),[Document](https://dx.doi.org/10.18653/v1/2022.aacl-short.47)Cited by:[§5](https://arxiv.org/html/2605.15304#S5.p4.1)\.
- D\. Pisarevskaya, M\. Ananyeva, M\. Kobozeva, A\. Nasedkin, S\. Nikiforova, I\. Pavlova, and A\. Shelepov \(2017\)Towards building a discourse\-annotated corpus of russian\.InProceedings of the 23rd International Conference on Computational Linguistics and Intellectual Technologies “Dialogue\-2017”,Moscow, Russia\.Cited by:[18th item](https://arxiv.org/html/2605.15304#A4.I1.i18.p1.1)\.
- A\. Potter \(2008\)Interactional coherence in asynchronous learning networks: a rhetorical approach\.The Internet and Higher Education11\(2\),pp\. 87–97\.Cited by:[5th item](https://arxiv.org/html/2605.15304#A4.I1.i5.p1.1)\.
- R\. Prasad, B\. Webber, and A\. Joshi \(2014\)Reflections on the Penn Discourse TreeBank, comparable corpora, and complementary annotation\.Computational Linguistics40\(4\),pp\. 921–950\.External Links:[Link](https://aclanthology.org/J14-4007),[Document](https://dx.doi.org/10.1162/COLI%5Fa%5F00204)Cited by:[§1](https://arxiv.org/html/2605.15304#S1.p3.1),[§4](https://arxiv.org/html/2605.15304#S4.SS0.SSS0.Px1.p1.1)\.
- P\. Prasertsom, A\. Jaroonpol, and A\. T\. Rutherford \(2024\)The Thai Discourse Treebank: Annotating and classifying Thai discourse connectives\.Transactions of the Association for Computational Linguistics12,pp\. 613–629\.External Links:ISSN 2307\-387X,[Document](https://dx.doi.org/10.1162/tacl%5Fa%5F00650),[Link](https://doi.org/10.1162/tacl_a_00650),https://direct\.mit\.edu/tacl/article\-pdf/doi/10\.1162/tacl\_a\_00650/2369332/tacl\_a\_00650\.pdfCited by:[19th item](https://arxiv.org/html/2605.15304#A4.I1.i19.p1.1)\.
- G\. Redeker, I\. Berzlánovich, N\. van der Vliet, G\. Bouma, and M\. Egg \(2012\)Multi\-layer discourse annotation of a Dutch text corpus\.InProceedings of the Eighth International Conference on Language Resources and Evaluation \(LREC’12\),N\. Calzolari, K\. Choukri, M\. U\. Declerck, B\. Maegaard, J\. Mariani, A\. Moreno, J\. Odijk, and S\. Piperidis \(Eds\.\),Istanbul, Turkey,pp\. 2820–2825\.External Links:[Link](https://aclanthology.org/L12-1528/)Cited by:[14th item](https://arxiv.org/html/2605.15304#A4.I1.i14.p1.1)\.
- M\. C\. Scholman, M\. Marchal, A\. Brown, and V\. Demberg \(2025\)DiscoNaija: A discourse\-annotated parallel Nigerian Pidgin\-English corpus\.Language Resources and Evaluation,pp\. 1–37\.Cited by:[15th item](https://arxiv.org/html/2605.15304#A4.I1.i15.p1.1)\.
- S\. Shahmohammadi, H\. Veisi, and A\. Darzi \(2021\)Persian rhetorical structure theory\.External Links:[Document](https://dx.doi.org/10.48550/arXiv.2106.13833),2106\.13833,[Link](https://arxiv.org/abs/2106.13833)Cited by:[11st item](https://arxiv.org/html/2605.15304#A4.I1.i11.p1.1)\.
- M\. Stede and A\. Neumann \(2014\)Potsdam commentary corpus 2\.0: annotation for discourse research\.InProceedings of the Ninth International Conference on Language Resources and Evaluation \(LREC’14\),N\. Calzolari, K\. Choukri, T\. Declerck, H\. Loftsson, B\. Maegaard, J\. Mariani, A\. Moreno, J\. Odijk, and S\. Piperidis \(Eds\.\),Reykjavik, Iceland,pp\. 925–929\.External Links:[Link](https://aclanthology.org/L14-1468/)Cited by:[1st item](https://arxiv.org/html/2605.15304#A4.I1.i1.p1.1)\.
- F\. Tamburini \(2017\)Semgrex\-plus: a tool for automatic dependency\-graph rewriting\.InProceedings of the Fourth International Conference on Dependency Linguistics \(Depling 2017\),S\. Montemagni and J\. Nivre \(Eds\.\),Pisa, Italy,pp\. 248–254\.External Links:[Link](https://aclanthology.org/W17-6528/)Cited by:[§2](https://arxiv.org/html/2605.15304#S2.p1.1)\.
- K\. Thompson, J\. Hunter, and N\. Asher \(2024\)Discourse structure for the Minecraft corpus\.InProceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation \(LREC\-COLING 2024\),N\. Calzolari, M\. Kan, V\. Hoste, A\. Lenci, S\. Sakti, and N\. Xue \(Eds\.\),Torino, Italia,pp\. 4957–4967\.External Links:[Link](https://aclanthology.org/2024.lrec-main.444/)Cited by:[8th item](https://arxiv.org/html/2605.15304#A4.I1.i8.p1.1)\.
- S\. Tonelli, G\. Riccardi, R\. Prasad, A\. Joshi, E\. A\. Stepanov, and S\. A\. Chowdhury \(2010\)LUNA Corpus Discourse Data Set\.ELRA\(italian\)\.External Links:[Link](http://universal.elra.info/product%5C_info.php?-cPath=37%5C_38%5C&products%5C_id=1832)Cited by:[13rd item](https://arxiv.org/html/2605.15304#A4.I1.i13.p1.1)\.
- A\. Yang and S\. Li \(2018\)SciDTB: discourse dependency TreeBank for scientific abstracts\.InProceedings of the 56th Annual Meeting of the Association for Computational Linguistics \(Volume 2: Short Papers\),I\. Gurevych and Y\. Miyao \(Eds\.\),Melbourne, Australia,pp\. 444–449\.External Links:[Link](https://aclanthology.org/P18-2071/),[Document](https://dx.doi.org/10.18653/v1/P18-2071)Cited by:[3rd item](https://arxiv.org/html/2605.15304#A4.I1.i3.p1.1)\.
- C\. Yi, L\. Sujian, and L\. Yueyuan \(2021\)Unifying discourse resources with dependency framework\.InProceedings of the 20th Chinese National Conference on Computational Linguistics,S\. Li, M\. Sun, Y\. Liu, H\. Wu, K\. Liu, W\. Che, S\. He, and G\. Rao \(Eds\.\),Huhhot, China,pp\. 1058–1065\(eng\)\.External Links:[Link](https://aclanthology.org/2021.ccl-1.94/)Cited by:[23rd item](https://arxiv.org/html/2605.15304#A4.I1.i23.p1.1)\.
- K\. Zaczynska and M\. Stede \(2024\)Rhetorical strategies in the UN security council: Rhetorical Structure Theory and conflicts\.InProceedings of the 25th Annual Meeting of the Special Interest Group on Discourse and Dialogue,T\. Kawahara, V\. Demberg, S\. Ultes, K\. Inoue, S\. Mehri, D\. Howcroft, and K\. Komatani \(Eds\.\),Kyoto, Japan,pp\. 15–28\.External Links:[Link](https://aclanthology.org/2024.sigdial-1.2/),[Document](https://dx.doi.org/10.18653/v1/2024.sigdial-1.2)Cited by:[7th item](https://arxiv.org/html/2605.15304#A4.I1.i7.p1.1)\.
- A\. Zeldes, T\. Aoyama, Y\. J\. Liu, S\. Peng, D\. Das, and L\. Gessler \(2025\)ERST: a signaled graph theory of discourse relations and organization\.Computational Linguistics51\(1\),pp\. 23–72\.External Links:[Link](https://aclanthology.org/2025.cl-1.3/),[Document](https://dx.doi.org/10.1162/coli%5Fa%5F00538)Cited by:[§4](https://arxiv.org/html/2605.15304#S4.SS0.SSS0.Px1.p1.1)\.
- A\. Zeldes \(2017\)The GUM corpus: Creating multilayer resources in the classroom\.Language Resources and Evaluation51\(3\),pp\. 581–612\.Cited by:[§3\.2](https://arxiv.org/html/2605.15304#S3.SS2.p2.1)\.
- D\. Zeyrek and M\. Kurfalı \(2017\)TDB 1\.1: extensions on Turkish discourse bank\.InProceedings of the 11th Linguistic Annotation Workshop,N\. Schneider and N\. Xue \(Eds\.\),Valencia, Spain,pp\. 76–81\.External Links:[Link](https://aclanthology.org/W17-0809/),[Document](https://dx.doi.org/10.18653/v1/W17-0809)Cited by:[20th item](https://arxiv.org/html/2605.15304#A4.I1.i20.p1.1)\.
- D\. Zeyrek, A\. Mendes, Y\. Grishina, M\. Kurfalı, S\. Gibbon, and M\. Ogrodniczuk \(2020\)TED multilingual discourse bank \(ted\-mdb\): a parallel corpus annotated in the pdtb style\.Language Resources and Evaluation54\(2\),pp\. 587–613\.External Links:[Document](https://dx.doi.org/10.1007/s10579-019-09445-9),[Link](https://doi.org/10.1007/s10579-019-09445-9),ISSN 1574\-0218Cited by:[4th item](https://arxiv.org/html/2605.15304#A4.I1.i4.p1.1)\.
- Y\. Zhou, J\. Lu, J\. Zhang, and N\. Xue \(2014\)Chinese Discourse Treebank 0\.5\.LDC\(chinese\)\.External Links:[Link](https://catalog.ldc.upenn.edu/LDC2014T21)Cited by:[22nd item](https://arxiv.org/html/2605.15304#A4.I1.i22.p1.1)\.

## Appendix ADataset details

Table[2](https://arxiv.org/html/2605.15304#A1.T2)gives an overview of the datasets that are currently searchable using the system\. Datasets marked by an asterisk \(∗\\ast\) require LDC licenses; annotations for this data and a small subset ofeng\.erst\.gumcoming from RedditBehzad and Zeldes \([2020](https://arxiv.org/html/2605.15304#bib.bib19)\)can be obtained from the DISRPT shared task repository, along with scripts to reconstruct the underlying text\.

CorpusLanguageFrameworkLabelsRelationsSentencesTokensDocumentsSignalsces\.rst\.crdtCzechRST171,24983514,66454–deu\.pdtb\.pccGermanPDTB112,1092,19333,222176typesdeu\.rst\.pccGermanRST162,8821,94432,836176–eng\.dep\.covdtbEnglishdependencies114,9852,34360,907300–eng\.dep\.scidtbEnglishdependencies149,9034,202102,534798–eng\.erst\.gentleEnglisheRST172,5521,33417,97926subtypeseng\.erst\.gumEnglisheRST1730,74714,158254,890255subtypeseng\.pdtb\.gentleEnglishPDTB127861,33417,97926typeseng\.pdtb\.gumEnglishPDTB1313,87914,158254,890255types∗\\asteng\.pdtb\.pdtbEnglishPDTB1347,79248,6301,173,3792,162typeseng\.pdtb\.tedmEnglishPDTB135293818,1856typeseng\.rst\.ollEnglishRST172,7512,15646,471327–∗\\asteng\.rst\.rstdtEnglishRST1719,7788,318208,912385–eng\.rst\.stsEnglishRST173,0582,59171,206150–eng\.rst\.umucEnglishRST154,9972,42461,59087–eng\.sdrt\.msdcEnglishSDRT1027,84814,744231,352440–eng\.sdrt\.stacEnglishSDRT1112,2717,39452,2711,101–eus\.rst\.ertBasqueRST163,6322,38045,780164–fas\.rst\.prstcFarsiRST145,1912,17966,926150–fra\.sdrt\.annodisFrenchSDRT123,3211,50732,69986–ita\.pdtb\.lunaItalianPDTB111,5253,75025,24260typesnld\.rst\.nldtDutchRST162,2641,65124,89880–pcm\.pdtb\.disconaijaNaijaPDTB139,9039,242140,729176typespol\.iso\.pdcPolishISO128,5439,142156,980556typespor\.pdtb\.crpcPortuguesePDTB1211,3275,194186,849302typespor\.pdtb\.tedmPortuguesePDTB135543948,1906typespor\.rst\.cstnPortugueseRST154,9932,22163,332140–rus\.rst\.rrtRussianRST1525,09513,131262,495234–spa\.rst\.rststbSpanishRST163,0492,08958,717267–spa\.rst\.sctbSpanishRST1669251616,51550–tha\.pdtb\.tdtbThaiPDTB1210,8616,534256,523180–∗\\asttur\.pdtb\.tdbTurkishPDTB133,17631,197496,358197–tur\.pdtb\.tedmTurkishPDTB135744106,2866typeszho\.dep\.scidtbMandarindependencies141,29750018,761109–zho\.pdtb\.cdtbMandarinPDTB95,2702,89173,314164–zho\.pdtb\.tedMandarinPDTB1513,3088,671181,91072typeszho\.rst\.gcdtMandarinRST178,4132,69262,90550–zho\.rst\.sctbMandarinRST1769258015,49650–Total16617311,796257,7055,139,5649,89014 datasets

Table 2:DISRPT 2025 datasets searchable in DiscoExplorer \(∗\\astmarks datasets requiring an LDC license\)\.Additional datasets can be added to the system as long as they conform to the DISRPT shared task format, meaning that relations are serialized in the\.relsformat and token annotations are available in a corresponding\.conllufile\. In particular, the DISRPT format assumes that relations apply between flat spans of text, meaning that hierarchical information as found in formalisms such as RST is lost\. Instead, relations are interpreted as a dependency conversion of constituent structures, as illustrated in Figure[6](https://arxiv.org/html/2605.15304#A1.F6)\.

The figure shows the system representation of a single relation, in this case aconcessionbetween two head units:

\\ex

\. \[this is a terrific opportunity\]\.¯\\b\{\.\}\[butwe will have to wait until after the event\]Theseunitsareonlypartsofthesentencestheycomefrom,asshownintheeRSTgraphfragment\.ModifiersofthoseunitswhichappearinthesamesentencesarerepresentedinDiscoExploreraspre−,inter−andpost−context,dependingonwhethertheyappearbeforethefirstargument,betweenthetwoarguments,orafterthesecond\.NotealsothatwhiletheeRSTgraphonthelefthighlightsmultiplesignals,onlythereddiscoursemarker‘‘but′′isattachedtotheconcessionrelation,andthatsamesignalisvisualizedintheDiscoExplorersearchresults\.Thecyanhighlightedsyntacticsignalsforpurposerelations\(‘opportunity\.\.toimprove′and‘wait\.\.toassess′\)belongtothoserespectiverelations,andwouldbehighlightedinqueriesactuallyretrievingtheassociatedrelation,ratherthanbeinghighlightedinaqueryretrievingadifferentrelationthathappenstooverlapthesametext\.Figure 6Figure 66Figure 66Original eRST graph fragment for a concession relation, visualized using rstWeb Gessler et al\. \(2019\) and the corresponding output in DiscoExplorer\.Figure 6Original eRST graph fragment for a concession relation, visualized using rstWeb Gessler et al\. \(2019\) and the corresponding output in DiscoExplorer\.\\par Theseunitsareonlypartsofthesentencestheycomefrom,asshownintheeRSTgraphfragment\.ModifiersofthoseunitswhichappearinthesamesentencesarerepresentedinDiscoExploreraspre\-,inter\-andpost\-context,dependingonwhethertheyappearbeforethefirstargument,betweenthetwoarguments,orafterthesecond\.\\par NotealsothatwhiletheeRSTgraphonthelefthighlightsmultiplesignals,onlythereddiscoursemarker\`\`but^\{\\prime\\prime\}isattachedtothe\\textsc\{concession\}relation,andthatsamesignalisvisualizedintheDiscoExplorersearchresults\.Thecyanhighlightedsyntacticsignalsfor\\textsc\{purpose\}relations\(\`opportunity\.\.toimprove^\{\\prime\}and\`wait\.\.toassess^\{\\prime\}\)belongtothoserespectiverelations,andwouldbehighlightedinqueriesactuallyretrievingtheassociatedrelation,ratherthanbeinghighlightedinaqueryretrievingadifferentrelationthathappenstooverlapthesametext\.\\par\\par\\par\\par\\par\\begin\{figure\*\}\[tbh\]\\centering\\includegraphics\[width=345\.0pt,frame\]\{latex/data\-conversion\.png\} \\@@toccaption\{\{\\lx@tag\[ \]\{\{6\}\}\{Original eRST graph fragment for a \{concession\} relation, visualized using rstWeb \\cite\[cite\]\{\\@@bibref\{Authors Phrase1YearPhrase2\}\{gessler\-etal\-2019\-discourse\}\{\\@@citephrase\{\(\}\}\{\\@@citephrase\{\)\}\}\} and the corresponding output in DiscoExplorer\.\}\}\}\\@@caption\{\{\\lx@tag\[: \]\{\{Figure 6\}\}\{Original eRST graph fragment for a \{concession\} relation, visualized using rstWeb \\cite\[cite\]\{\\@@bibref\{Authors Phrase1YearPhrase2\}\{gessler\-etal\-2019\-discourse\}\{\\@@citephrase\{\(\}\}\{\\@@citephrase\{\)\}\}\} and the corresponding output in DiscoExplorer\.\}\}\} \\@add@centering\\end\{figure\*\}\\par

## Appendix BAdditional use cases

![Refer to caption](https://arxiv.org/html/2605.15304v1/latex/when.png)Figure 7:Frequencies oftemporal‘when’ clauses in left\-to\-right vs\. right\-to\-left directionsIn addition to exploring relation labels, DiscoExplorer can be used to study the distribution of relation signals \(in datasets with signal annotations, indicated in the dataset list by a lightning bolt icon in Figure[1](https://arxiv.org/html/2605.15304#S3.F1)\), relation directions and signal subtypes\. Figure[8\(a\)](https://arxiv.org/html/2605.15304#A2.F8.sf1)shows a breakdown of relation

Figure[7](https://arxiv.org/html/2605.15304#A2.F7)shows the disparity of right\-to\-left versus left\-to\-right temporal relations signaled by ‘when’ in English, where source of the relation tends to precede the target, but not always – the proportion is about 7:3 in favor of placing thetemporalclause first\.

![Refer to caption](https://arxiv.org/html/2605.15304v1/latex/sigtypes.png)\(a\)Residual plot for DISRPT labels versus major signal types\.
![Refer to caption](https://arxiv.org/html/2605.15304v1/latex/dms.png)\(b\)Residual plot for discourse markers appearing 20\+ times versusexplanationrelation labels

Figure 8:Contingency tables for signal types and subtypes ineng\.erst\.gum\.Figure[8](https://arxiv.org/html/2605.15304#A2.F8)shows two panels demonstrating the breakdown of signaling devices\. Panel[8\(a\)](https://arxiv.org/html/2605.15304#A2.F8.sf1)shows major signal types signaling three relations classes:alternationrelations \(such as ‘A or B’\) are typically signaled by a discourse marker \(dm\) such as ‘\(either\) or’, ‘\(or\) else’, ‘alternatively’ etc\.;attributionrelations, indicating the source of information, are typically signaled semantically by the presence of a speaker noun phrase, or lexically through a speech verb, but very rarely using a dm \(for example ‘AsSmith had it, …’\) or with no signals \(implicitly\)\. By contrastcausalrelations usually appear with a dm, and appear with no explicit signaling more often than the other relations \(column ‘None’\)\.

Finally, Panel[8\(b\)](https://arxiv.org/html/2605.15304#A2.F8.sf2)shows discourse marker subtypes used to signal three original labels forexplanationrelations in the data, using a cutoff to show only items appearing 20 or more times\. Although the DISRPT relation labels do not include subtypes, we can get breakdowns for relation subtypes if the original labels of the underlying dataset include them, which is the case here\. The biggest disparity is the preference ofexplanation\-evidencerelations to be marked by ‘for example’ and ‘as’\. By contrast,explanation\-justifyfavors the use of ‘and’, as in[B](https://arxiv.org/html/2605.15304#A2)\.

\\ex

\.\[\[The record is replete with case law that says exactly that,\]1\]\_\{1\}\[\[andI’m not here to dispute that today\.\]2\]\_\{2\}

Meanwhileexplanation\-motivationrelations appear disproportionally often with ‘so’, attempting to convince someone to do something using a supporting argument, as in[B](https://arxiv.org/html/2605.15304#A2)\.

\\ex

\.\[\[Good jokes have a lot of details and personality,\]1\]\_\{1\}\[\[sodon’t be afraid to embellish\.\]2\]\_\{2\}

However using the interface, it is easy to find examples of ‘so’ as a discourse marker with any of the three labels\.

## Appendix CResource consumption

While the interface runs very quickly and requires no server\-side compute resources to run, an anonymous reviewer has inquired about the memory footprint of loading a large dataset\. This is difficult to quantify exactly, since we cannot trivially access browser memory management internals, but to get a rough benchmark, we ran Chrome on a Windows 11 64 bit machine and compared memory usage for the browser with different datasets loaded, as shown in Table[3](https://arxiv.org/html/2605.15304#A3.T3)\.

Table 3:Memory usage in several scenarios using Chrome\.The dataset ‘loaded’ state by default means that a query runs to retrieve all relations, since no filter has been selected\. As the table shows, loading a fairly large and richly annotated corpus such aseng\.erst\.gumrequires about 400 MB of RAM\. The largest dataset in tokens and relations,eng\.pdtb\.pdtb, takes only slightly more, at 444 MB – we suspect the small difference is due to the amount of space taken up by the signal annotations in the former dataset, which the latter lacks\. Using the comparison function on the two datasets does not add much more memory \(∼\\sim465 MB total\)\. We suspect that this is because a new full result set is not actually loaded into memory \- only the information being compared, by default the relation label statistics, is added to the main memory, while indexing specific matching tokens or sentence spans is unnecessary for the comparison data, since no detailed search results are shown for the second dataset\.

## Appendix DCorpus resources

The system described in this paper would be useless without the datasets it makes searchable\. In addition to the datasets and papers cited above, we would like to acknowledge the projects that have produced the remaining datasets in Table[2](https://arxiv.org/html/2605.15304#A1.T2), all of which can be searched in our publicly available instance of DiscoExplorer and are available under their original licenses from the DISRPT shared task:

- •deu\.rst\.pccanddeu\.pdtb\.pcc– the Potsdam Commentary Corpus,Stede and Neumann \([2014](https://arxiv.org/html/2605.15304#bib.bib24)\)
- •eng\.dep\.covdtb– the COVID\-29 Discourse Treebank,Nishida and Matsumoto \([2022](https://arxiv.org/html/2605.15304#bib.bib25)\)
- •eng\.dep\.scidtb– SciDTB,Yang and Li \([2018](https://arxiv.org/html/2605.15304#bib.bib26)\)
- •eng\.pdtb\.tedm,por\.pdtb\.tedmandtur\.pdtb\.tedm– TED Multilingual Discourse Bank,Zeyreket al\.\([2020](https://arxiv.org/html/2605.15304#bib.bib37)\)
- •end\.rst\.ollandend\.rst\.sts– RST Online\-Learning and Science, Technology, and Society corpora,Potter \([2008](https://arxiv.org/html/2605.15304#bib.bib42)\)
- •end\.rst\.rstdt– RST Discourse Treebank,Carlsonet al\.\([2003](https://arxiv.org/html/2605.15304#bib.bib27)\)
- •end\.rst\.umuc– University of Potsdam Multilayer UNSC Corpus,Zaczynska and Stede \([2024](https://arxiv.org/html/2605.15304#bib.bib43)\)
- •eng\.sdrt\.msdc– Minecraft Structured Dialogue Corpus,Thompsonet al\.\([2024](https://arxiv.org/html/2605.15304#bib.bib44)\)
- •eng\.sdrt\.stac– Strategic Conversation Corpus,Asheret al\.\([2016](https://arxiv.org/html/2605.15304#bib.bib28)\)
- •eus\.rst\.ert– Basque RST Treebank,Iruskietaet al\.\([2012](https://arxiv.org/html/2605.15304#bib.bib29)\)
- •fas\.rst\.prstc– Persian RST Corpus,Shahmohammadiet al\.\([2021](https://arxiv.org/html/2605.15304#bib.bib30)\)
- •fra\.sdrt\.annodis– the ANNODIS corpus,Afantenoset al\.\([2012](https://arxiv.org/html/2605.15304#bib.bib31)\)
- •ita\.pdtb\.luna– LUNA corpus,Tonelliet al\.\([2010](https://arxiv.org/html/2605.15304#bib.bib38)\)
- •nld\.rst\.nldt– Dutch DTB,Redekeret al\.\([2012](https://arxiv.org/html/2605.15304#bib.bib32)\)
- •pcm\.pdtb\.disconaija– DiscoNaija corpus,Scholmanet al\.\([2025](https://arxiv.org/html/2605.15304#bib.bib45)\)
- •pol\.iso\.pdc– Polish Discourse Corpus,Ogrodniczuket al\.\([2024](https://arxiv.org/html/2605.15304#bib.bib46)\)
- •por\.rst\.cstn– CST News Corpus,Cardosoet al\.\([2011](https://arxiv.org/html/2605.15304#bib.bib33)\)
- •rus\.rst\.rrt– Russian RST Treebank,Pisarevskayaet al\.\([2017](https://arxiv.org/html/2605.15304#bib.bib34)\)
- •tha\.pdtb\.tdtb– Thai Discourse Treebank,Prasertsomet al\.\([2024](https://arxiv.org/html/2605.15304#bib.bib47)\)
- •tur\.pdtb\.tdb– Turkish Discourse Bank,Zeyrek and Kurfalı \([2017](https://arxiv.org/html/2605.15304#bib.bib39)\)
- •spa\.rst\.sctbandzho\.rst\.sctb– the RST Spanish\-Chinese Treebank,Caoet al\.\([2018](https://arxiv.org/html/2605.15304#bib.bib35)\)
- •zho\.pdtb\.cdtb– Chinese Discourse Treebank,Zhouet al\.\([2014](https://arxiv.org/html/2605.15304#bib.bib40)\)
- •zho\.dep\.scidtb– Chinese SciDTB,Yiet al\.\([2021](https://arxiv.org/html/2605.15304#bib.bib36)\)