CHILDES is a widely used resource of transcribed child and child-directed speech. This paper introduces UD-English-CHILDES, the first officially released Universal Dependencies (UD) treebank. It is derived from previously dependency-annotated CHILDES data, which we harmonize to follow unified annotation principles. The gold-standard trees encompass utterances sampled from 11 children and their caregivers, totaling over 48K sentences (236K tokens). We validate these gold-standard annotations under the UD v2 framework and provide an additional 1M silver-standard sentences, offering a consistent resource for computational and linguistic research.
In this paper, we discuss the challenges that we faced during the construction of a Universal Dependencies treebank for Abaza, a polysynthetic Northwest Caucasian language. We propose an alternative to the morpheme-level annotation of polysynthetic languages introduced in Park et al. (2021). Our approach aims at reducing the number of morphological features, yet providing all the necessary information for the comprehensive representation of all the syntactic relations. Besides, we suggest to add one language-specific relation needed for annotating repetitions in spoken texts and present several solutions that aim at increasing cross-linguistic comparability of our data.
This paper discusses the need for including morphological features in Japanese Universal Dependencies (UD). In the current version (v2.11) of the Japanese UD treebanks, sentences are tokenized at the morpheme level, and almost no morphological feature annotation is used. However, Japanese is not an isolating language that lacks morphological inflection but is an agglutinative language. Given this situation, we introduce a tentative scheme for retokenization and morphological feature annotation for Japanese UD. Then, we measure and compare the morphological complexity of Japanese with other languages to demonstrate that the proposed tokenizations show similarities to synthetic languages reflecting the linguistic typology.
This paper reports the development of the first dependency treebank for the Sinhala language (STB). Sinhala, which is morphologically rich, is a low-resource language with few linguistic and computational resources available publicly. This treebank consists of 100 sentences taken from a large contemporary written text corpus. These sentences were annotated manually according to the Universal Dependencies framework. In this paper, apart from elaborating on the approach that has been followed to create the treebank, we have also discussed some interesting syntactic constructions found in the corpus and how we have handled them using the current Universal Dependencies specification.
This paper presents the harmonisation process carried out on the five treebanks available for Latin in Universal Dependencies, with the aim of eliminating the discrepancies in their annotation styles. Indeed, this is the first issue to be addressed when parsing Latin, as significant drops in parsing accuracy on different Latin treebanks have been repeatedly observed. Latin syntactic variability surely accounts for this, but parsing results are as well affected by divergent annotation choices. By analysing where annotations differ, we propose a Python-based alignment of the five UD treebanks. Consequently, the impact of annotation choices on accuracy scores is assessed by performing parsing experiments with UDPipe and Stanza.
This article presents a comparative analysis of four different syntactic typological approaches applied to 20 different languages. We compared three specific quantitative methods, using parallel CoNLL-U corpora, to the classification obtained via syntactic features provided by a typological database (lang2vec). First, we analyzed the Marsagram linear approach which consists of extracting the frequency word-order patterns regarding the position of components inside syntactic nodes. The second approach considers the relative position of heads and dependents, and the third is based simply on the relative position of verbs and objects. From the results, it was possible to observe that each method provides different language clusters which can be compared to the classic genealogical classification (the lang2vec and the head and dependent methods being the closest). As different word-order phenomena are considered in these specific typological strategies, each one provides a different angle of analysis to be applied according to the precise needs of the researchers.
Pomak is an endangered oral Slavic language of Thrace/Greece. We present a short description of its interesting morphological and syntactic features in the UD framework. Because the morphological annotation of the treebank takes advantage of existing resources, it requires a different methodological approach from the one adopted for syntactic annotation that has started from scratch. It also requires the option of obtaining morphological predictions/evaluation separately from the syntactic ones with state-of-the-art NLP tools. Active annotation is applied in various settings in order to identify the best model that would facilitate the ongoing syntactic annotation.
Recent efforts to consolidate guidelines and treebanks in the Universal Dependencies project raise the expectation that joint training and dataset comparison is increasingly possible for high-resource languages such as English, which have multiple corpora. Focusing on the two largest UD English treebanks, we examine progress in data consolidation and answer several questions: Are UD English treebanks becoming more internally consistent? Are they becoming more like each other and to what extent? Is joint training a good idea, and if so, since which UD version? Our results indicate that while consolidation has made progress, joint models may still suffer from inconsistencies, which hamper their ability to leverage a larger pool of training data.
Technical documents present distinct challenges when used in natural language processing tasks such as part-of-speech tagging or syntactic parsing. This is mainly due to the nature of their content, which may differ greatly from more studied texts like news articles, encyclopedic extracts or social media entries. This work contributes an English corpus composed of software requirement texts annotated in Universal Dependencies (UD) to study the differences, challenges and issues encountered on these documents when following the UD guidelines. Different structural and linguistic phenomena are studied in the light of their impact on manual and automatic dependency annotation. To better cope with texts of this nature, some modifications and features are proposed in order to enrich the existing UD guidelines to better cover technical texts. The proposed corpus is compared to other existing corpora to show the structural complexity of the texts as well as the challenge it presents to recent processing methods. This contribution is the first software requirement corpus annotated with UD relations.
This study discusses the way different numerals and related expressions are currently annotated in the Universal Dependencies project, with a specific focus on the Uralic language family and only occasional references to the other language groups. We analyse different annotation conventions between individual treebanks, and aim to highlight some areas where further development work and systematization could prove beneficial. At the same time, the Universal Dependencies project already offers a wide range of conventions to mark nuanced variation in numerals and counting expressions, and the harmonization of conventions between different languages could be the next step to take. The discussion here makes specific reference to Universal Dependencies version 2.8, and some differences found may already have been harmonized in version 2.9. Regardless of whether this takes place or not, we believe that the study still forms an important documentation of this period in the project.
UDWiki is an online environment designed to make creating new UD treebanks easier. It helps in setting up all the necessary data needed for a new treebank up in a GUI, where the interface takes care of guiding you through all the descriptive files needed, adding new texts to your corpus, and helping in annotating the texts. The system is built on top of the TEITOK corpus environment, using an XML based version of UD annotation, where dependencies can be combined with various other types of annotations. UDWiki can run all the necessary or helpful scripts (taggers, parsers, validators) via the interface. It also makes treebanks under development directly searchable, and can be used to maintain or search existing UD treebanks.
We describe ud2rrg, a rule-based approach for converting UD trees to Role and Reference Grammar (RRG) structures. Our conversion method aims at facilitating the annotation of multilingual RRG treebanks. ud2rrg uses general and language-specific conversion rules. In order to evaluate ud2rrg, we approximate the subsequent annotation effort via measures of tree edit distance. Our evaluation, based on English, German, French, Russian, and Farsi, shows that the ud2rrg transformation of UD-parsed data constitutes a highly useful starting point for multilingual RRG treebanking. Once a sufficient amount of data has been annotated in this way, the automatic conversion can be replaced by a statistical parser trained on that data for an even better starting point.
We attempt to shed some light on the various ways how languages specify date and time, and on the options we have when trying to annotate them uniformly across Universal Dependencies. Examples from several language families are discussed, and their annotation is proposed. Our hope is to eventually make this (or similar) proposal an integral part of the UD annotation guidelines, which would help improve consistency of the UD treebanks. The current annotations are far from consistent, as can be seen from the survey we provide in appendices to this paper.
Nonfinite verb forms are a crosslinguistically widespread phenomenon that poses a challenge to universal annotation formalisms like Universal Dependencies (UD), often clashing with traditionally established, language-specific conventions and terminologies. This paper, using Latin as a concrete case study, aims to give a survey on the VerbForm feature distribution among UD treebanks and to suggest a restructuring thereof in a universal perspective.
We describe the design and use of a web-based system for helping the validation of English or French Universal Dependencies corpora by sentence regeneration. A symbolic approach is used to transform the dependency tree into a constituency tree which is then regenerated as a sentence in the original language. The comparison between regenerated sentences and the original ones from version 2.8 of Universal Dependencies revealed some annotation errors which are discussed and give rise to suggestions for improvement.
We introduce the first treebank for Old Turkic script Old Turkish texts, consisting of 23 sentences from Orkhon corpus and transliterated texts such as poems, annotated according to the Universal Dependencies (UD) guidelines with universal part-of-speech tags and syntactic dependencies. Then, we propose a text processing pipeline for the script that makes the texts easier to encode, input and tokenize. Finally, we present our approach to tokenization and annotation from a crosslingual perspective by inspecting linguistic constructions compared to other languages.
This paper presents a first attempt to apply Universal Dependencies (Nivre et al., 2016; de Marneffe et al., 2021) to Bribri, an Indigenous language from Costa Rica belonging to the Chibchan family. There is limited previous work on Bribri NLP, so we also present a proposal for a dependency parser, as well as a listing of structures that were challenging to parse (e.g. flexible word order, verbal sequences, arguments of intransitive verbs and mismatches between the tense systems of Bribri and UD). We also list some of the challenges in performing NLP with an extremely low-resource Indigenous language, including issues with tokenization, data normalization and the training of tools like POS taggers which are necessary for the parsing. In total we collected 150 sentences (760 words) from publicly available sources like grammar books and corpora. We then used a context-free grammar for the initial parse, and then applied the headfloating algorithm in Xia and Palmer (2001) to automatically generate dependency parses. This work is a first step towards building a UD treebank for Bribri, and we hope to use this tool to improve the documentation of the language and develop language-learning materials and NLP tools like chatbots and question answering-systems.
Ligurian is a minority Romance language spoken in the homonymous region of Northern Italy and the Principality of Monaco, amongst others. In this paper we present the first Universal Dependencies treebank for Ligurian, consisting of 316 sentences and 6 928 tokens, extracted from a wide variety of sources to reflect variation in syntax and register. Along with the corpus, we contribute a short analysis of the varieties and spelling systems of Ligurian, as well as a set of recommendations and annotation guidelines for certain constructions with non-trivial analyses. We hope that these will serve as a foundation for further research, to encourage the development of NLP technologies for a language that has so far been under-served.
On the basis of four small scale studies on corpora of English, German and Modern Greek, this paper points out problems with the lack of annotation guidelines for adnominal pronoun constructions like we linguists in treebanks employing the Universal Dependencies framework. I propose that a more uniform strategy of annotating these constructions will improve the internal consistency of corpora and better facilitate crosslinguistic comparability. Specifically, I argue against the use of the APPOS(ition) relation for these constructions and in favour of employing the DET(erminer) relation as a default annotation strategy.
