Search

This paper describes the first steps in creating a Lemma Bank for Old Irish (600-900CE) within the Linked Data paradigm, taking inspiration from a similar resource for Latin built as part of the LiLa project (2018{--}2023). The focus is on the extraction and RDF conversion of nouns from Goidelex, a novel and highly structured morphological resource for Old Irish. The aim is to strike a good balance between retaining a representative level of morphological granularity and at the same time keeping the amount of lemma variants within workable limits, to facilitate straightforward resource interlinking for Old Irish, planned as future work.

When it comes to the quantitative evaluation of word embeddings, there are two main strategies: extrinsic, i.e. using pre-trained embeddings as input vectors in a downstream ML task, such as language modelling, and intrinsic, i.e. through analogy and similarity tasks that require special datasets (Bakarov, 2018). Extrinsic evaluation Language modelling seems to be the easiest way to evaluate historical word embeddings, since it is language independent, scalable and does not require dataset creation. Hypothetically, using pre-trained embeddings must lower the perplexity of a language model, even if these embeddings were trained on a different period of the same language. However, language modelling, as well as the majority of modern NLP tasks, is not very relevant to historical linguistics, so we might want to find a better downstream task or turn to intrinsic evaluation. Intrinsic evaluation There are two major tasks used for intrinsic evaluation of word embeddings: similarity and analogy. The similarity task consists in comparing similarity scores of two words yielded by an embedding model to those calculated based on experts’ judgment. We did not explore this option, because it requires too much manual work by definition. The analogy task is simply asking an embedding model “What is to a′ as b is to b′ ?”, and expecting a as an answer. Analogy datasets can be created automatically or semi- automatically if there exists a comprehensive historical dictionary of a language in question in machine readable format or a WordNet. Traditionally, analogy datasets are based on pairwise semantic proportion and therefore every question has a single correct answer. Given the high level of variation in historical languages, such a strict definition of a correct answer seems unjustified. Therefore, in our Early Irish analogy dataset we follow the authors of BATS (Gladkova et al., 2016) providing several correct answers for each analogy question and evaluating the performance with

This paper describes the creation of a workbench tool designed to make technologies developed throughout the lifespan of the Cardamom project easily accessible to researchers who could most benefit from them, but who may not have the technical expertise to apply bleeding edge technologies to their own datasets. The workbench provides an intuitive graphical user interface (GUI) and workflow which abstract users away from underlying technical tasks, while providing them with a suite of powerful NLP tools developed by the Cardamom team. These include tokenisers, POS-taggers, various annotation tools, and ML models. The performance of workbench tools can be improved as text and annotations are added by users. It is envisioned that this workbench will provide a simple route to digital publication for academics in the humanities, or more specifically, for linguists working with under-resourced or historical languages, who have collected text data but are unable to make it available online as a result of financial or technical restraints. This has the added benefit of increasing the availability of high quality, annotated text data to NLP researchers, thereby providing value to both communities of researchers.

This paper describes the structure and findings of the SIGTYP 2023 shared task on cognate and derivative detection for low-resourced languages, broken down into a supervised and unsupervised sub-task. The participants were asked to submit the test data's final prediction. A total of nine teams registered for the shared task where seven teams registered for both sub-tasks. Only two participants ended up submitting system descriptions, with only one submitting systems for both sub-tasks. While all systems show a rather promising performance, all could be within the baseline score for the supervised sub-task. However, the system submitted for the unsupervised sub-task outperforms the baseline score.

Exploiting cognates for transfer learning in under-resourced languages is an exciting opportunity for language understanding tasks, including unsupervised machine translation, named entity recognition and information retrieval. Previous approaches mainly focused on supervised cognate detection tasks based on orthographic, phonetic or state-of-the-art contextual language models, which under-perform for most under-resourced languages. This paper proposes a novel language-agnostic weakly-supervised deep cognate detection framework for under-resourced languages using morphological knowledge from closely related languages. We train an encoder to gain morphological knowledge of a language and transfer the knowledge to perform unsupervised and weakly-supervised cognate detection tasks with and without the pivot language for the closely-related languages. While unsupervised, it overcomes the need for hand-crafted annotation of cognates. We performed experiments on different published cognate detection datasets across language families and observed not only significant improvement over the state-of-the-art but also our method outperformed the state-of-the-art supervised and unsupervised methods. Our model can be extended to a wide range of languages from any language family as it overcomes the requirement of the annotation of the cognate pairs for training.

While Old Irish (c. 600–900 A.D.) is extensively documented, it remains digitally under- resourced, lacking the range of digital resources available for other older Indo-European languages (e.g., Latin, see Pellegrini and Passarotti, 2018). We report on the development of a fully inflected lexicon of Old Irish nouns, provided in both phonemic and orthographic notation. This involved a computer-assisted, systematic, and reproducible grapheme-to- phoneme conversion pipeline and generating morphological forms through a finite-state transducer. The inflected lexicon we develop will better enable computational studies in Old Irish morphology, further research into diachronic developments, and have a wide range of Natural Language Processing (NLP) applications. We began by extracting noun lemmata from the Old Irish Würzburg glosses (Kavanagh, 2001) and the Corpus PalaeoHibernicum (CorPH) ‘Old Irish Corpus’ (Stifter et al., 2021). We then devised a set of rules for orthography-to-phonology conversion, subsequently implemented using the Python package Epitran (Mortensen, Dalmia, and Littell, 2018). The resulting transcriptions act as the data input for a finite-state transducer (FST) adapted from Fransen (2019), allowing us to generate inflected forms of Old Irish nouns. Finally, we derived orthographic forms (and their variants) by applying conversion rules to the generated forms. Old Irish presents considerable challenges for the development of a resource of this nature, given its opaque and inconsistent orthography, complex phonology, elaborate system of morphophonological alternations, and intricate patterns of morphological inflection (Anderson, 2016; Stifter, 2009; Thurneysen, 1946; Pedersen, 1909–1913). We report on how we dealt with these problems in the development of the inflectional lexicon. While this study focused on the Old Irish nouns in the Würzburg glosses, we intend to extend the lexicon by applying this pipeline to further corpora and other parts- of-sp

The digitisation of historical texts has provided new horizons for NLP research, but such data also presents a set of challenges, including scarcity and inconsistency. The lack of editorial standard during digitisation exacerbates these difficulties.This study explores the potential for temporal domain adaptation in Early Modern Irish and pre-reform Modern Irish data. We describe two experiments carried out on the book subcorpus of the Historical Irish Corpus, which includes Early Modern Irish and pre-reform Modern Irish texts from 1581 to 1926. We also propose a simple orthographic normalisation method for historical Irish that reduces the type-token ratio by 21.43{\%} on average in our data.The results demonstrate that the use of out-of-domain data significantly improves a language model{'}s performance. Providing a model with additional input from another historical stage of the language improves its quality by 12.49{\%} on average on non-normalised texts and by 27.02{\%} on average on normalised (demutated) texts. Most notably, using only out-of-domain data for both pre-training and training stages allowed for up to 86.81{\%} of the baseline model quality on non-normalised texts and up to 95.68{\%} on normalised texts without any target domain data. Additionally, we investigate the effect of temporal distance between the training and test data. The hypothesis that there is a positive correlation between performance and temporal proximity of training and test data has been validated, which manifests best in normalised data. Expanding this approach even further back, to Middle and Old Irish, and testing it on other languages is a further research direction.

In this paper, we describe how we unearthed some fundamental problems while building an analogy dataset modelled on BATS (Gladkova et al., 2016) to evaluate historical Irish embeddings on their ability to detect orthographic, morphological and semantic similarity.The performance of our models in the analogy task was extremely poor regardless of the architecture, hyperparameters and evaluation metrics, while the qualitative evaluation revealed positive tendencies. We argue that low agreement between field experts on fundamental lexical and orthographic issues, and the lack of a unified editorial standard in available resources make it impossible to build reliable evaluation datasets for computational models and obtain interpretable results. We emphasise the need for such a standard, particularly for NLP applications, and prompt Celticists and historical linguists to engage in further discussion. We would also like to draw NLP scholars{'} attention to the role of data and its (extra)linguistic properties in testing new models, technologies and evaluation scenarios.

We present the findings of the LoResMT 2021 shared task which focuses on machine translation (MT) of COVID-19 data for both low-resource spoken and sign languages. The organization of this task was conducted as part of the fourth workshop on technologies for machine translation of low resource languages (LoResMT). Parallel corpora is presented and publicly available which includes the following directions: English↔Irish, English↔Marathi, and Taiwanese Sign language↔Traditional Chinese. Training data consists of 8112, 20933 and 128608 segments, respectively. There are additional monolingual data sets for Marathi and English that consist of 21901 segments. The results presented here are based on entries from a total of eight teams. Three teams submitted systems for English↔Irish while five teams submitted systems for English↔Marathi. Unfortunately, there were no systems submissions for the Taiwanese Sign language↔Traditional Chinese task. Maximum system performance was computed using BLEU and follow as 36.0 for English–Irish, 34.6 for Irish–English, 24.2 for English–Marathi, and 31.3 for Marathi–English.

In “Automatic morphological analysis and interlinking of historical Irish cognate verb forms”, Theodorus Fransen describes a computational approach to understanding how the Irish verbal system develops diachronically. The author’s major contribution is to propose a morphological analyser for Old Irish verbs and to discuss ways this analyser can be incorporated into a framework of computational resources for various stages of Irish. This proposal dovetails with Jøhndal’s and Meelen’s chapters in dealing with ways of expanding the current computational toolset for a historical language (specifically historical stages of Irish) and in its concerns with scalability. These concerns are reflected in his detailed investigation of the challenges encountered by a methodology that incorporates finite-state morphology as it applies to Old Irish. The challenges he details are twofold. The first challenge has to do with word and morpheme division as encountered in “real” text, i.e. editions or manuscript transcriptions. In many cases, multiple morphemes may be written as a concatenated string, resulting in the need to find a way to encode licit combinatorial possibilities of multiple morphemes. This is a so-called generation problem, where generation means the ability of the analyser to generate all and only the licit inflected forms of any given stem. In other cases, whitespace is found between morphemes leading to potential parsing ambiguities since the analyser is word-based (where a word is understood to be an element between whitespace). This is a so-called analysis problem, which may result in the wrong morphological tag being assigned to any given string. The second challenge has to do with the complex interaction between phonology (especially stress) and morphology in Old Irish since stress alternations can result in syncope and the presence or absence of palatalisation of stem-final and ending-initial consonants. These challenges impinge on the choices made for implemen

The topic of this paper constitutes the main part of a recently finished Ph.D. project carried out by the author which investigates how computational methods can be employed to map cognate verb forms in Early Irish (ca. 7th–12th centuries A.D.) and Modern Irish (ca. 1200 onwards). This paper discusses the development of a finite-state morphological transducer using foma (Hulden, 2009) for the Old Irish language (ca. 7th–9th centuries A.D.), focusing on verbs. Two main challenges are discussed. First, different practices of word segmentation have significant repercussions for the encoding of dependencies both on and beyond the word level. A second challenge is complex verb stem formation and considerable stem allomorphy. This has been tackled by operating with “monolithic stem” entries for each verb lemma, i.e., synchronic, invariable hard-coded stems, representing a semi-surface-level base form.

This paper describes an inflectional lexicon of Old Irish nouns, and the tools developed for its cre- ation. While Old Irish (c. 600–900 A.D.) is extensively documented, it remains digitally under-resourced. We develop a morphological description in the form of a fully inflected lexicon of Old Irish nouns, pro- vided in both phonemic and orthographic notation. This entailed devising a computer-assisted, systematic, and reproducible grapheme-to-phoneme conversion pipeline and generating morphological forms through a finite-state transducer. We report on the considerable challenges posed by Old Irish in terms of its mor- phophonological complexities and its intransparent and inconsistent orthography. The inflected lexicon we develop will better enable computational studies in Old Irish morphology, further research into diachronic developments, and have a wide range of Natural Language Processing (NLP) applications. Despite the fact that “Old Irish is the earliest period of Irish –– or of any Celtic language –– for which the extant record is sufficiently full and varied to permit a full synchronic description” (Stifter, 2009, p. 59), the language still lacks the range of digital resources available for other Indo-European languages (e.g., Latin, see Pellegrini and Passarotti, 2018). While there are a number of independent projects focusing on Old Irish lexicography (Griffith, Stifter, and Toner, 2018), the most comprehensive resource, both in terms of contemporary source material included and the level of grammatical annotation, is Corpus PalaeoHiber- nicum (CorPH) ‘Old Irish Corpus’ (Stifter et al., 2021). However, in spite of the richness of the linguistic annotation in CorPH, it cannot be used as the basis for a morphological generator without considerable pre- processing, due to its inconsistent orthography for lemmata and the way it segments complex morphological structures. Old Irish presents many challenges for the development of computational resources. The lan

Word senses are the fundamental unit of description in lexicography, yet it is rarely the case that different dictionaries reach any agreement on the number and definition of senses in a language. With the recent rise in natural language processing and other computational approaches there is an increasing demand for quantitatively validated sense catalogues of words, yet no consensus methodology exists. In this paper, we look at four main approaches to making sense distinctions: formal, cognitive, distributional, and intercultural and examine the strengths and weaknesses of each approach. We then consider how these may be combined into a single sound methodology. We illustrate this by examining two English words, “wing†and “fish,†using existing resources for each of these four approaches and illustrate the weaknesses of each. We then look at the impact of such an integrated method and provide some future perspectives on the research that is necessary to reach a principled method for making sense distinctions.

This paper gives an overview of the Cardamom project, which aims to close the resource gap for minority and under-resourced languages by means of deep-learning-based natural language processing (NLP) and exploiting similarities of closely-related languages. The project further extends this idea to historical languages, which can be considered as closely related to their modern form, and as such aims to provide NLP through both space and time for languages that have been ignored by current approaches.

Multilingual sentence embeddings capture rich semantic information not only for measuring similarity between texts but also for catering to a broad range of downstream cross-lingual NLP tasks. State-of-the-art multilingual sentence embedding models require large parallel corpora to learn efficiently, which confines the scope of these models. In this paper, we propose a novel sentence embedding framework based on an unsupervised loss function for generating effective multilingual sentence embeddings, eliminating the need for parallel corpora. We capture semantic similarity and relatedness between sentences using a multi-task loss function for training a dual encoder model mapping different languages onto the same vector space. We demonstrate the efficacy of an unsupervised as well as a weakly supervised variant of our framework on STS, BUCC and Tatoeba benchmark tasks. The proposed unsupervised sentence embedding framework outperforms even supervised state-of-the-art methods for certain under-resourced languages on the Tatoeba dataset and on a monolingual benchmark. Further, we show enhanced zero-shot learning capabilities for more than 30 languages, with the model being trained on only 13 languages. Our model can be extended to a wide range of languages from any language family, as it overcomes the requirement of parallel corpora for training.

Dr. Theodorus Fransen & Dr. John McCrae explore how digital language tools can potentially resolve the underrepresentation of minority languages in terms of digital technology and the Web.

We continue this month’s theme, Languages, with a piece in which Theodorus Fransen introduces us to the basics of historical Irish linguistics and Natural Language Processing. Theodorus is a PhD student in the School of Linguistic, Speech and Communication Sciences, Trinity College Dublin and is currently an Early Career Research Fellow in the Trinity Long Room Hub Arts & Humanities Research Institute.

Automatic Language Identification (LI) or Dialect Identification (DI) of short texts of closely related languages or dialects, is one of the primary steps in many natural language processing pipelines. Language identification is considered a solved task in many cases; however, in the case of very closely related languages, or in an unsupervised scenario (where the languages are not known in advance), performance is still poor. In this paper, we propose the Unsupervised Deep Language and Dialect Identification (UDLDI)method, whichcan simultaneously learn sentence embeddings and cluster assignments from short texts. The UDLDI modelunderstands the sentence constructions of languages by applying attention to character relations which helps to optimize the clustering of languages. We have performed our experiments on three short-text datasets for different language families, each consisting of closely related languages or dialects, with very minimal training sets. Our experimental evaluations on these datasets have shown significant improvement over state-of-the-artunsupervised methods and our model has outperformed state-of-the-art LI and DI systems in supervised settings.

Aligning senses across resources and languages is a challenging task with beneficial applications in the field of natural language processing and electronic lexicography. In this paper, we describe our efforts in manually aligning monolingual dictionaries. The alignment is carried out at sense-level for various resources in 15 languages. Moreover, senses are annotated with possible semantic relationships such as broadness, narrowness, relatedness, and equivalence. In comparison to previous datasets for this task, this dataset covers a wide range of languages and resources and focuses on the more challenging task of linking general-purpose language. We believe that our data will pave the way for further advances in alignment and evaluation of word senses by creating new solutions, particularly those notoriously requiring data such as neural networks. Our resources are publicly available at https://github.com/elexis-eu/MWSA., Aligning senses across resources and languages is a challenging task with beneficial applications in the field of natural language processing and electronic lexicography. In this paper, we describe our efforts in manually aligning monolingual dictionaries. The alignment is carried out at sense-level for various resources in 15 languages. Moreover, senses are annotated with possible semantic relationships such as broadness, narrowness, relatedness, and equivalence. In comparison to previous datasets for this task, this dataset covers a wide range of languages and resources and focuses on the more challenging task of linking general-purpose language. We believe that our data will pave the way for further advances in alignment and evaluation of word senses by creating new solutions, particularly those notoriously requiring data such as neural networks. Our resources are publicly available at https://github.com/elexis-eu/MWSA.

Language and mathematics meet in the dynamic field of computational linguistics. Using the latest deep learning techniques and transferring results from better resourced languages, Theodorus Fransen of the Cardamom project in Galway hopes to develop an algorithm capable of analysing the various forms of the Irish verb across time and space, adding to the linguistic tools available to both students and scholars of historical Irish.

The historical periods of the Irish language can be generally classified as follows: Old Irish (c.600–900), Middle Irish (c.900–1200), Early Modern Irish (c.1200–1650) and Modern Irish (c.1650– present). The last ten to fifteen years have seen the rise in the number and variety of digital resources supporting the study of historical Irish texts, e.g. CELT [1] and RIA Online Archive [2]. In this paper, we will provide an inventory of the resources available. Special attention will be paid to two lexicographical projects: the electronic Dictionary of the Irish Language [3] and Foclóir na Nua-Ghaeilge [4]. The electronic Dictionary of the Irish Language (eDIL), covering the period 600–1650, is available online and was most recently revised in 2013. Foclóir na Nua-Ghaeilge (FNG), a project of the Royal Irish Academy, aims at producing a historical dictionary of Irish covering the period 1600–2000. Currently texts for the period 1600 to 1926 are being lemmatized. In this paper we will also report on a study currently being undertaken which looks at the possibilities of interconnecting eDIL and FNG, focussing initially on verbs. Challenges from a computational point of view, as well as benefits for historical textual and linguistic research, will be discussed. [1] www.ucc.ie/celt/ [2] http://research.dho.ie/fng/index.php [3] http://edil.qub.ac.uk/about.php [4] www.ria.ie/research/focloir-na-nua-ghaeilge.aspx