Proxecto Nós is an initiative aimed at providing the Galician language with openly licensed resources, tools, and demonstrators in the area of intelligent technologies. The Project has two main scientific and technological objectives: (i) to integrate the Galician language into cuttingedge AI and language technologies, thus enabling the natural use of Galician in human-machine interactions; and (ii) to improve the state of the art of language technologies for Galician. Language technologies, linguistic rights, Galician, low-resource languages.
Proxecto Nós (The Nós Project) is an initiative promoted by the Galician Government (Xunta de
aimed at providing the
language with openly licensed resources, tools, demonstrators, and use cases in the area of intelligent technologies.
The execution of Proxecto Nós has been entrusted to the University being carried out by a research team comprising members of the Instituto da Lingua Galega (ILG) and the Centro Singular de Investigación en
The first stage, spanning from the final trimester of 2021 to 2025, will lay the foundations and provide the resources that will help place Galician among the languages that are fully active in the digital society and economy.
The development of language technologies is a strategic innovation area geared towards the digital society and economy, and it has been a priority in both Spanish (Plan Estatal de Investigación Científica y Técnica y de Innovación, Estrategia Española de Ciencia y Tecnología y de Innovación) and European (Horizon 2020) scientific planning. Technologies such as machine translation (MT), information extraction (IE), text analytics, and dialogue systems are essential in the digital society, culture, and economy.
Languages in high demand worldwide (especially English) benefit from a large variety of computational resources that can contribute to developing new automatic language processing technologies and tools. Such is the case due to the long-standing research tradition in these areas (e.g., the variety of projects financed by USA’s DARPA) and the need to incorporate such languages into the AI applications associated with the latest electronic devices (such as the conversational AI or automatic dictation software developed by Google, Amazon or Apple). Other languages that have joined AI research later, such as Chinese, are currently following in the footsteps of English, through projects such as Baidu’s Qian Yan, which improve significantly the computational resources available in their respective language varieties.
Notwithstanding, language technologies are also necessary for languages in lower international demand. Consequently, different languages have developed similar initiatives to Nós. Among others, we can highlight Projecte AINA, which will develop computational resources for Catalan until 2024, or the work carried out at the HiTZ Research Center, focusing on languages technologies for Basque. Other projects, such as CorCenCC (in Great Britain, for Welsh) or UQAILAUT (in Canada, for Inuktitut) were considered success cases in the promotion of the digital use of socially threatened languages.
The democratization of language technologies has a great social and cultural impact on the communities that use them. For instance, MT increases access to contents in different languages, thus facilitating intercultural relations; dialogue systems allow us to communicate with machines in our own language; and semantic technologies enable advances in the automatic comprehension of texts, thus making it possible to process enormous quantities of documents. In the case of Galician, incorporating the language into state-of-the-art AI applications can not only significantly favor its prestige (a decisive factor in language normalization), but also guarantee citizens’ language rights and reduce social inequality.
In economic terms, the global Natural
Language Processing (NLP) market size was
valued at more than USD 10 billion in 2020 and
is expected to reach USD 41 billion by 2025
In 2012, the White Paper The Galician
Language in the Digital Age
Furthermore, Galician is currently part of
multilingual crowdsourced data collection
initiatives carried out by important companies on
the global IT market, which have resulted in
speech databases such as Google’s SLR77
Despite the quality of these resources, it should
be noted that not all are freely and publicly
available for the development of LT. The LT field
has undergone profound changes over the last few
years since the introduction of neural network
systems. Generally, training models using these
state-of-the-art technologies requires large
quantities of data and has high energetic and
computational costs, which continues to be a
challenge for low-resource languages. However,
as many recent studies show, end-to-end
technologies and open-source multilingual
pretrained models created using large quantities of
data from high-resource languages
The tasks that are to be carried out as part of the Project can be included in the following areas, corresponding to some of the major NLP fields: An example of numbered list is as following. 1. Speech synthesis (TTS) 2. Speech recognition (ASR) 3. Automatic text generation 4. Dialogue systems 5. MT 6. IE 7. Opinion mining and fact checking 8. Language correction and assessment These broad, mutually interdependent areas fall within the three strategic lines jointly identified by the Project’s research team and the Xunta de Galicia (in particular, with the Axencia para a Modernización Tecnolóxica de Galicia): (i) spoken or written conversation with people, (ii) language quality, and (iii) information management.
In accordance with the funding agreement signed by the Xunta de Galicia and the USC, the organization of the tasks included in Nós follows a yearly schedule. Each year, resources, language models and demonstrators from different areas will be made publicly available.
More information on the organization of Proxecto Nós can be found in de-Dios-Flores et al, 2022.
Proxecto Nós has two main scientific and technological objectives: (i) to integrate the Galician language into cutting-edge AI and language technologies, thus enabling the natural use of Galician in human-machine interactions; and (ii) to improve the state of the art of language technologies for Galician.
For this purpose, resources, tools, and applications will be developed and distributed under open licenses, which will allow them to be integrated into existing devices and services (such as smart speakers or conversational agents) and future technologies. To this end, specific objectives directly related to some of the major tasks of NLP have been established.
Each of these technological objectives will be executed in a different subproject, which will allow the parallel development of different tasks and, overall, a more effective organization of the work. However, a set of general objectives are shared by all the tasks. These objectives are: (i) the compilation of high-quality linguistic resources (annotated reference corpora, web-scale corpora, specialized corpora by tasks and domains, parallel corpora, knowledge bases, dictionaries, etc.); (ii) the elaboration of language and acoustic models (both general-purpose and task-specific models); and (iii) the development of applications based on these models. The project will also have a general coordination mechanism through which resources will be distributed and shared among its subprojects.
The resources and language models developed for each task will be made available to the public, thus allowing their use in all kinds of applications, services, and products, by the scientific community, companies, institutions, and society in general. The results will be disseminated through a repository available at the project’s web portal (which can be hosted on internal servers), as well as other established and internationally recognized repositories, such as HuggingFace, GitHub, Zenodo, etc.
Finally, the project contemplates the complete development of applications based on these resources, which will act as visible and accessible demonstrators of the developed technology and will produce a tractor effect that will lead to the development of new products.
Among the initial results of Nós, we can
highlight the first crawl of a web-based Galician
corpus and a language model based on the CCNet
tools and data
For the current year, Proxecto Nós aims to keep generating linguistic and computational resources to explore different subprojects. Specifically, in the first half of 2022 work will be carried out on the design of a high-quality speech corpus of sufficient size so as to allow training TTS state-of-the-art models, to be released in the last trimester. The second half of the year will also see the publication of a speech corpus for ASR. In the same timeframe, the project will publish several text corpora: parallel Galician-Spanish, Galician-English, and Galician-Portuguese corpora; a web-scale Galician text corpus, larger than the one already compiled, to be used in all the subprojects working with written text included in Nós; and a domain-specific corpus for automatic text generation. Based on these resources, new language models will be developed using different state-of-the-art techniques, as well as demonstrators or prototypes of a TTS system, NMT system, and automatic text generator for Galician. At the same time, throughout 2022 efforts will focus on extending and improving the first systems developed, and on validating the results obtained via the creation of high-quality gold standards.
This research was funded by the project “Nós: Galician in the society and economy of artificial intelligence” (Proxecto Nós: O galego na sociedade e economía da intelixencia artificial 2021-CP080), agreement between Xunta de Galicia and University of Santiago de Compostela, and grant ED431G2019/04 by the Galician Ministry of Education, University and Professional Training, and the European Regional Development Fund (ERDF/FEDER program). J. Dunne (eds.), Report on the Galician Language (Deliverable D1.15), ELE, 2022. [12] J. Shen, R. Pang, R. J. Weiss, M. Schuster, N. Jaitly, Z. Yang, Z. Chen, Y. Zhang, , Y. Wang, R. J. Skerry-Ryan, R. A. Saurous, Y. Agiomyrgiannakis, Y. Wu, Natural TTS Synthesis By Conditioning Wavenet On Mel Spectrogram Predictions, in: Proceedings of ICASSP, 2018. [13] T. Wolf, L. Debut, V. Sanh, J. Chaumond, C.
Delangue, A. Moi, P. Cistac, T. Rault, R. Louf, M. Funtowicz, J. Davison, S. Shleifer, P. von Platen, C. Ma, Y. Jernite, J. Plu, C. Xu, T. Le Scao, S. Gugger, et al., Transformers: State-of-the-Art Natural Language Processing. In: Proceedings of the 2020 Conference on Empirical Methods in NLP: System Demonstrations, 2020, pp. 38– 45.