TL;DR
This paper explores the critical challenges and innovative approaches in developing assistive technologies for Irish (Gaelic), an endangered language. It highlights how the unique linguistic features of Irish and the socio-linguistic context necessitate a community-driven, language-sensitive approach to ensure inclusivity for users with disabilities.
What Problem Does This Paper Address?
The paper addresses the significant gap in assistive technology (AT) development for minority and endangered languages, specifically focusing on Irish (Gaelic). While speech and language technologies are rapidly advancing, their benefits are not equitably distributed across languages. This creates a "digital timebomb" where speakers of endangered languages, particularly those with disabilities, are excluded from digital participation, further accelerating language loss. The authors emphasize that for languages like Irish, AT development cannot be driven by commercial interests but must prioritize the urgent needs of the community, especially individuals with speech and communication disabilities who face a "minority within a minority" situation.
Methods
The paper outlines three key areas of AT development for Irish, each with unique methodological considerations:
- Screenreading Facility for Visually Impaired People: This involved developing a plugin for the NVDA screenreader. The approach focused on creating an interface that leverages existing ABAIR synthetic voices and integrates with Braille translation software (Liblouis). User control over speed and visual highlighting for partial vision were key features.
- Phonological Awareness and Early Literacy for Dyslexic People: This area required a deeper dive into linguistic research. The method involved developing an interactive platform, "Lón don Leon," which uses specially designed stories, graphics, and games to teach phonemic awareness and phonic rules. The platform integrates prerecorded content with ABAIR's synthetic voices to provide a rich learning experience, addressing the opaque mapping between Irish orthography and its sound system, which is particularly challenging for dyslexic learners.
- Speech-Enabled AAC System for Non-Speaking People: This involved adapting an open-source system (Coughdrop) for Irish. The core challenge here was addressing the complex morphology, syntax, and semantics of Irish, which differ significantly from English. The methodology incorporates a multi-stage grammar correction process, including a morphological corrector, an open-source grammar checking engine ("An Gramadóir"), and a genitive corrector, to ensure grammatically correct sentence generation from user input.
Datasets
Not explicitly detailed in the paper, but the development of the ABAIR initiative's resources (speech synthesis and recognition systems) forms the foundational dataset for these AT applications. The paper mentions corpus collection focusing on native speakers from different Gaeltacht areas for speech recognition.
Key Insights and Ideas
- The "Digital Timebomb" for Endangered Languages: The rapid advancement of speech and language technologies, without parallel development for minority languages, poses a significant threat to their survival.
- Community-Driven Development is Paramount: For endangered languages, AT development must be guided by the specific needs and priorities of the community, not commercial interests.
- Language Sensitivity is Crucial: The unique linguistic structures of Irish (e.g., dialectal variation, complex morphology, opaque orthography) pose significant challenges that require deep linguistic knowledge to overcome.
- User-Centric Design is Non-Negotiable: Users and their support networks must be active partners throughout the entire development lifecycle, from design to dissemination.
- Bilingualism Adds Complexity: The interaction between Irish and English, including code-switching, requires AT systems to be designed with this bilingual context in mind.
- Addressing the "Minority within a Minority": Individuals with disabilities who are also speakers of an endangered language face a double layer of exclusion, making AT development for them particularly critical.
Key Results
- Screenreading Facility: A functional NVDA plugin has been developed and tested with visually impaired school children, allowing them to access written Irish and participate in the digital world.
- Phonological Awareness Platform: The "Lón don Leon" platform has been developed to address the lack of resources for children with dyslexia learning Irish, aiming to improve phonemic awareness and early literacy skills.
- Speech-Enabled AAC System: A prototype Irish AAC system has been developed, incorporating sophisticated grammar correction mechanisms to handle the complexities of the Irish language.
Limitations and Future Work
- Dialectal Diversity: While current synthetic voices cater to three main dialects, further development is needed to include more dialects and specifically children's voices.
- ASR for Non-Native Speakers and Children: The current ASR system is optimized for native speakers; extending it to non-native speakers and children is a future research direction.
- Assessment Materials: The "Lón don Leon" platform is currently focused on training and intervention, with future work planned to incorporate assessment materials for dyslexia.
- Optimizing Motor Plans for Bilingual Users: Further research and testing are needed to determine the optimal layout of symbols in the AAC system to cater to both Irish and English motor plans.
- Interdisciplinary Researcher Shortage: Finding skilled researchers with expertise in linguistics, speech technology, and AT development remains a significant challenge.
- Cost-Free and Accessible Outputs: A guiding principle is to ensure all developed AT outputs are cost-free and readily available on the ABAIR website.
Summary updated