The African Storybook (ASb) is a multilingual literacy initiative that works with educators and children to publish openly licensed picture storybooks for early reading in the languages of Africa. An initiative of Saide, the ASb has an interactive website that enables users to read, create, download, translate, and adapt storybooks. The initiative addresses the dire shortage of children’s storybooks in African languages, crucial for children’s literacy development. As of September 2020 website had nearly 6,700 storybooks in 222 languages spoken in Africa, consisting of nearly 2000 original titles and translations of those titles.

We created a novel dataset, ANTC — African News Topic Classification for 4 African languages. We obtained data from three different news sources: VOA, BBC6 and isolezwe7 . From the VOA data we created datasets for Lingala and Somali. We obtained the topics from data released by Palen-Michel et al. (2022) and used the provided urls to get the news category from the websites. For pidgin and isiZulu, we scrapped news topic from the respective news website (BBC Pidgin and isolezwe respectively) directly base on their category. We noticed that some news topics are not mutually exclusive to their categories, therefore, we filtered such topics with multiple labels. Also, we ensured that each category has at least 200 samples. The categories include but not limited to, Africa, Entertainment, Health, and Politics. The pre-processed datasets were divided into training, development, and test sets using stratified sampling with a ratio of 70:10:20. Appendix A.2 has more details about the dataset size and news topic information.

The story of Linguarena
I am Samba Kamara, founder of Linguarena.

In 2009, before a trip to Dakar (Senegal), I decided to learn wolof language.

It was not my first trip to Senegal but this time I wanted to be able to talk with locals in wolof.



I was used to learning languages on my IPhone 3GS at the time, so naturally I launched the App Store app to look for a wolof app or translator.

I found one wolof app, it was poorly designed, buggy and abandoned by its creator.

Back from my trip in Dakar, I told myself that someone have to solve this problem, it should possible to learn wolof and any other language with our SmartPhone.



I went to meet some publishing companies in Paris, which had hundreds of books in their catalogues to learn languages including wolof, swahili, lingala, soninke, etc…., Ali Baba’s cave of languages, in short.



I suggested to them that they should have all their language catalogue on App Store and Google Play Store because there was a real need and the future of language learning will be on electronic device (mobile, tablet and computer) and not on paper books.

They were very excited about the project but did not follow up.



That’s when the idea of Linguarena started to sprout in my head, what if I made my own Ali Baba cave of languages?

I decided! I will make my own language collections, I would call it the « language arena » in latin (lingua: language , arena: arena), hence Linguarena.



Ten years later, Linguarena has 5 languages, 300,000 application downloads and more to come, with my team we are working to make as many languages available on Smartphone as possible.



I work with language teachers, native speakers for audio voices, designers, graphic designers and illustrators located in the four corners of the world.



If your preferred language is not yet available at Linguarena, write to us via the contact page, or leave us comments on our social networks, our next language will be the one that has been most requested.

Palatalization of coronals and stridents is well-known and widespread, and is most commonly associated with front vowels or glides as triggers. In some dialect(s) of Setswana, a much different type of palatalization occurs: alveolar stridents /s ts tsʰ/ become pre-palatal [ʃ tʃ tʃʰ] before back vowels and the glide [w]. Clear empirical support for this pattern comes from productive alternations induced by the nominalizing suffix /-ɔ/, as well as alternations with an assortment of less productive morphemes, and lexical evidence. If palatalization before front vocoids is phonetically natural, then palatalization before back vocoids seems like it must be phonetically unnatural. However, this paper suggests that it is not the case: palatalization before back vowels actually makes phonetic sense, as a consequence of using lip rounding as a phonetic enhancement of the S~Š distinction.

“Application of Ensemble methods for solving imbalanced data problems”

Poster presented at the Deep Learning Indaba 2022 by Henock Makumbu Mboko

As language and speech technologies become more advanced, the lack of fundamental digital resources for African languages, such as data, spell checkers and PoS taggers, means that the digital divide between these languages and others keeps growing. This work details the organisation of the AI4D - African Language Dataset Challenge, an effort to incentivize the creation, curation and uncovering to African language datasets through a competitive challenge, particularly datasets that are annotated or prepared for use in a downstream NLP task.

Adapter modules have emerged as a general parameter-efficient means to specialize a pretrained encoder to new domains. Massively multilingual transformers (MMTs) have particularly benefited from additional training of language-specific adapters. However, this approach is not viable for the vast majority of languages, due to limitations in their corpus size or compute budgets. In this work, we propose MAD-G (Multilingual ADapter Generation), which contextually generates language adapters from language representations based on typological features. In contrast to prior work, our time- and space-efficient MAD-G approach enables (1) sharing of linguistic knowledge across languages and (2) zero-shot inference by generating language adapters for unseen languages. We thoroughly evaluate MAD-G in zero-shot cross-lingual transfer on part-of-speech tagging, dependency parsing, and named entity recognition. While offering (1) improved fine-tuning efficiency (by a factor of around 50 in our experiments), (2) a smaller parameter budget, and (3) increased language coverage, MAD-G remains competitive with more expensive methods for language-specific adapter training across the board. Moreover, it offers substantial benefits for low-resource languages, particularly on the NER task in low-resource African languages. Finally, we demonstrate that MAD-G{'}s transfer performance can be further improved via: (i) multi-source training, i.e., by generating and combining adapters of multiple languages with available task-specific training data; and (ii) by further fine-tuning generated MAD-G adapters for languages with monolingual data.