Developing an English-Efik Corpus and Machine Translation System for Digitization Inclusion
Organizations: University of Cross River State · ML Collective · Arthur Jarvis University · University of Calabar
Abstract
Low-resource languages serve as invaluable repositories of human history, preserving cultural and intellectual diversity. Despite their significance, they remain largely absent from modern natural language processing systems. While progress has been made for widely spoken African languages such as Swahili, Yoruba, and Amharic, smaller indigenous languages like Efik continue to be underrepresented in machine translation research. This study evaluates the effectiveness of state-of-the-art multilingual neural machine translation models for English-Efik translation, leveraging a small-scale, community-curated parallel corpus of 13,865 sentence pairs. We fine-tuned both the mT5 multilingual model and the NLLB200 model on this dataset. NLLB-200 outperformed mT5, achieving BLEU scores of 26.64 for English-Efik and 31.21 for Efik-English, with corresponding chrF scores of 51.04 and 47.92, indicating improved fluency and semantic fidelity. Our findings demonstrate the feasibility of developing practical machine translation tools for low-resource languages and highlight the importance of inclusive data practices and culturally grounded evaluation in advancing equitable NLP.
Figures & tables
| Domain | Number of Sentences |
|---|---|
| Daily Conversation | 7,245 |
| Family and Relationship | 1,885 |
| Food and Agriculture | 1,237 |
| Nature and Environment | 855 |
| Religion | 512 |
| Time and Number | 509 |
| Split | Number of Sentences |
|---|---|
| Train | 10,398 |
| Dev | 1,387 |
| Test | 2,080 |
| Total | 13,865 |
| English | Count | Efik | Count |
|---|---|---|---|
| house | 298 | ke | 3,457 |
| good | 236 | enye | 2,094 |
| want | 220 | mi | 1,407 |
| thing | 211 | oro | 1,087 |
| money | 211 | ye | 830 |
| one | 202 | mme | 791 |
| Hyperparameter | mT5 | NLLB |
| Learning rate | 1e-4 | 2e-5 |
| Train batch size per device | 8 | 2 |
| Eval batch size per device | 8 | 2 |
| Gradient accumulation steps | 1 | 2 |
| Number of epochs | 20 | 6 |
| Warmup ratio | - | 0.1 |
| Model | BLEU | chrF | ||
|---|---|---|---|---|
| en–ef | ef–en | en–ef | ef–en | |
| Zero-shot | ||||
| mT5-small | 0.01 | 0.01 | 2.61 | 3.15 |
| NLLB-200 | 0.11 | 0.22 | 10.95 | 15.16 |
| Encoder–Decoder | ||||
| mT5-small FT | 14.32 | 18.87 | 35.55 | 36.73 |
| Source (En) | Reference (Ef) | mT5-small FT Output | NLLB-200 FT Output |
|---|---|---|---|
| My mother is very sick and she is in the hospital. | Eka mi ọdọñọ eti eti ndien enye odu ke ufọkibọk. | Eka mi ọsọñọ [ eti eti omitted ] ndien enye ọdiọñọ ke ufọkibọk. | Eka mi ọdọñọ eti eti ndien enye odu ke ufọkibọk. |
| Are you carrying that umbrella for the rain or just to look fancy? | Ndi akama ufukeyo ke ntak edịm mme onyuñ akama ke asian? | Ndi afo ebet ufukeyo oro ọnọ ed i m mme ñkpọ etiede ? | Ndi emen ufukeyo oro ke ntak ed i m mme ndikut ọkpọsọñ ñkpọ ? |
| Source (Ef) | Reference (En) | mT5-small FT Output | NLLB-200 FT Output |
|---|---|---|---|
| Udia usenubọk amakebe, mmọ ewọrọ anwa ndinwam ekpri eyeneka esie ndimen akpasa nya inwañ mbakara kiet nka efe urua. | After breakfast, they went outside to help his younger brother carry a basket of tomatoes to the stall. | The morning went down , they went to help his little sister pick up a basket of yams to the market . | After breakfast, they set off to help his younger brother carry a basket of tomatoes to the stall. |
| William Otu ekedi akwa owo ke akamba usiakifia aran isọñ ke Lagos. | William Otu was a seasoned manager at a big oil company in Lagos. | William group was a very heavy trader in the mining companies . | William Otu was a senior official in the Lagos oil company. |
Machine Translation
Machine translation develops computational methods for converting text, speech, and visual content between languages while preserving meaning and purpose.
See all 267 papers