Language Pairs

English to Yoruba: AI Translation Comparison

Updated 2026-03-10

English to Yoruba: AI Translation Comparison

Yoruba is spoken by over 45 million people, primarily in southwestern Nigeria, Benin, and Togo, with significant diaspora communities worldwide. It is a tonal language with three lexical tones (high, mid, low) that are essential for meaning — the same sequence of consonants and vowels can produce entirely different words depending on tone. Demand for English-to-Yoruba translation is driven by Nigerian government services, education, media, Nollywood film subtitling, and digital content localization for Nigeria’s massive internet population.

This comparison evaluates five leading AI translation systems on English-to-Yoruba accuracy, naturalness, and suitability for different use cases.

Translation comparisons are based on automated metrics and editorial evaluation. Quality varies by language pair and content type.

Accuracy Comparison Table

SystemBLEU ScoreCOMET ScoreEditorial Rating (1-10)Best For
Google Translate18.30.7245.4General-purpose, broadest data
DeepL14.60.6914.4Very limited Yoruba support
GPT-420.90.7425.9Contextual accuracy, tonal awareness
Claude18.80.7285.5Long-form content
NLLB-20022.10.7536.2Best Yoruba coverage, self-hosted

Translation Quality Metrics: BLEU, COMET, and Human Evaluation Explained

Example Translations

Formal Business Email

Source: “We are pleased to inform you that your application has been approved. Please find the relevant documentation attached.”

SystemTranslation
GoogleA ni idunnu lati fi to yin leti pe ohun elo yin ti je itewogba. Ejowo, e wa awon iwe afikun ti a so po.
DeepLA ni idunnu lati so fun yin pe ohun elo yin ti gba ifọwọsi. Jọwọ wa awọn iwe ti o yẹ ti a so mọ.
GPT-4A ni inudidun lati fi to yin leti pe iwe ibere yin ti gba ifọwọsi. Ẹ jọwọ wo awọn iwe afikun ti a so pọ mọ lẹta yii.
ClaudeA ni inudidun lati fi to yin leti pe ohun elo yin ti je itẹwọgba. Ẹ jọwọ, ẹ wo awọn iwe to yẹ ti a fi kun.
NLLB-200A ni inudidun lati fi to yin leti pe iwe ibere yin ti gba ifọwọsi. Ẹ jọwọ wo awọn iwe afikun ti a so pọ mọ.

Assessment: NLLB-200 and GPT-4 produce the most natural formal Yoruba, with correct use of diacritics (subdots and tone marks). Google’s output omits most diacritics, which significantly impairs readability and can change word meanings. The presence of proper Yoruba diacritics (ẹ, ọ, ṣ) is a basic quality indicator for this language.

Casual Conversation

Source: “Hey, I was thinking we could grab some food later. What do you feel like eating?”

SystemTranslation
GoogleHey, mo n ro pe a le lo mu ounje nigbamii. Kini o fe je?
DeepLHey, mo n ro pe a le gba ounje nigbamii. Kini o fe je?
GPT-4Ẹ n lẹ, mo n ro pe a le lọ jẹun nigbamii. Kini o fẹ jẹ?
ClaudeBawo, mo n ro pe a le lọ mu ounjẹ nigbamii. Kini o fẹ jẹ?
NLLB-200Mo n ro pe a le lọ mu ounjẹ nigbamii. Kini o fẹ jẹ?

Assessment: GPT-4 and Claude attempt Yoruba greetings instead of transliterating “Hey.” Claude’s “Bawo” is a natural casual greeting. Google and DeepL keep “Hey” untranslated, which is actually common in modern Nigerian Yoruba, reflecting English influence. All systems struggle to fully capture casual Nigerian Yoruba, which is heavily influenced by code-switching with English.

Technical Content

Source: “The API endpoint accepts POST requests with a JSON body containing the source text and target language code.”

SystemTranslation
GoogleAPI endpoint gba awon ibere POST pelu JSON body ti o ni orisun ọrọ ati koodu ede ibi-afẹde.
DeepLAaye ipari API gba awọn ibeere POST pẹlu ara JSON ti o ni ọrọ orisun ati koodu ede ibi-afẹde.
GPT-4API endpoint n gba awọn POST requests pẹlu JSON body ti o ni source text ati target language code ninu.
ClaudeAPI endpoint n gba awọn ibeere POST pẹlu JSON body ti o ni ọrọ orisun ati koodu ede ti a n wa.
NLLB-200Aaye ipari API n gba awọn ibeere POST pẹlu ara JSON ti o ni ọrọ orisun ati koodu ede afojusun ninu.

Assessment: GPT-4 keeps English technical terms intact, reflecting how Nigerian Yoruba developers actually communicate. DeepL and NLLB-200 translate “endpoint” as “aaye ipari” (end point), which has no established usage in Yoruba tech communities. Best Translation AI for Technical Documentation

Strengths and Weaknesses

Google Translate

Strengths: Accessible and free. Benefits from Nigerian web content. Weaknesses: Frequently omits Yoruba diacritics (subdots and tone marks), which can change word meanings entirely. Quality below the other systems for this pair.

DeepL

Strengths: Basic grammatical structure is usually correct. Weaknesses: Very limited Yoruba support. Lowest quality overall. Poor diacritical mark handling. Unnatural word choices.

GPT-4

Strengths: Best contextual understanding. Better diacritical mark usage than Google. Can handle English-Yoruba code-switching naturally. Best tonal awareness in text output. Weaknesses: Expensive. Not always consistent with tone marks across longer texts.

Claude

Strengths: Consistent output for long documents. Reasonable diacritical mark handling. Weaknesses: Less natural than GPT-4 for idiomatic Yoruba. Limited casual register capability.

NLLB-200

Strengths: Best free option for Yoruba by a significant margin. Meta’s NLLB project made Yoruba a priority language. Consistent diacritical mark usage. Self-hostable. Weaknesses: No register control. Over-translates English terms in technical content. Cannot handle code-switching.

Recommendations

Use CaseRecommended System
Quick personal translationGoogle Translate (free, but verify diacritics)
Government / official documentsGPT-4 with human review
Educational materialNLLB-200
Media / Nollywood subtitlesGPT-4
Technical documentationGPT-4
High-volume, cost-sensitiveNLLB-200 (self-hosted)
Long-form contentClaude

Best Translation AI in 2026: Complete Model Comparison

Key Takeaways

  • NLLB-200 leads as the best free option, outperforming Google Translate on Yoruba thanks to Meta’s targeted investment in African languages. GPT-4 offers the highest contextual quality at a cost.
  • Diacritical marks are not optional in Yoruba — they distinguish different words. Systems that omit subdots and tone marks produce ambiguous or incorrect text.
  • Code-switching between English and Yoruba is extremely common in Nigerian communication. GPT-4 handles this best; NMT systems struggle with mixed-language input.
  • All systems produce lower quality for Yoruba than for high-resource languages. Human review is essential for published content.

Next Steps