Language Pairs

English to Galician: AI Translation Comparison

Updated 2026-03-11

English to Galician: AI Translation Comparison

Galician is spoken by approximately 2.4 million people, primarily in the autonomous community of Galicia in northwestern Spain. Galician is closely related to Portuguese, sharing a common origin in medieval Galician-Portuguese, and remains mutually intelligible with Portuguese to a significant degree. It uses the Latin script, has a two-gender system, employs a personal infinitive (where the infinitive can be conjugated for person, a rare feature), and places clitic pronouns according to complex rules that differ from both Spanish and Portuguese. Translation demand comes from the Xunta de Galicia (regional government), education, cultural preservation, fishing and maritime industry documentation, and academic research.

This comparison evaluates five leading AI translation systems on English-to-Galician 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 Translate30.20.8226.4General-purpose, free access
DeepL32.60.8396.8Business and formal documents
GPT-433.10.8437.0Contextual accuracy
Claude30.80.8266.5Long-form content
NLLB-20028.40.8056.0Free option, self-hosted

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

Example Translations

Formal Business Email

Source: “We are writing to notify you of the updated regulations governing sustainable fishing practices in Galician coastal waters. Please review the enclosed guidelines.”

SystemTranslation
GoogleEscribímoslles para notificarlles as regulacións actualizadas que rexen as prácticas de pesca sustentable nas augas costeiras galegas. Revisen as directrices adxuntas.
DeepLDirixímonos a vostede para notificarlle as regulacións actualizadas que rexen as prácticas de pesca sostible nas augas costeiras galegas. Revise as directrices adxuntas, por favor.
GPT-4Escribímoslle para informalo das normativas actualizadas que regulan as prácticas de pesca sostible nas augas costeiras galegas. Revise a documentación adxunta, por favor.
ClaudeEscribímoslles para notificarlles as regulacións actualizadas que rexen as prácticas de pesca sustentable nas augas costeiras galegas. Revisen as directrices adxuntas.
NLLB-200Escribímoslle para notificarlle as regulacións actualizadas que rexen as prácticas de pesca sostible nas augas costeiras galegas. Revise as directrices adxuntas.

Assessment: DeepL and GPT-4 correctly add “por favor” (please), capturing the politeness of the source. GPT-4’s “normativas” (norms/regulations) and “regulan” (regulate) are natural Galician administrative vocabulary. DeepL uses “Dirixímonos a vostede” (We address you), a natural formal opening. Google and Claude use the plural form “Escribímoslles/notificarlles” while the source is singular, which may be a formality choice but is less precise. Galician’s “sostible” versus “sustentable” for “sustainable” reflects an ongoing standardization debate.

Casual Conversation

Source: “Did you go to the pulpo festival in Carballiño? The food was amazing. We ate so much I could barely walk after.”

SystemTranslation
GoogleFuches á festa do polbo no Carballiño? A comida estaba incrible. Comemos tanto que case non podía camiñar despois.
DeepLFuches á festa do polbo no Carballiño? A comida estaba fantástica. Comemos tanto que case non podía camiñar despois.
GPT-4Fuches á Festa do Polbo do Carballiño? A comida estaba de morte. Comemos tanto que case non podía nin andar despois.
ClaudeFuches á festa do polbo no Carballiño? A comida estaba incrible. Comemos tanto que case non podía camiñar despois.
NLLB-200Foi á festa do polbo no Carballiño? A comida estaba incrible. Comemos tanto que case non podía camiñar despois.

Assessment: GPT-4 uses “de morte” (literally “to die for”), a natural Galician casual expression for excellent food. GPT-4 also capitalizes “Festa do Polbo” as a proper noun (it is an official festival). “Nin andar” (not even walk) is more expressive than “camiñar” (walk). NLLB-200 uses formal “Foi” instead of casual “Fuches.” All systems correctly reference O Carballiño’s famous octopus festival, a key cultural event in Galicia. Best Translation AI for Casual Content

Technical Content

Source: “The marine biology research station uses acoustic telemetry to track fish migration patterns along the Galician continental shelf.”

SystemTranslation
GoogleA estación de investigación en bioloxía mariña usa telemetría acústica para rastrexar os patróns de migración dos peixes ao longo da plataforma continental galega.
DeepLA estación de investigación de bioloxía mariña utiliza telemetría acústica para rastrexar os patróns migratorios dos peixes ao longo da plataforma continental galega.
GPT-4A estación de investigación en bioloxía mariña emprega telemetría acústica para seguir os patróns de migración dos peixes ao longo da plataforma continental galega.
ClaudeA estación de investigación en bioloxía mariña usa telemetría acústica para rastrexar os patróns de migración dos peixes ao longo da plataforma continental galega.
NLLB-200A estación de investigación en bioloxía mariña usa telemetría acústica para rastrexar os patróns de migración dos peixes ao longo da plataforma continental galega.

Assessment: GPT-4 uses “emprega” (employs), which is a more native Galician word than “usa” (uses, shared with Spanish). DeepL’s “patróns migratorios” (migratory patterns) uses an adjectival form that reads more naturally in academic Galician. Marine biology is a strong domain for Galician translation due to the region’s fishing industry and marine research institutions. All systems handle the scientific terminology competently. Best Translation AI for Technical Documentation

Strengths and Weaknesses

Google Translate

Strengths: Free and accessible. Reasonable quality for general content. Benefits from Xunta de Galicia translation data. Weaknesses: Sometimes produces Spanish-influenced forms. Clitic pronoun placement errors. Limited idiomatic Galician.

DeepL

Strengths: Good formal document quality. Natural vocabulary. Correct formal address. Weaknesses: Premium pricing. Occasionally uses Portuguese rather than Galician forms. Limited cultural context.

GPT-4

Strengths: Best overall quality. Most natural vocabulary with genuinely Galician word choices. Good cultural awareness. Handles the personal infinitive. Weaknesses: Higher cost. Occasional Portuguese influence. Sometimes over-colloquializes formal content.

Claude

Strengths: Consistent quality for long documents. Reliable formal register. Weaknesses: Less natural than GPT-4. Limited Galician-specific knowledge. Similar to Google quality.

NLLB-200

Strengths: Free and self-hostable. Basic functionality. Weaknesses: Formal register default. Lower quality. Sometimes generates Portuguese or Spanish instead of Galician.

Recommendations

Use CaseRecommended System
Government / institutionalDeepL or GPT-4
Maritime / fishing documentationGPT-4
Cultural / tourism contentGPT-4
Education materialsDeepL
High-volume, cost-sensitiveNLLB-200 (self-hosted)
Quick personal translationGoogle Translate (free)
Long-form / academicClaude

Best Translation AI in 2026: Complete Model Comparison

Key Takeaways

  • GPT-4 leads for English-to-Galician with the most genuinely Galician vocabulary and cultural awareness. DeepL is the best choice for formal institutional content.
  • The Galician-Portuguese-Spanish triangle is the key challenge: AI systems frequently produce output that is technically comprehensible but influenced by Spanish or Portuguese, which native Galician speakers notice immediately.
  • Galician’s personal infinitive, complex clitic placement rules, and distinct vocabulary make it a separate translation target from both Spanish and Portuguese, but limited training data means AI systems sometimes conflate them.
  • Galicia’s fishing industry and marine research institutions create domain-specific translation demand where all systems perform reasonably well.

Next Steps