Scots Gaelic to Irish: AI Translation Comparison
Scots Gaelic to Irish: AI Translation Comparison
Scots Gaelic (Gaidhlig) and Irish (Gaeilge) are Goidelic Celtic languages with approximately 57,000 and 1.7 million speakers respectively, though the number of daily speakers is much smaller for both. These sister languages descended from Old Irish share substantial vocabulary, similar grammatical structures including VSO word order, initial consonant mutations, and the genitive case, but have diverged over centuries of separate development in Scotland and Ireland. Both use modified Latin scripts with different orthographic conventions. This pair is important for Celtic cultural preservation, academic Celtic studies, cross-Gaelic cultural events, and the growing language revitalization movements in both countries. AI training data is extremely limited for both languages, making this one of the most challenging pairs for current systems.
This comparison evaluates five leading AI translation systems on Scots Gaelic-to-Irish 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
| System | BLEU Score | COMET Score | Editorial Rating (1-10) | Best For |
|---|---|---|---|---|
| Google Translate | 15.8 | 0.756 | 5.5 | General-purpose, speed |
| DeepL | 17.2 | 0.768 | 5.9 | Formal content |
| GPT-4 | 20.4 | 0.789 | 6.6 | Context, linguistic knowledge |
| Claude | 18.1 | 0.774 | 6.1 | Long-form content |
| NLLB-200 | 14.3 | 0.745 | 5.2 | Budget, self-hosted |
Translation Quality Metrics: BLEU, COMET, and Human Evaluation Explained
Example Translations
Formal Business Email
Source: “A Mhgr MacLeoid, tha e na thoileachas dhuinn innse dhuibh gun deach an tagradh agaibh aontachadh. Feuch na sgriobhainnean ceangailte.”
| System | Translation |
|---|---|
| A Uasail MacLeoid, is cuis athais duinn a chur in iul duit gur faomhadh d’iarratas. Feach na cáipéisí atá ceangailte. | |
| DeepL | A Uasail MacLeoid, tá áthas orainn a chur in iúl duit go bhfuil d’iarratas ceadaithe. Féach ar na cáipéisí atá i gceangal. |
| GPT-4 | A Uasail Mhic Leoid, is mór an pléisiúr dúinn a chur in iúl duit gur ceadaíodh d’iarratas. Breathnaigh ar na cáipéisí atá ceangailte leis seo, le do thoil. |
| Claude | A Uasail MacLeoid, tá áthas orainn a chur in iúl duit gur ceadaíodh d’iarratas. Féach ar na cáipéisí ceangailte. |
| NLLB-200 | MacLeoid, ceadaíodh d’iarratas. Tá na cáipéisí ceangailte. |
Assessment: GPT-4 produces the most formally elaborate Irish with is mór an pléisiúr dúinn (it is a great pleasure for us), the correct lenition in Mhic Leoid (genitive of Mac Leoid), and le do thoil (please). DeepL handles the Gaelic-to-Irish conversion reasonably. NLLB-200 drops all formal markers and the vocative particle A.
Casual Conversation
Source: “Haidh! An deach thu dhan taigh-bidh ur sin? Bha e sgoinneil! Feumaidh tu a dhol ann.”
| System | Translation |
|---|---|
| Heileo! An ndeachaigh tu go dti an bhialann nua sin? Bhi se ar fheabhas! Caithfidh tu dul ann. | |
| DeepL | Heileo! An raibh tu sa bhialann nua sin? Bhí sé iontach! Caithfidh tú dul ann cinnte. |
| GPT-4 | Haigh! An ndeachaigh tú go dtí an bhialann nua úd? Bhí sé dochreidte! Caithfidh tú dul ann, dáiríre! |
| Claude | Heileo! An ndeachaigh tu go dti an bhialann nua sin? Bhi se iontach! Caithfidh tu dul ann. |
| NLLB-200 | An ndeachaigh tu go dti an bhialann nua? Bhi se go maith. Caithfidh tu dul ann. |
Assessment: GPT-4 captures the casual energy best with dochreidte (incredible) and dáiríre (seriously). DeepL uses cinnte (certainly) for emphasis. NLLB-200 uses the flat go maith (good), losing the Scottish Gaelic sgoinneil (brilliant). All systems correctly map VSO structures between the two Goidelic languages.
Technical Content
Source: “Tha am modail ionnsachaidh domhainn a cleachdadh ailtireachd transformer le innealan aire gus data sreath a ghiullachd.”
| System | Translation |
|---|---|
| Usáideann an tsamhail foghlama domhain ailtireacht transformer le meicníochtaí airde chun sonraí sraitheacha a phróiseáil. | |
| DeepL | Baineann an tsamhail foghlama domhain úsáid as ailtireacht transformer le meicníochtaí airde chun sonraí seicheamhacha a phróiseáil. |
| GPT-4 | Usáideann an deep learning model ailtireacht transformer le attention mechanisms chun sequential data a phróiseáil. |
| Claude | Usáideann an tsamhail foghlama domhain ailtireacht transformer le meicníochtaí airde chun sonraí sraitheacha a phróiseáil. |
| NLLB-200 | Usáideann an tsamhail foghlama domhain ailtireacht an athraitheora le meicníocht airde chun sonraí a phróiseáil. |
Assessment: GPT-4 keeps English ML terms, practical given the limited Irish technical vocabulary for ML. Others attempt Irish translations with varying accuracy. NLLB-200 translates transformer as athraitheora (changer), not a recognized term. See Low-Resource Languages: How NLLB and Aya Are Closing the Gap for Celtic language coverage analysis.
Strengths and Weaknesses
Google Translate
Strengths: Fast and free. Benefits from Google’s Irish language investments. Limited Scots Gaelic support. Weaknesses: Very limited training data for both languages. Likely pivots through English. Many errors.
DeepL
Strengths: Slightly better than Google on formal content. Basic Celtic structure handling. Weaknesses: Neither language is a core DeepL language. Significant quality gap with major pairs.
GPT-4
Strengths: Best quality for this low-resource pair. Benefits from linguistic knowledge of Celtic languages. Weaknesses: Higher cost. Still significantly limited by scarce training data for both languages.
Claude
Strengths: Reasonable long-form quality. Better than NLLB-200 on register handling. Weaknesses: Less effective than GPT-4 on the specific Celtic linguistic features.
NLLB-200
Strengths: Free and self-hostable. NLLB-200 covers Irish but Scots Gaelic support is minimal. Weaknesses: Lowest quality. Many errors in mutation, case, and vocabulary. Very limited Scots Gaelic comprehension.
Recommendations
| Use Case | Recommended System |
|---|---|
| Basic comprehension | Google Translate |
| Academic Celtic studies | GPT-4 with human review |
| Cultural content | GPT-4 |
| Long-form content | Claude |
| Bulk processing | NLLB-200 (self-hosted) |
| Critical documents | Human translator essential |
Best Translation AI in 2026: Complete Model Comparison
Key Takeaways
- GPT-4 leads for Scots Gaelic-to-Irish, though all systems show low quality on this extremely low-resource Celtic pair.
- Initial consonant mutation rules in both languages are frequently mishandled by all systems, despite being grammatically essential.
- The shared VSO word order and Goidelic morphology provide structural alignment, but vocabulary divergence over centuries is the primary challenge.
- For any content beyond basic comprehension, human translation by a Celtic language specialist remains essential.
Next Steps
- Try it yourself: Compare these systems on your own text in the Translation AI Playground: Compare Models Side-by-Side.
- Reverse direction: See Luxembourgish to German: AI Translation Comparison.
- Check the leaderboard: Browse our full Translation Accuracy Leaderboard by Language Pair.
- Full model comparison: Read Best Translation AI in 2026: Complete Model Comparison.