A statistical machine translation (SMT) system exhibits systematically lower BLEU scores when translating technical documents with heavy jargon. Which mechanism explains this effect?

💡 Explanation

The lower BLEU scores occur because the SMT system’s vocabulary lacks the domain-specific jargon, leading to frequent unknown word tokens and inaccurate translations; therefore, poor vocabulary coverage is the reason, rather than issues related to context vectors or network architecture which affect fluency more than accuracy.

• Why does a message sent over a noisy radio channel degrade in intelligibility when the signal-to-noise ratio falls below a critical threshold?
• Why does coarticulation alter consonant production differently across languages?
• Why does spatial reasoning exhibit variations across cultures, considering language relativity in map-reading tasks?
• Why does emoji usage differ significantly across online platforms, leading to potential misinterpretations?
• If a language experiences decreased usage in scientific publications and academic discourse, which consequence follows regarding its social status?
• Why does early cuneiform writing exhibit high polysemy?