The present experiment is motivated by two strands of research. On the one hand, it builds on contrastive and cognitive linguistic work on article systems in translation, which raise the broader question of how determinacy is encoded in languages without articles. On the other hand, it engages with recent discussions in machine translation regarding candidate selection methods that do not rely exclusively on model probabilities. The initial assumption is that Polish word order itself encodes determinacy. Specifically, the generic hypothesis predicts that preverbal noun phrases are more likely to trigger definite articles in English translations, whereas postverbal noun phrases tend toward indefinite interpretations. To test this configuration, we employ conservative selections of verbs and nouns, combining corpus-based examples with synthetic samples. While the issue is fundamentally one of sentence pragmatics, the present study restricts itself to word order mapping from source to target as a proxy for determinacy construal, its relation to sentence alignment and translation quality metrics. Evidence is drawn from multiple MT systems trained with different objectives and explored under a range of decoding strategies (beam search, top-k sampling, minimum Bayes-risk decoding). Together, this design allows us to evaluate whether article choice in English translations is systematically influenced by Polish source word order.
Paper
The full text of this publication is not hosted on 44B due to licensing.
Read it at OpenAlex