A Survey of Orthographic Information in Machine Translation

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Machine Translation

Background:

• Machine translation (MT) systems face challenges with under-resourced and closely related languages due to linguistic and orthographic variations.
• Traditional MT approaches struggle with differing orthographic conventions, hindering comparability and performance.
• Orthographic information, often overlooked, presents an opportunity to improve MT systems for low-resource scenarios.

Purpose of the Study:

• To survey research on the influence of orthography on machine translation for under-resourced languages.
• To explore how orthographic knowledge can be leveraged to enhance MT performance.
• To highlight recent trends linking orthographic information with established MT methods, including multilingual neural MT.

Main Methods:

• Literature review of existing research on orthography and machine translation.
• Analysis of studies utilizing orthographic information, including cognate data, to improve MT.
• Discussion of different MT types and their integration with orthographic knowledge.

Main Results:

• Orthographic knowledge demonstrably improves machine translation performance for under-resourced languages.
• Cognate information at various levels aids MT, offering valuable insights.
• Multilingual neural machine translation of closely related languages shows promise when incorporating orthographic features.

Conclusions:

• Orthographic information is a crucial factor for advancing machine translation, particularly for languages with limited resources.
• Future directions involve integrating orthographic knowledge into multilingual MT settings and bilingual lexicon induction.
• Leveraging orthographic nuances can bridge linguistic gaps and enhance cross-lingual communication.