A new way for computers to recognise and translate complex place names is reported in the International Journal of Information and Communication Technology. The approach offers a roadmap to address a long-standing weakness in digital language systems used for mapping, navigation, and international communication.
Place names often carry historical, geographical, and cultural significance, and errors in translation can lead to confusion or loss of context. More accurate handling of such names could improve digital maps, navigation systems, logistics platforms, and multilingual communication tools.
The research focuses on English-derived place names, those created by adding prefixes, suffixes, or descriptive elements to existing names. While common in geographic data, these constructions are hard for automated systems to work with because they combine meaning and pronunciation in ways that do not transfer neatly across languages.
To address this, the researchers developed a computational model that integrates two complementary approaches: a knowledge graph and a phonetic generation algorithm. A knowledge graph is a structured representation of information that maps relationships between concepts, allowing the system to understand how place names are formed and how their components relate to one another. This captures the semantic dimension of language, its meaning and contextual associations.
The phonetic generation algorithm focuses on the sound of the spoken names. It converts written words into standardised representations of pronunciation, enabling the system to align how a place name is written with how it is spoken. This is particularly important in translation, where names often need to preserve recognisable sounds alongside meaning.
These two elements interact using what the team refers to as a bidirectional dynamic interaction fusion mechanism. In this system, the semantic and phonetic information feed each other to improve recognition and translation. The system also uses a Long Short-Term Memory (LSTM) network, a type of neural network commonly used for language processing.
The model demonstrated an error rate of just 1.3 per cent in recognising place names and 0.8 per cent in translating them. Its outputs are more than 95 per cent fluent and consistent.
Ma, D. (2026) ‘English-derived place name recognition and translation based on knowledge graph and phonetic generation algorithm’, Int. J. Information and Communication Technology, Vol. 27, No. 27, pp.109–132.
No comments:
Post a Comment