Machine map label translation

A computer-implemented process for translating a map label from a first language to a second language, comprising:

• receiving an entity's map label in a first language that is to be translated into a second language;
• generating translated terms in the second language for constituent parts of the map label in the first language;
• generating translation candidate sequences by using combinations of translated terms for each constituent part of the received map label;
• selecting a number of the highest interim ranking generated translation candidate sequences by performing an interim ranking of generated translation candidate sequences by using transition probabilities between terms of each translation candidate sequence to determine a resulting transition probability for that translation candidate sequence;
• extracting features from the entity's map label in the first language and the number of generated highest interim ranking translation candidate sequences by using geospatial and linguistic context information;
• using the extracted features and a trained classifier trained at least in part with the extracted features re-ranking the interim ranking of the translation candidate sequences;
• and
• post-ranking re-ranked top translation candidate sequences using neighboring proximity information to disclose the highest ranking sequence as the translated map label.

A system comprising:

• a processor;
• and
• a memory, the memory including a word breaker, a translation candidate generator, a translation candidate sequence selector, a feature extractor, a candidate sequence re-ranker and a post ranker, wherein:
• the word breaker receives an entity's map label in a first language that is to be translated into a second language and breaks the map label into constituent parts;
• the translation candidate generator generates translation candidates for each constituent part in the entity's map label and uses these translation candidates to generate translation candidate sequences for the map label;
• the candidate sequence selector selects a number of highest interim ranking translation candidate sequences by using transition probabilities between terms in each of the translation candidate sequences;
• the feature extractor extracts features from the highest interim ranking translation candidate sequences and the entity's map label by using geospatial and linguistic context information;
• the candidate sequence re-ranker uses the extracted features to re-rank the highest interim ranking translation candidate sequences using a probabilistic classifier trained at least in part using the extracted features, wherein each translation candidate sequence is assigned a ranking score;
• and
• the post ranker re-ranks the re-ranked translation candidate sequences by using rules based on the map labels of entities in the neighboring proximity of the received entity's map label location.