System and method for active learning based multilingual semantic parser

A multilanguage semantic parser apparatus, comprising:

• a processor;
• and
• a memory storing instructions that, when executed by the processor, configure the multilanguage semantic parser apparatus to:
• receive a multilingual training dataset, wherein the multilingual training dataset includes pairs of utterances and meaning representations from at least one high-resource language and at least one low-resource language and wherein the multilingual training dataset is initially a machine-translated dataset;
• train a multilanguage semantic parser by translating the utterances in the multilingual training dataset to a target language;
• and
• iteratively perform:
• select a subset of the multilingual training dataset for human translation;
• update the multilingual training dataset with a human-translated subset of the multilingual training dataset;
• and
• retrain the multilanguage semantic parser with the updated multilingual training dataset,
• wherein scores, which are used to select the subset, for each of the utterances in the multilingual training dataset or the updated multilingual training dataset are based on a translation bias score, a translation error score, a semantic diversity score, and a semantic density score.