Cross-lingual information retrieval and information extraction

A training apparatus, the apparatus comprising:

• processing circuitry and memory;
• the processing circuitry to:
• access a dataset comprising annotated documents in a first natural language and translations of the annotated documents into a second natural language, wherein each annotated document in the first natural language is annotated with one or more events, wherein the second natural language is different from the first natural language, wherein each of the one or more events event is represented, in the first natural language, using a least one trigger word and at least one argument;
• generate a second natural language document-event map by mapping, to each translation of each annotated document into the second natural language, the one or more events with which the annotated document in the first natural language is annotated;
• and
• train, using a neural network training technique, a cross-lingual information retrieval module to associate a new document in the second natural language to at least one event using the second natural language document-event map;
• and
• provide an output representing the trained cross-lingual information retrieval module.

A non-transitory machine-readable medium storing instructions which, when executed by processing circuitry of one or more machines, cause the processing circuitry to:

• access a query in a first natural language;
• identify an event corresponding to the query, wherein the event is represented, in the first natural language, using a least one trigger word and at least one argument;
• compute, using a cross-lingual information retrieval module and without translating the query or the event corresponding to the query into a second natural language, a ranked list of documents in the second natural language that are related to the event, wherein at least a portion of documents in the ranked list are selected from a collection of documents in the second natural language that are not annotated with events, wherein the cross-lingual information retrieval module is trained using a dataset comprising annotated documents in the first natural language and translations of the annotated documents into the second natural language, wherein each annotated document in the first natural language is annotated with one or more events, wherein the dataset is generated by propagating the one or more events from a given annotated document in the first natural language to a translation of the given annotated document into the second natural language;
• and
• provide an output representing at least a portion of the ranked list of documents in the second natural language, wherein the second natural language is different from the first natural language.

A cross-lingual search method, the method comprising:

• accessing a query in a first natural language;
• identifying an event corresponding to the query, wherein the event is represented, in the first natural language, using a least one trigger word and at least one argument;
• computing, using a cross-lingual information retrieval module and without translating the query or the event corresponding to the query into a second natural language, a ranked list of documents in the second natural language that are related to the event, wherein at least a portion of documents in the ranked list are selected from a collection of documents in the second natural language that are not annotated with events, wherein the cross-lingual information retrieval module is trained using a dataset comprising annotated documents in the first natural language and translations of the annotated documents into the second natural language, wherein each annotated document in the first natural language is annotated with one or more events, wherein the dataset is generated by propagating the one or more events from a given annotated document in the first natural language to a translation of the given annotated document into the second natural language;
• and
• providing an output representing at least a portion of the ranked list of documents in the second natural language, wherein the second natural language is different from the first natural language.