Machine translation method and apparatus

Abstract

In a machine translation method and apparatus, an input text analyzer analyzes an input text in a first language to form a graph structure of nodes and edges. An equivalent selector introduces variables to ambiguous components of the graph structure by a variable introducer. A constraint solver propagates constraints on the variables to other components. Solutions are then found for equivalent determination and locally consistent solutions are also found for the respective components. A variable realizer finds solutions to the respective locally consistent components according to the patterns of the variables. An output text generator then generates text in a second language from a set of these solutions. Thus, the method and apparatus enable avoidance of an explosion of combinations between source words and equivalents capable of occurring in the analysis phase and also enables the optimum use of information through deferring the determination of words or phrases in the second language of ambiguous components to the text generation phase.

Dependencies are inferred from the claim text.