Fault-tolerant information extraction

A computer system for fault-tolerant information extraction, comprising:

• a processor;
• and
• a non-transitory computer-readable medium storing instructions that, when executed by the processor, cause the computer system to carry out a method for automatically analyzing a natural language input for fault-tolerant named-entity recognition, the method including:
• (i) a step of receiving the natural language input;
• (ii) a step of providing a grammar model, the grammar model being represented as an augmented transition network with outputs and comprising:
• a local grammar model used for modelling a syntactic and semantic structure of at least a part of the natural language input, with a finite set of state transitions between states, the state transitions being defined by:
• a current state, a following state, an input label and an output label;
• a set of external functions for information extraction, linked to the local grammar model, wherein one or more of the external functions are attached to the state transitions, each external function being external in the sense that each external function is not part of the local grammar model and allows the grammar model to be fault-tolerant against morphology, syntax, semantic and/or discursive issues present in the natural language input;
• a finite set of read/write memory registers used by a finite state machine or a finite state transducer to parse expressions and by the one or more external functions, in which are stored the natural language input, the states, the outputs of the grammar model, and/or other data;
• and
• each of the one or more external functions being an operation over:
• at least the part of the natural language input, one or more states, the outputs of the grammar model already stored on the memory registers, and/or other data;
• (iii) a step of applying the grammar model to at least a part of the natural language input using the finite state machine or the finite state transducer, comprising:
• a call of the one or more external functions by the finite state machine or the finite state transducer, evaluating the external functions with a register-based virtual machine;
• and
• the grammar model changing from the current state to the following state whenever a condition is fulfilled that the one or more external functions attached to a state transition returns a return value, which is a non-empty string, a number or a boolean value equal to true;
• (iv) a step of extracting information from the at least part of the natural language input using at least an output of the grammar model being built based on:
• at least one return value of the one or more external functions from evaluating the one or more external functions in step (iii), wherein the at least one return value has been stored in a memory register;
• and
• one or more input labels and/or output labels;
• the grammar model being able to:
• access information from the memory registers and/or from an external computer-implemented resource, method or service, to generate the outputs;
• and
• identify and classify named entities in the natural language input into predetermined categories.