OPERA: Operations-oriented Probabilistic Extraction, Reasoning, and Analysis

The OPERA system (for Operations-oriented Probabilistic Extraction, Reasoning, and Analysis) developed jointly by CMU and USC/ISI is an integrated solution to the challenges of DARPAs Active Interpretation of Disparate Alternatives (AIDA)program in the form of: (i) high-performance media analysis (TA1) for text, speech, and image/video data, (ii) semantic representation and reasoning support (TA1 and TA2), (iii) cross-medium and cross-language integration (TA2), and (iv) hypothesis creation, management, and hypothesis exploration (TA3). Given that all required components of such a systemare still active areas of research, the creation of a single system (pipelined or otherwise) has the potential for a substantialrate of compounded errors. Early versions of the system created had strong abstraction boundaries for limited informationsharing between systems. Later incarnations benefited from allowing for the output of extractors to be coupled with raw textstrings and embedding vectors. These prove especially advantageous in the presence of large-scale language models thatencode world knowledge, and when aligning predictions to an open-domain ontology, like that of WikiData.