Controlled Natural Language Framework for Generating Assertions from Hardware Specifications

摘要

In this paper, we present a controlled natural language (CNL) framework for automatic processing and generation of assertions from hardware design specification. Current CNL systems have limitations in mapping differently worded sentences with the same meaning to the same logic structures. We aim to mitigate this limitation by developing a dependency grammar based CNL where the constructed parse tree does not follow strict surface-structure dependencies and instead extract additional relationship based on semantic information that is embedded in the grammar. In addition, current translation schemes for creating executable assertions from hardware design specifications do not provide feedback on wrongly or ambiguously written input sentences. Our natural language understanding algorithm is guided by the dependencies in the parse tree and has the capability to offer useful feedback for sentences that are not fully understood. We reported results on natural language assertions extracted from UART, Memory and AMBA AXI protocol specification documents. We successfully tested syntactic variations of these specifications as well.