In software development, ambiguities in requirements described in natural language (NL) prevent the application of formal approaches, posing a difficulty that has been avoided in two main ways: discovery based on formal specifications generated from NL requirements, and the creation of non-ambiguous NL requirements. In the former, NL is more expressive and does not rely on the user’s expertise, but instead makes the automatic generation of formal specifications difficult. The latter facilitates the automatic generation of formal specifications and has the advantage of reduced syntactic complexity, but in exchange for reduced expressiveness of NL. In this paper, we take an approach that allows users to describe highly expressive NL requirements and reduces syntactic complexity to support the automatic generation of formal specifications from NL requirements. Particularly in the automatic generation of model-based specifications, we propose a method for recognizing syntactic patterns that represent semantic relations from which model components are extracted. Applying our method to practical requirement sentences reveals that hypotheses necessary for recognizing syntactic patterns denoting conditional relations of NL requirements can be simplified, which is shown to be effective in reducing the complexity of information extraction rules. Additionally, the results from extracting the elements matching the recognized hypotheses show that the hypotheses are matched multiple times in a sentence. We expect that our method will support the generation of formal specifications from NL requirements without compromising the expressive power of the language.