The SignCom project uses motion capture (mocap) data to animate a virtual French Sign Language (LSF) signer. An important part of any signing avatar project is to ensure that a computer animation engine has a large quantity of interesting and on-topic signs from which to build novel signing sequences. In this article, we detail the process of selecting an adequate range of signs and situations to be included in our corpus: from controlling discourse topic to including signs that can accept modified movements or handshapes, we describe how an avatar corpus has a different motivation than traditional signed language corpora.