This paper concerns the transmission of two independent Gaussian sources over a two-user decentralized interference channel, assuming that the transmitters are unaware of the instantaneous CSIs. The availability of the channel state information at receivers (CSIR) is considered in two scenarios of perfect and imperfect CSIR. In the imperfect CSIR case, we consider a more practical assumption of having an MMSE estimation of the channel gain at the receivers. In this case, minimizing the expected achievable distortion associated with each link is considered. Due to the absence of CSI at the transmitters, the Gaussian sources are encoded in a successively refinable manner and the resulting code words are transmitted over the channel using a multi-layer coding technique. Accordingly, the optimal power assignment between code layers leading to the least expected achievable distortion, under a mean-square error criterion is derived for both, the perfect and imperfect CSIR scenarios. Finally, some numerical examples are provided and it is demonstrated that the proposed method results in better performance as compared with the conventional single-layer approach, termed as outage approach.