Automatic image captioning research has some very impressive results in the recent years. Also there are other ideas in this domain that can handle language qui

Automatic image captioning research has some very impressive results in the recent years. Also there are other ideas in this domain that can handle language quite well (e.g. visual question answering). Interpreting classes in images as symbols, it appears as if symbolic and statistical approaches can in principle be combined. Wondering whether this can somehow be transferred to learning language for automatic translation.