the mental image of this object (svalakṣaṇākāra),
which imparts its image on
the conceptually constructed object (santāna)
The order is inversed in the case of inference (and consequently of Linguistic Communication, which is a case of inference):
an āropitākāra (a superimposed image,
which appears as external), out of which,
an adhyavasita object
Therefore, in the latter case there is no connection whatsoever with the svalakṣaṇa. But, interestingly enough, the inference is not exactly like an error. Dharmottara (in Jayanta) explains that, though mistaken, it keeps some loose connection with the object. In other words, even if the object is not actually in touch with the epistemological process, the inference works because of an indirect link with it.
The above is part of what I understood during many interesting discussions during the Apoha Workshop. I am particularly grateful to a debate between Shoyu Katsura and Larry McCrea on the last day. Errors are entirely my fault.
On the Apoha Workshop (and for further links on apoha), see this post.