Thanks for the example – perhaps the documentation at http://moin.delph-in.net/ErgSemantics/Essence needs to be expanded to include some examples of non-equivalence.
Semantic equivalence isn’t the same as logical equivalence. For example, (3a) and (3b) are logically equivalent, meaning that they are true in exactly the same situations (assuming that, if a person hasn’t left, they’ve stayed). However, in the general case, establishing logical equivalence is hard.
(3a) Everyone left.
(3b) No one stayed.
Semantic equivalence is better understood in terms of compositionality. Semantically equivalent sentences involve the same semantic objects combined in the same ways. If we compare (1a) and (1b), only (1b) refers to entering. For example, if we compare (1a) with (1c) below, we can see that “into” doesn’t necessarily entail entering:
(1c) Marie glanced into the room.
Even if we assume that (1a) and (1b) are logically equivalent, they have composed different semantic objects. Because (1b) introduces an “entering” event into the semantics, it also means that they can be modified in diferent ways. Compare the following:
(4a) Marie suddenly danced into the room.
(4b) Marie suddenly entered the room dancing.
(4c) Marie entered the room, suddenly dancing.
Based on (1b), we have the two sentences (4b) and (4c). In (4b), the entering is sudden. This could describe a situation where no one in the room was expecting Marie to enter, and they only noticed her once she entered. Meanwhile, in (4c), the dancing is sudden. This could describe a situation where everyone could see Marie coming, but she only started dancing as she entered. However, based on (1a), it is not possible to set up this contrast, and we only have (4a). This suggests that (1a) and (1b) are not semantically equivalent.
However, it is clear that (1a) and (1b) have similar semantics, and it is certainly true that different languages have different strategies for indicating the direction and manner of motion. If there is a consistent difference between two languages (e.g. English prefers a sentence like (1a) while Hindi prefers a sentence like (1b)), it might be useful to explicitly express this fact. When using Delph-in grammars for machine translation, it is possible to use transfer rules to modify the structure of an MRS, and I think that would work in a case like this.