Collisions are integral to the way that Ethernet works. If you consider a single Ethernet segment, working away on it's own and not connected to anything else, then any collisions that occur will propogate to all points on that segment and be detected by all the stations. In this case, the collision domain (the area/region/extent within the whole network to which collisions propogate and have effect) extends over the whole network.
If you now connect this segment to a second identical segment using a bridge, the bridge combines the two segments into a single network but isolates the two halves into two separate collision domains. This is useful because as you add more stations and/or traffic to an Ethernet segment the collision rate increases until the segment becomes unusable. Breaking the network up into multiple collision domains allows you to add more load than would be possible with just one