On today’s Morning Edition on NPR, in the story “Without Heat, Sandy Victims ['victims of the storm Sandy', not 'victims who are covered with sand'] Guard Their Homes”:
He’s living in a house that was partially flooded so it doesn’t get robbed – for a second time.
The sentence adverbial so it doesn’t get robbed … is clearly intended to modify the main clause (he’s living in a house …) — it offers a reason for this man to live in a house that was partially flooded — but some listeners probably had a moment of wondering about partially flooding the house so it doesn’t get robbed. The intended interpretation involves “high attachment” (HA), to the main clause preceding the so-adverbial, rather than “low attachment” (LA), to the relative clause within the main clause. It’s been noted again and again that LA is preferred in syntactic processing, but also noted (see here, for example) that this is only a default, with context, real-world knowledge, and discourse organization often favoring HA instead.
In the cases that people have looked at in terms of LA vs. HA, the issue is how some constituent C is parsed with respect to preceding material: is it parsed with a lower, smaller predecessor constituent B or with a higher, more inclusive predecessor A (ending in B)? Since the head word of B (was (flooded) in the hurricane example above) will of necessity be nearer to C (the so-adverbial in this example) than the head of A (is (living) in this example) is, this preference is often thought of as a preference for attachment to the nearest, but it’s the structural relationships that are key here.