compounding on multiwords

[unhammer]

Currently, we only compound when a sequence is marked as unknown. A sequence is delimited by inconditionals or space, so if på togløpet. were analysed as ^på<pr>$ ^*togløpet$^.<sent>$ without compounding, we may get ^på<pr>$ ^tog<n><cmp>+løp<n>$^.<sent>$ with compounding – in this case, the compounded sequence had a space first and an inconditional after.

But this rule doesn't let us compound on multiwords that are in the dictionary, e.g. even though formel 1<n> is in dix, we get ^formel<n>$ ^1<det>$^-<guio>$^løp<n>$ as four elements instead of ^formel 1<n><cmp>+løp<n>$ – the whole string cannot be one unknown sequence ready for compounding since it contains a space.

Would it be possible to try compounding when we are reading a prefix with at least one space that ends in compound-only-L, and the following character is alphabetic? (This will happen when we've analysed formel 1-, the FST has an analysis that ends in compound-only-L, and the next character is l.) Should we only do it when the last one is dash, or can we do it in general?

Other examples where it'd be nice: "La Liga-målet" (La Liga<np> + mål<n>), "a cappella-konsertene"

---

A complication is that we still want the longest possible analysis if there is one in the dictionary, so if formel 1-løpet as a whole is in dix, we don't want to do compounding. So we can't make the decision to do compounding until we've first seen formel 1-/<compound-only-L> and then read the whole formel 1-løpet up until where regular analysis would give up, then try compound, and if that doesn't work try the regular analysis (which in this example gives four elements).