Northern Sámi and Norwegian/bidix

The apertium-sme-nob bidix makes heavy use of bidix pardefs. The main uses for these are:

• To change the tag format from the Giellatekno standard to the apertium standard
• To mark certain sme verbs as inherently passive/causative/reflexive
  • these markings again triggers certain transfer rules, most of them in the chunker (t1x)
• To transfer from one part of speech to another

The most complex part of the bidix is probably the verb section. A typical one looks like:

<e><p><l>vurket<s n="V"/><s n="TV"/></l><r>oppbevare<s n="vblex"/><s n="pers"/></r></p><par n="__verb"/></e>

where "pers" marks that the agent is typically animate, and __verb handles the changes in tags for person, number, temps. When translating vurken, the tags <V><IV><Ind><Prs><Sg1> are turned into <vblex><pers><pres><sg><p1> by bidix, then the transfer rules distribute the tags <vblex><pres> onto the verb lemma, creating oppbevarer (and perhaps insert a pronoun using the other tags, creating jeg oppbevarer). Additionally, the pardef handles certain derivations, so when translating vurkejuvvot, the tags <V><TV><Der3><Der_PassL><V><Inf> will turn into <vblex><pers><inf><pass>, transfer rules add <vblex><inf><pass> to the verb lemma, creating oppbevares.

However, we can also have another pardef which, in addition to the above, also adds a causative tag <caus> which is picked up by transfer:

<e><p><l>divuhit<s n="V"/><s n="TV"/></l><r>reparere<s n="vblex"/><s n="pers"/></r></p><par n="caus__verb"/></e>

Here transfer will try to make a causative construction with this verb, by prepending "la" and distributing the finite temps tag there, while making the verb infinite. Thus given divuhin, tagged <V><TV><Ind><Prt><Sg1>, bidix will output ^reparere<vblex><pers><caus><pret><sg><p1>$, and when transfer sees the verb is tagged >caus>, it creates ^la<vblex><pret>$ ^reparere<vblex><inf>$ (perhaps also inserting a pronoun as above).

Similarly, with

<e><p><l>viidánit<s n="V"/><s n="IV"/></l><r>spre<s n="vblex"/><s n="pers"/></r></p><par n="refl__verb"/></e>

we get a reflexive (seg/meg/...) appended by transfer on seeing the <refl> tag added by refl__verb.

With

<e><p><l>suovganit<s n="V"/><s n="IV"/></l><r>slite<s n="vblex"/><s n="pers"/></r></p><par n="pass__verb"/></e>

we get a "pass" tag and a passive construction, with a participle (here: bli slitt). However, with the passive, the predicate can also be an adjective, which we mark like this:

<e><p><l>viessat<s n="V"/><s n="IV"/></l><r>trøtt<s n="adj"/><s n="pers"/></r></p><par n="pass__verb"/></e>

(other parts of speech for the passive predicates are currently TODO-marked in bidix)

It's up to transfer (mainly the chunker, t1x) to make sense of and clean up these tag combinations.

Pardef	Description	Example	Notes
__verb	Regular verb transfer	vurken → (jeg) oppbevarer
pass__verb	sme verb to norwegian dynamic passive	áibat → bli forsinket	use lemma forsinke in the <r>; this pardef also works with adjectives (e.g. čuččodit translates to <r>stående<s n="adj"/><s n="pers"/></r></p><par n="pass__verb"/>, bli stående)
pstv__verb	sme verb to norwegian lexicalised passive	čoggot → samles	use lemma samles in the <r>

PlcSur__np

For all Plc-tagged proper noun lemmas in bidix, we have to have a Sur-tagged entry too. Even though "Hammerfeasta" is never used as a Sur, sme-dis.rle (and thus apertium-sme-nob.sme-nob.rlx) has a rule that can change arbitrary Plc-tagged proper nouns to Sur. So bidix has to be able to handle that.

If the translation is identical no matter whether it's Plc or Sur, we use a pardef:

<e><p><l>Isuzu<s n="N"/><s n="Prop"/></l><r>Isuzu<s n="np"/><s n="top"/></r></p><par n="PlcSur__np"/></e>

If it's not, we do like this:

<e>       <p><l>Ádjáčohkka<s n="N"/><s n="Prop"/><s n="Plc"/></l><r>Emmenesveten<s n="np"/><s n="top"/></r></p><par n="__np"/></e>
<e r="LR"><p><l>Ádjáčohkka<s n="N"/><s n="Prop"/><s n="Sur"/></l><r>Ádjáčohkka<s n="np"/><s n="top"/></r></p><par n="__np"/></e>

(since we should never change surnames in translation).

• Idea: run a huge corpus through CG with --trace, grep for the rule that changes Plc to Sur, add any such lemmas into lexc as Sur, and get rid of the whole PlcSur mess.
  • or just add Sur from http://www.census.gov/genealogy/names/ http://www.ssb.no/navn/alf/main.html http://no.wikipedia.org/wiki/Kategori:Etternavn and remove the rule