140918

• FST
• The physical meeting in Tartu
• Next meeting

tag conversion

Tag conversion is almost ready. ... coming to a window close to us. There are some points for which "make check" fails, but these poins were there earlier as well. The last results are not in svn yet, but will come in after the meeting.

remaining issues

• Overanalysis, compounding
• Extra derivation tags(?): _lt _us _ke _kene _ini _ prefix = &-&
• verb generation - could have been an issue also in the PLAMK version
• errors in noun generation (N+Sg+Nom): hernel, poisl, meli, käli ( l instead of s)

The tradition of lexicalisation of compounds ... of lexicalisation of derivated processes

Compounds are separated with '#':

• lexicalised: #
• dynamic: +Cmp#

From the North Saami analyser:

echo skuvlabusse | usme
skuvlabusse        
    skuvlabusse+N+G3+Sg+Gen  <==== the lexicalised one will be picked prior to disambiguation
    skuvla+N+Cmp/SgNom+Cmp#busse+N+G3+Sg+Acc

With middleprocessing:

echo skuvlabusse | usme | lookup2cg
"<skuvlabusse>"
         "skuvlabusse" N G3 Sem/Veh Sg Gen
         "skuvlabusse" N G3 Sem/Veh Sg Nom
         "skuvlabusse" N G3 Sem/Veh Sg Acc

In the lexc file:

Now: lemma is "skuvlabusse":
skuvlabusse+G3+Sem/Veh:skuvla#bus'se GOAHTI ;
Earlier lemma was "skuvla#busse":
skuvla#busse+G3+Sem/Veh:skuvla#bus'se GOAHTI ;

derivation boundary: » and «

• surface side: »
• tags: Der/foo Der/bar

Examples of analysing some Estonian derived forms:

tähtsusega
tähtsusega        tähtsus+N+Sg+Com
tähtsusega        tähtsus+N+Sg+Nom&ega+Adv

täielikult
täielikult        täi+N+Sg+Nom&esi+N+Sg+Nom&kult+N+Sg+Nom
täielikult        täi+N+Sg+Gen&esi+N+Sg+Nom&kult+N+Sg+Nom
täielikult        täielikult+Adv
täielikult        täielik+A+Sg+Nom&ült+Adv
täielikult        täielik+A+Sg+Abl

There is a lot of garbage.

A North Saami example:

         
tf-hsl-m0016:~ ttr000$ echo ustitlaš | usme
ustitlaš        ustitlaš+A+Sg+Nom
ustitlaš        ustitlaš+A+Attr
ustitlaš        ustit+N+Der/laš+A+Sg+Nom
ustitlaš        ustit+N+Der/laš+A+Attr

The same s: l error (probably in the two-level rules) shows up in the analysis of eli:

eli
eli        esi+N+Sg+Nom
eli        esi+N+Pl+Par

Example of compound analysis:

loodusõnnetuste loodus+N+Sg+Nom&õnn+N+Sg+Gen&tust+N+Pl+Par
loodusõnnetuste loodus+N+Sg+Nom&õnnetus+N+Pl+Gen

A relatively simple but powerful idea to try: prefer the analysis with the smallest number of compound boundaries.

• dynamic derivations: marked with the "=" sign
• dynamic compounds: marked with the "&" sign

Suggestion: change to the symbols used by GTDivvun infra: +Der for derivations, +Cmp followed by # for compounds.

• The program will be ready and e-mailed to all participants next week. (Heiki, Heli, Trond)
• Heiki will book the hotel rooms as follows:

Fran double room sun, mon
Sjur single room mon
Trond single room sun, mon, tue
Jack single room sun, mon, tue

i.e.

Fran double room sun-tue
Sjur single room mon-tue
Jack single room sun-wed
Trond single room sun-wed

Next week on fst-remaining, perhaps without Heli and Trond. Fran, Jaak and Sjur on IRC?