North Saami and Finnish

From Apertium
Revision as of 20:25, 12 June 2010 by Francis Tyers (talk | contribs)
Jump to navigation Jump to search

This page is for discussing the Northern Sámi and Finnish translator (apertium-fin-sme). Some pending things to think about:

  • How are compounds dealt with in Omorfi and in the GTSVN analysers ? Do they always split in the same places ? If not, we probably have to add those that don't as lexicalised entries in the transducers.
    • Compounds in sme and fin are similar, and we should strive at translating dynamic compounds.
  • Adding subcategories (Dem, Itg, etc.) to pronouns in Omorfi
  • Fred Karlsson's constraint grammar for Finnish has been GPL'd, and is available and undergoing conversion to CG3 here:
    • This should be converted in an Apertium-compatible manner from the start! No using reserved symbols (e.g. <, > and /)
  • How can we restrict generation of alternative forms in the Sámi generator ? In lttoolbox this is done with LR (only analyse)/RL (only generate) markings.
    • As follows: The RL forms should be marked as such in the source code. The tag for it is +Use/NG. All forms given this tag will be included in the analyser sme.fst but excluded from the generator isme.fst
  • hfst-lookup or something similar to _generate_ analyses that come in with ^ and $
  • Can we rig up SVN to pull in the twol file from GT svn directly ?
    • No we can't. SVN doesn't allow file externals. Francis Tyers 20:25, 12 June 2010 (UTC)
  • Some tags do not get replaced by the relabel script: olleet olla+V[GEN=ACT]+Pcp1+Pos+Pl+Nom
  • Sub-categorise conjunctions into CC/CS ?
  • Another possible source for paired words and sentences: Open-tran. Contains translation strings for linux software with GUIs, allows searching in any language pair, and contains Finnish and Northern Sámi. Ryan can contact them if it seems like their data would be of use.
  • Lex choice build xsl script: Add colon number on the Finnish side.
  • Find frequent multiwords, perhaps take advantage of mwetoolkit. Are there any multiword resources for Finnish existing ?
  • Generation with correct case. At the moment the North Sámi generator cannot generate words with initial caps.
  • Syntax tags should not use > and <, until these are replaced, the translator should not run the syntax section of the CG (section #5). See modes.xml file.

Comparisons of Northern Sámi and Finnish

Noun phrases

Both Northern Sámi and Finnish order noun suffixes in this way:


Possessives markers are much less common in Northern Sámi, but morphological analyzers will handle them.

Constituent order within noun phrases is similar:

Det Num Adj+ Noun

Where Det can be either a demonstrative pronoun or a pronoun denoting possession (i.e., a personal pronoun in the genitive).


Northern Sámi has 7 cases: nominative, accusative, genitive, locative, illative, comitative, essive.

  • Accusative and Genitive are often syncretic, except in some numbers and some pronouns.
  • Comitative and Essive are the same in singular and plural

Finnish has 15 cases (and several additional case-like suffixes only applied to adverbials). This is alot, here are the significant facts to avoid a string of opaque latinate terms:

  • Structural cases: 4. nominative, partitive, accusative, genitive
  • Locative cases: 6. An internal and external set (3 cases each) that show goal, location, and source.
  • Stative cases: 2. state, goal state; rarely a third - source state
  • Additional: 2 instructive/instrumental cases (with, without), 1 comitative case (plural only)

Where Finnish distinguishes internality and externality with locative and stative cases, there is no such distinction in Northern Sámi. Northern Sámi uses locative for source and location, and illative for goal. Thus, cases can roughly be transfered this way:

  • (fin) Internal Source, Internal Location, External Source, External Location → Locative
  • (fin) Internal Goal, External Goal → Illative
  • (fin) Partitive, Accusative, Genitive → AccGen

Of course, the last set ending in AccGen will have to be distinguished with certain numbers and pronouns.


Adjectives in Northern Sámi can have two separate forms depending on whether they are attributive or predicative. The attributive adjectives mostly do not agree in number with the head noun, but predicative adjectives do. Attributive adjectives do not agree in case with the head noun.

In Finnish, adjectives always agree in number and case with the head noun, and agree in number when they occur in predicates (although there is some variation as to whether or not the predicative adjective is partitive plural or nominative plural).


Tag Type Example Analysis in North Sámi Gloss
Der/inen N→Adj "muovinen" muovi+N+Der/inen+Pos+Sg+Nom plastihkas ráhkaduvvon plastihkka+n.loc build+v.pass.pp
Der/ja V→N "kirjoja" kirjoa+V+Der/ja+Sg+Nom kirjoa-ja = write-er (writer) ?
Der/lainen N→Adj "saamelainen" saame+N+Der/lainen+Pos+Sg+Nom sápmelaš -laš
Der/llinen N→Adj "kirjallinen" kirja+N+Der/llinen+Sg+Nom kirja-llinen = book-ish (literary)?
Der/minen marks deverbal nouns ?
Der/sti Adj→Adv derives an adverb from an adjective ? -ly
Der/ton N→Adj "rahaton" raha+N+Der/ton+Sg+Nom ruđaheapme ruht + -heapme

There are some cases where both a derived and a lexicalised entry might be in one analyser, but only one or the other in the other analyser. For example:

saamelainen	[LEMMA='saamelainen'][POS=ADJECTIVE][KTN=38][CMP=POS][NUM=SG][CASE=NOM]

saamelainen	saamelainen+A+Pos+Sg+Nom
saamelainen	saame+N+Der/lainen+Pos+Sg+Nom
saamelainen	saame+N+Sg+Nom#lainen+N+Sg+Nom


sápmelaš	sápmelaš+A+Sg+Nom
sápmelaš	sápmelaš+A+Attr

How to deal with this will be one of the main challenges. E.g. do we add more entries, or do we remove entries ? Is there a way to do either of those automatically ?

The reason why the sme analysis gives only the lexicalised analysis is that there is a postprocessor choosing the lexicalised one, the perl file lookup2cg. Run through the same file the fin output is compatible:

$echo saamelainen|ufin|lookup2cg
	 "saamelainen" A Pos Sg Nom


Source files
File Description Notes
apertium-sme-fin.sme-fin.dix Transfer lexicon / Bilingual dictionary
apertium-sme-fin.sme.twol Morphophonology for Sámi This file is copied as is from Giellatekno SVN. No changes should be made to the local version.
apertium-sme-fin.fin-sme.rlx Constraint Grammar for Finnish
apertium-sme-fin.fin-sme.t1x Chunker file for Finnish→Northern Sámi
Compiled and binary files
File Description Notes
fin-sme.prob Tagger HMM probability file This file needs to be trained when the CG is fully converted.
fin-sme.rlx.bin Compiled Constraint Grammar for Finnish
fin-sme.autobil.bin Compiled transfer lexicon
fin-sme.t1x.bin Compiled transfer rules These are first-stage transfer rules, mostly for chunking and local reordering.


Some thoughts on when to just adjust tagsets, transfer lexically (in the bilingual dictionary) or with transfer rules.

  • If there is a 1:1 correspondence between tags that mean the same thing in all contexts, then the tagset should be changed. For example [NEG=CON] should be just relabelled as <ConNeg> (the tag as it appears in the Sámi analysers).
  • If there is a need to insert or remove tags based on other tags, because of a difference in what is tagged then this can be done in the bilingual dictionary. For example in Finnish, adjectives in the positive comparison are tagged with A.Pos, where in Sámi the positive is unmarked A. In this case, a paradigm in the bilingual dictionary can be used to remove all <Pos> tags from adjectives in the target language.
  • If the tag changes depend on syntactic or morphological context of more than one word, then it should be done in transfer rules.

See also