Romanian Language Technology

Dan Tufiº, Ana-Maria Barbu * A Reversible and Reusable Morpho-Lexical Description of Romanian

As shown before, using the relational abstractions one can define implicit values for grammatical categories, that is values inherited by all the members of the respective category, unless otherwise specified. In the following example, such a generic description of the verb class is provided.

# Define lexical Verb
	<head sem pred>		=<form>
	<head sem voice>	=active/passive/reflexive
	<head agr>		==VAgr
	<head agr num>		=singular/plural
	<head agr pers>		=1/2/3
	<head agr gen>		=masculine/feminine
	<head tensed>		=no/yes
	<head prd>		=no/yes
	<head type>		=aux/main

This type-like definition specifies the main structure of the verb category and describes the valid values for some of the attributes. In the case of the attribute expressed by the <head agr> path, its value is restricted to being a feature structure of the type Vagr (verbal agreement). A refinement of the general category of verb (please note the inheritance from the Verb) would be the tensed verb:

# Define lexical VTensed(TENSE,Verform)
	!Verb
	<head sem mood>		=Verform
	<head sem tense>	=TENSE
	<head tensed>		=yes
	<head prd>		=yes
	<head sem voice>	=active/reflexive

The previous examples (Figures 2-5) have shown how the actual feature structure corresponding to a given lexical item is hidden by means of relational abstractions. If one would like for specific reasons to choose a different structuring of the attributes and their associated values then only the files containing the definitions of the relational abstractions (romanian_lexis and partially romanian_morphology) should be modified, while the word definitions (making more than 98% of the lexicon definition) would remain as such.

The word-forms analysis and generation both use a "left to right" strategy, verifying the combination restrictions appropriate to each morpheme in the composition of a word. Let us consider the entries in Figure 5, repeated below:

# Lexicon lemmas
...
	abroga	*	v/n/adj
...
# Lexicon stems
...
abrog	v	@abroga/lemmas !pred_form !prefix (none) $verb1
abrog	v	@abroga/lemmas !non_pred_form !prefix (ne/none) $verb1
abrog	n/adj	@abroga/lemmas !rol(denom/patient) !prefix (ne/none) 
		$verb_suf3 $verb_suf4 {+part} $verb_suf_part1
...

These descriptions specify that the (implicit) verbal lemma "abroga" allows for analysis and generation of the inflected forms of the verb "a abroga" as well as of some derivated nouns (abrogare) and adjectives (abrogat, abrogabil). The negative prefix "ne" is legal for non-predicative forms of the verbal forms (neabrogat, neabrogând) and for the derived nouns and adjectives (neabrogare, neabrogat, neabrogabil). As one can see in Figure 7, the word-form "abrogat" is rather ambiguous having two interpretations: it can have either a participle reading as in "am abrogat o serie de legi nedemocratice" - or an adjectival reading as in "un act normativ abrogat este inoperant". On the other hand both the nominal form "abrogarea" and the verbal one "abrogaserãm" are uniquely interpretable.

In generation (see Figure 8), the same descriptions as above would allow the system to generate all the inflected forms of the verb "a abroga", of the noun "abrogare" and of the adjectives "abrogat/ã" and "abrogabil/ã" plus the only legal negatively prefixed word-forms (the non-predicative verbal forms "neabrogând"," neabrogându-", "neabrogat" and the inflected forms of "neabrogare", "neabrogat/ã" and "neabrogabil/ã").