Version classiqueVersion mobile

Formaliser les langues avec l’ordinateur : de INTEX à Nooj

Svetla Koeva
Denis Maurel
Max Silberztein

Troisieme partie. Analyse syntaxique

19. Word Sense Disambiguation for Purposes of Machine Translation – the nature of Bulgarian clitics

Svetla Koeva et Svetlozara Lesseva

Texte intégral

I. Word sense disambiguation for the purposes of machine translation

1Translation (especially automatic one) is a very difficult task due to a number of reasons such as cultural differences, differences in the inventory of concepts shared by the members of a society, as well as in the concepts’ lexicalization, and even in respect to grammatical structures. The development of an automatic knowledge-based translation system generally requires morphological analysis and morphological disambiguation tools – a Part-Of-Speech (POS) tagger, domain specific and general translation lexicons for the source language, transfer rules and a morphological generator for the target language. In this process a word sense disambiguation module decides on the correct concept to which each translation unit belongs.

2Word Sense Disambiguation (WSD) consists of selecting the semantic sense of a word (either single or compound) among all the possible senses given by a dictionary, as well as taking into account the context in which this word appears.

3It is well-known that the meaning of a word in a particular usage can only be determined in dependence of its context. The problem is that words often have more than one meaning, sometimes very close ones. Word Sense Disambiguation is well-known as one of the most difficult problems in the field of natural language processing (Ide at all 1998 ; Resnik at all 2000). The difficulties stem from several sources, including the lack of means to formalize the properties of context that characterize the usage of an ambiguous word in a given sense, the lack of a standard (and possibly exhaustive) sense inventory, and the subjectivity of the human evaluation of such algorithms. The ambiguity of Bulgarian clitics and their word sense disambiguation is considered in terms of Finite State Transducers (FSTs).

II. A typology of Bulgarian clitics

4A clitic is usually defined as a morpheme that has the syntactic characteristics of a word, but shows evidence of being phonologically bound to another word. Clitics are the object of much discussion in linguistics ; however their definition is still not clear enough. In this paper we are not going to define a coherent and unified concept of clitics and to give a clear definition of what clitics are for all categories of clitics in different languages. Instead we present a typology of Bulgarian clitics (our target units for sense disambiguation being among them) according to several distinguishing features :

  • Not receiving a full word-level stress – generally, Bulgarian clitics fail to receive a stress and are treated as phonologically bound with an adjacent word for stress purposes regardless of the fact that they are written as separate words. Nevertheless there are sentence positions in which some classes of Bulgarian clitics can be stressed – e.g. pronominal clitics after some complementizers ;
  • Subject of Wackernagel cliticization – generally, Bulgarian clitics occur after the first constituent although some Bulgarian clitics may begin the sentence – e.g. grammatical particle forming future tense ;
  • Degree of fixing of the clitics’ position in the sentence – some clitics’ word order is related only to the position of the stress sed word they belong to, the order of other clitics depends on more factors – e.g. the dependency of word order between negative, interrogative and pronominal clitics with respect to the verb ;
  • Attachment to a particular lexical class – Bulgarian clitics (even focused) cannot be integrated into the standard discourse without being bound to some other word. The syntactic categories of those words are very different – nouns, adjectives, verbs, pronouns, numerals or adverbs, in fact all open classes ;
  • Expressing meaning – like morphemes Bulgarian clitics can express either grammatical or lexical meaning ;
  • Difference in part of speech – Bulgarian clitics belong to closed classes like pronouns, particles, prepositions, auxiliary verbs, and conjunctions). The more detailed classification separates them into :
    • Pronouns – possessive, reflexive, reciprocal ;
    • Particles :
    • Lexical – reflexive, reciprocal, third personal reflexive, third personal dative reflexive ; impersonal reflexive, impersonal dative reflexive ;
    • Morphological – negative, positive, interrogative, future, appellative, dative ethic ;
    • Syntactical – optative, impersonal optative, passive, impersonal passive ;
    • Auxiliary and copula verbs ;
    • Prepositions ;
    • Conjunctions – coordinative, subordinate.
  • Functioning at different syntactic level – either phrase or clause. The difference in part of speech reflects the difference in clitics’ function in the sentence.

III. Clitics’ ambiguity

5Some clitics in Bulgarian are graphically identical and thus semantically ambiguous. The focus of our investigation is the semantic interpretation of the Bulgarian se / si clitics – two of the most ambiguous words that are traditionally defined as clitics. As it is shown below POS ambiguity of the examined words is shared between pronouns, particles and verbs.

6Bulgarian pronominal clitics se / si are semantically bound by a word (a subject) in the sentence. There are some other peculiarities that reflect pronominal clitics only :

  • The strictly defined position in relation to the stressed word – we assume that words entirely depending on the position of the stressed word (personal pronominal clitics in some positions can be focused) are clitics regardless of their function. On the other hand, the notion of strictly defined position is not an absolute one : the short reflexive possessive pronoun is normally attached to the determiner-bearing word in the noun phrase. However, with NPs selected as internal arguments of verbs the short possessive pronoun can move out of the NP and in this case it respects the rules which VP internal short pronouns and particles obey.
  • The difference in semantic function of pronominal clitics – they can be arguments and non-arguments.
  • The difference in the class of nouns to which pronominal clitics refer – abstract, concrete, animate, non-animate, person, etc.
  • The scope of syntactic phenomena that can be observed with clitics – generally, pronominal clitics can be doubled with a non-clitic alternant and are subject to diatheses – reflexive, reciprocal, passive, optative, impersonal).
  • The difference in syntactic function – Bulgarian personal and reflexive pronominal clitics can be objects or adjuncts co-referent with the clause subject, Bulgarian possessive personal and reflexive clitics can be only noun modifiers, and Bulgarian reciprocal clitics can be objects or adjuncts.

7Particles are considered clitics regardless of whether they are obligatory (lexical particles) or non-obligatory – stylistic, emotive, etc. particles. Particles are not fixed in position with respect to the verb and move in conformity to certain rules concerning the movement of all clitics in the environment of a verb. The examined particles se / si cannot be focused in any position in the sentence.

Table 1 : Classification the different senses of se / si clitics

Table 1 : Classification the different senses of se / si clitics

8Se / Si clitics are an example of grammatically and / or sense ambiguous words that function either within the VP, the PartP (participle phrase), or the NP. With a view to their frequency and functionality se / si’s ambiguity poses an interesting task to automatic translation oriented research with respect to the possibility for disambiguation rules drawing on the immediate context. As it is shown in Table 1 si can be a pronoun – personal (reflexive or reciprocal) or possessive, a lexical particle and a morphological particle. Se can be a pronoun (reflexive or reciprocal), a lexical particle, with which a compound word is built, or a syntactic particle that forms the verbal diatheses.

1. Reflexive, possessive reflexive and reciprocal pronouns

9Reflexive interpretation is possible if the verb takes at least two arguments (subject and complement) specified as animate and one and the same being can be their lexical projection. Bulgarian reflexive personal pronominal clitics (accusative – se and dative – si) are short forms of the Bulgarian reflexive pronouns, which substitute VP internal animate constituents (respectively NP or PP) and are semantically bound by the subject of the clause. Not every (reflexive) PP constituent is subject to clitization – only those with dative designation.

10In addition to the personal reflexive pronouns, there is also a possessive version in Bulgarian. These pronouns are used to refer to a possessor identical with the subject of a sentence. Again, since the clitic is reflexive in this sense, it cannot occur in the subject NP but only inside other NPs in a clause. Bulgarian reflexive possessive clitic si substitutes all forms of Bulgarian reflexive possessive pronouns and appears only in definite phrases.

11Reciprocal interpretation is possible if the verb takes at least two arguments (subject and complement) specified as animate. Those verbs are subject to a syntactical transformation after which the reciprocal subject incorporates the initial arguments and an obligatory reciprocal operator is involved in a complement position – the reciprocal pronominal clitic (accusative – se and dative – si). As reciprocal meaning is possible only with plural verb forms, reciprocal pronouns express an interchangeable or mutual action or relationship and their antecedent is interpretable only as the subject of the clause. Bulgarian contains also a set of reciprocal adjunct phrases corresponding to English each other : edin (preposition) drug.

2. Lexical particles

12There are several types of verbs belonging to certain classes for which the form se / si is a lexicalized particle, part of the lexical form of the verb (Koeva 2004a). Such verbs fall into the following classes :

13Reflexiva tantum se verbs are personal verbs (with animate subject) that are lexically intransitive, 3rd personal (with non animate subject) intransitive (without any complements) verbs and corresponding impersonal intransitive verbs.

14Reflexiva tantum si verbs are personal transitive or intransitive verbs (with animate subject).

15Reflexiva dativa tantum se verbs are called intransitive verbs with non animate subject in 3rd person singular or plural and with an obligatory object, expressed by the short dative form of the personal pronouns and corresponding impersonal intransitive verbs.

16The obligatory lexical "reflexive" particles se / si with which the above listed verbs are built do not have true reflexive meaning and it is considered as a lexical part of the verb.

17Reciproca tantum se / si are personal intransitive verbs with animate subject which are constantly lexically reciprocal – the lexical reciprocity is expressed in the explication of two arguments with equivalent and symmetric relations to the predicate. The reciprocal se / si not have the status of a reciprocal pronoun but of a word-building lexical particle.

3. Dative ethic

18Traditionally, the term refers to non-obligatory particles originated from the short dative pronominal clitics. The one coming from the short reflexive personal pronoun si is widely used in Bulgarian. The dative ethic functions within the VP and is used for stylistic rather than syntactic reasons.

4. Syntactical particles

19Syntactical particle se is used in syntactical transformations in which the number of arguments and / or their syntactic realization is changed. We can observe the following cases in Bulgarian (Koeva 2004a) :

20The true optatives are intransitive 3rd person and impersonal verbs that do not describe the real situation but their semantics expresses willingness to do the action. The indicators for the transformation are the limited paradigm of the category person, imperfect aspect, obligatory dative personal clitics and obligatory lexical optative particle se.

21The true se-passives are intransitive 3rd person and impersonal verbs that express the transformation of subject – predicate relation to static or relative predicative attribute. The markers for the transformation are the limited paradigm of the category person, imperfect aspect for impersonal verbs, optional by-phrase (mostly with generic meaning) and obligatory lexical passive particle se.

22We are interested in the resolving of sense ambiguity inside the word classes : possessive, reflexive or reciprocal meaning of the pronoun si, reflexive or reciprocal meaning of the pronoun se, reflexive, reciprocal or dative ethic meaning of the lexical particle si, reflexive, reciprocal, third person reflexive, third person dative reflexive ; impersonal reflexive or impersonal dative reflexive meaning of the lexical particle se, and optative, impersonal optative, passive, impersonal passive particle se.

IV. Corpora methodology

23Our approach is based on using a tagged corpus for several purposes : for extracting certain contextual information, enhancing dictionaries with grammatical and semantic features and finally, providing recall and precision evaluation for the rules formulated on the ground of the analysis of the derived contextual information. The test corpus is a POS tagged and hand disambiguated structured corpus of 150 000 words extracted from a 1 mln. word corpus designed after the Brown methodology. The appropriate ambiguity tags are derived from the Bulgarian Grammar Dictionary (BGD) (Koeva 1998) and were further enhanced to the end of capturing tokens not included in the dictionary. The occurrences of a given graphic word, in this particular case a graphic word that is a se / siclitic in at least one of its senses, were extracted from the Bulgarian POS disambiguated corpus into several sub-corpora corresponding to the different senses, identified by the different tags assigned in the process of POS disambiguation.

24After that the resulting sub-corpora were used for linguistic analysis of the environments of the linguistic items in question. The investigation of the context lead both to the formulation of rules based on the syntactic properties of the environment and to establishing certain dependencies stemming from the semantic properties of the context, even more taking into consideration the fact that the syntactic contexts of the linguistic items dealt with here, often coincide (the items are in complementary distribution depending on the semantic context). These were then used in determining and enhancing the DELA dictionaries (Silberztein 1993) to the end of delimiting the scope of the rules. The DELAF and DELACF dictionaries are derived from the Bulgarian Grammatical Dictionary which comprises more than 1 000 000 word forms. The DELACF dictionaries are subsets of the BGD and identify the verbs and the proclitics that are part of the lexical meaning, thus resolving part of the ambiguous senses on the pre-processing level. The detailed dictionaries has facilitated the formulation of rules represented as FSTs that conform to the following criteria – being necessary and sufficient to capture only a specific sense where possible, and being broad enough in scope to yield all the possible interpretations in a context where the different interpretations are identifiable only in a larger grammatical or in the extra-linguistic context.

V. Sense disambiguation rules for lexical particles

25The context sensitive rules for lexical particles describe the word order dependencies between target clitics, particular types of verbs specified in the DELA dictionaries, dative pronominal clitics and the interrogative particle. We have formulated the following rules :

26Se is interpreted as a lexical reflexive (reciprocal) (third personal reflexive) (impersonal reflexive) particle if it :

  • immediately precedes a verb lexically marked as constantly reflexive (reciprocal) (third personal reflexive) (impersonal reflexive) ;
  • immediately follows a verb lexically marked as constantly reflexive (reciprocal) (third personal reflexive) (impersonal reflexive) ;
  • immediately follows the sequence of the constantly reflexive (reciprocal) (third personal reflexive) (impersonal reflexive) verb and the interrogative particle li.

27Si is interpreted as a lexical reflexive (reciprocal) particle if it :

  • immediately precedes a verb lexically marked as constantly reflexive (reciprocal) ;
  • immediately precedes the sequence of an accusative clitic and a verb lexically marked as constantly reflexive (reciprocal) ;
  • immediately follows a verb lexically marked as constantly reflexive (reciprocal) ;
  • immediately follows the sequence of the constantly reflexive (reciprocal) verb and the interrogative particle li.

28Se is interpreted as a lexical third person dative reflexive (impersonal dative reflexive) particle if it :

  • immediately follows dative pronominal clitic and immediately precedes a verb lexically marked as constantly third personal dative reflexive (impersonal dative reflexive) ;
  • immediately follows the sequences of a verb lexically marked as constantly third personal dative reflexive (impersonal dative reflexive) and dative pronominal clitic ;
  • immediately follows the sequences of a verb lexically marked as constantly third personal dative reflexive (impersonal dative reflexive), dative pronominal clitic ; and interrogative particle li.

29Part of those rules (where the word order variations are limited are presented in DELACF dictionaries, for the rest word sense disambiguation rules FSTs are formulated.

VI. Sense disambiguation rules for the rest of target clitics

1. Relevant grammatical features

30With a view to disambiguation, in this paper a classification of verbs is outlined according to the number and the syntactic realization of their complements :

31Intransitive verbs are verbs that do not take any internal constituent – lie, sit, sleep, shine.

32We consider transitive verbs that take one internal constituent, irrelevant of its syntactic realization (NP, PP) :

33• NP-transitive verbs are those that are traditionally classified as transitive : V [NP – NP]

341. Tya yade [NP torta].

35(She eats cake.)

36• PP-transitive verbs are those that take a complement realized as a PP : V [NP – PP]

372. Toy dosazhda [PP na Maria].

38(He bothers Maria.)

39Ditransitive verbs are considered verbs that take two internal constituents. In Bulgarian only one of these complements (exception are NP small clauses which are not relevant here) may be realized as a NP.

40• NP-PP-ditransitive verbs : [NP – NP PP]

413. Maykata dade [PP na decata] [NP yabalki].

42(The mother gave her children apples.)

43• PP-PP-ditransitive verbs : [NP – PP PP]

444. Blagodarih [PP na studentite] [PP za razbiraneto].

45(I thanked the students for their understanding.)

46Since in a number of contexts it is difficult to rely only on the verbs classification, several other grammatical criteria for the clitics se / si distinguishing are employed :

  • The definiteness of the relevant NPs ;
  • The person and number of the verb form the clitics belong to (Table 2) ;
  • The aspect of the verb the clitics belong to (Table 3) ; The selective restriction for subject involved (Table 3).

47The selective restriction for complements involved (Table 3).

Table 2 : Distribution of pronominal and syntactic se / si clitics according to the verb person and number

Table 2 : Distribution of pronominal and syntactic se / si clitics according to the verb person and number

Table 3. Distribution of the word sense distinguishing features

Table 3. Distribution of the word sense distinguishing features

48We show that the combination of some of the criteria for clitics’ distinguishing listed above can be used in context rules in order to solve clitics’ ambiguity among word classes. The dependencies of immediate word order that are in strong connection to the type of clitic are registered and the grammatical features of the adjacent context that are relevant for the word sense disambiguation are displayed.

2. Reflexive versus reciprocal pronouns se and si

49Personal transitive and ditransitive verbs which take at least two animate arguments (subject and complement) are subject to the reflexive transformation. The reflexiva tantum and reciproca tantum verbs are also productive for the reflexivization if they fulfill the respective syntactical and semantical requirements. The Bulgarian reflexive pronouns se / si are the accusative / dative clitics immediately adjacent to the verb and substitute the corresponding full-form personal reflexive pronoun if this is admissible.

Initial predicate : Selective restrictions :
[NP1 – NP2 (PP)] NP1 = animate (person)
[NP1 – (NP) PP1] NP2 = animate (person)
[NP1 – PP1 (PP2)] PP1 = animate (person)

50Transformed reflexive predicate :

[[NP1i → [V ← [sei /sii]]] se=NP2 si=PP1

51Co-indexation shows that the two constituents could be considered as one and the same being (reflexive meaning).

52Personal transitive and ditransitive verbs which take at least two animate arguments (subject and complement) are subject also to the reciprocal transformation. The reflexiva tantum and reciproca tantum verbs have to be considered also if they respect the requirements for the expression of the reciprocal meaning. After the reciprocal transformation a reciprocal pronominal clitic se / si takes place in a complement position immediately adjacent to the verb.

Initial predicate : Selective restrictions :
[NP1 – NP2 (PP)] NP1 = animate (person)
[NP1 – (NP) PP1] NP2 = animate (person)
[NP1 – PP1 (PP)] PP1 = animate (person)

53Transformed reciprocal predicate :

[[NP1 + NP2 | PP1] → [V ← [se /si]]]

54The reflexive and reciprocal se pronoun is a clitic that occurs with V NP transitive verbs regardless of the fact that there might be other PP complements. Respectively, the reflexive and reciprocal sipronoun can be found with transitive PP and ditransitive NP PP verbs that yield reflexive and reciprocal interpretation. Moreover NP PP verbs can take either se or si clitics if the respective constituents are both animate and both allow reflexive or reciprocal interpretation.

55One of the main factors playing role in reflexive versus reciprocal pronouns disambiguation is verb’s number – the only meaning of se / si clitics that occur with singular verb forms is reflexive personal pronoun, respectively the sense of se / si clitic with plural verb forms is ambiguous – either reflexive or reciprocal. The Graph 1 below represents the se clitics ambiguity in plural (reflexive {PRO+REF : zt : at} or reciprocal {PRO+REC : zt : at}) where the word order dependencies in the immediate context are taken into consideration – the place of the interrogative particle li, the position of the auxiliary verbs, past participles, and dative ethic clitics, if any. The FST for the si pronoun adjacent to the plural verb differs only in fact that accusative clitics may appear instead of dative which reflects to the word order, too.

56The adjunctive reciprocal phrase "edin-preposition-drug" (each-preposition-other) has two functions. It eliminates the reflexive reading of the sentence whenever it is possible, as well as it restores the original argument structure of the corresponding verb. That is why the preposition in the reciprocal phrase depends on the preposition with which the respective argument in the sentence is introduced and the structure of the reciprocal phrase is an exact copy of the sentence structure without reciprocal meaning. Thus when the reciprocal phrase is explicit in the sentence it eliminates ambiguity and only the reciprocal meaning of the se / si pronominal clitic is licensed.

Graph 1 : Bulgarian se pronominal clitic in plural

Graph 1 : Bulgarian se pronominal clitic in plural

57The other ambiguity restriction is that the NP PP verbs with both NP and PP complements can be used only with the reflexive sepronominal interpretation in plural, if the third animate PP constituent is explicit which impedes the option for reciprocal meaning – thus the mutual interaction of the other two NP animate constituents.

58Apart from these cases, certain ditransitive verbs impose semantic restrictions which block one interpretation or another. Two groups can be outlined in relation to this :1) verbs that do select an animate constituent that do not allow reflexive interpretation, that is the argument should necessarily have a referent different from the subject. Nevertheless, such verbs have a reciprocal interpretation in the plural (5.a., 5.b.), for example verbs with such semantic features that form a coherent semantic class are some verbs of distributing, allocating, etc ; 2) verbs that have only reflexive interpretation and block the reciprocal interpretation in the plural – that is they do not select an animate complement that may be interpretable as other than the subject, (6.a., 6.b.).

5.a. Razpredelihme si zadylzheniyata.
(We divided the responsibilities among each other.)

5.b. *Razpredelihme si zadylzheniyata mezhdu sebe si
(We divided the responsibilities among ourselves.)

6.a. Toy si zasluzhi nagradite.
(He won prizes for himself.)

6.b. *Te si zasluzhiha nagradite edin na drug
(They deserved the prizes for each other.)

59The verbs that impose such restrictions on the reflexive / reciprocal interpretations (in both cases the possessive interpolation is possible) are handled in the dictionary associating them with the relevant restrictions. The exhaustive description of the semantic classes and grammatical types to which these verbs belong is a large-scale task involving the creation of a comprehensive syntactic dictionary of the Bulgarian verbs – already under way (Koeva 2004b).

3. Short reflexive possessive pronoun si versus reflexive and reciprocal pronouns si

60The main peculiarity of the possessive clitic is that it normally occupies the position immediately after the determiner-bearing word in the NP. Determinacy can either be explicit – grammatically expressed by a word form incorporating the definite article, or inherent in the semantic structure of certain classes of words, such as demonstrative pronouns (an exception is the group of kinship nouns, which attach possessive clitic only in case they do not receive grammatical marker for determinacy, because they express it semantically). In an NP that has a branched structure (that is it has at least one more modifier except si) the interpretation of si as possessive pronoun {PRO+POSS : zt} relies on the immediate left and right context since the defined modifier categories are to the left, while the N head is to the right of the clitic. The group of the determiner-bearing words includes categories of modifiers and determiners / premodifiers which embrace any of the following categories marked as X below – categories incorporating the definite article : adjectives, ordinal numerals, cardinal numerals, indefinite pronouns denoting number and quantity, negative pronouns denoting number and quantity, interrogative pronouns denoting number and quantity, generalizing pronouns denoting number and quantity ; inherently definite categories : demonstrative pronouns denoting things and persons. The categories that are placed to the right of the clitic are the ones enumerated and marked below as Y : adjectives, numerals, indefinite pronouns denoting attributes, negative pronouns denoting attributes, interrogative pronouns denoting attributes, generalizing pronouns denoting attributes, as well as the head noun itself (1).

(1) X+det si {PRO+POSS : zt} Y*-det {N-det | N-det (kinship)}

61When the noun is the only constituent of the NP (besides the si) and fulfils the condition for determinacy, the following si is unambiguously a possessive clitic if the noun is preceded by a verb form. The right context in this case is irrelevant for disambiguation (2).

(2) V {N+det | N-det (kinship)} si {PRO+POSS : zt}

62When there is a definite noun, including definite kinship nouns, in the left context of si and any category other than a verb or accusative personal pronominal clitics, the target si is the possessive pronoun :

(3) {N+det|N+det (kinship)}si{PRO+POSS : zt} {< ! V >| ! PRO+PER : zt}

63Apart from these cases, there are verbs that favor only the possessive. These verbs can be generally classified as NP-transitive, as they obligatorily select an NP complement, as well as other complements that are not relevant to the interpretation of si. An accusative personal pronominal clitic can double the whole NP also which is formulated in (4) and (5). It must be noted that at the same time these verbs can assign the reflexive or reciprocal interpretation to the whole NP complement (substituted respectively by the reflexive and the reciprocal pronoun se).

(4) {N+det|N+det (kinship)}si
{PRO+POSS : zt} (<PRO+PER : zt>) <V – NP>

(5) <V – NP>si
{PRO+POSS : zt} (<PRO+PER : zt>){NP+det|N+det (kinship)}

64On the contrary, the adjacency of ditransitive NP PP verb and si can be ambiguous since si may be VP internal constituent (a pronoun or a particle) or may belong to the immediate definite NP. In the environment of verbs the short reflexive possessive pronoun can precede the noun phrase it is part of instead of following the determiner-bearing lexical category in the NP. As a result it may occupy the same linear position as the short reflexive (reciprocal) pronoun. Actually, often a distinction between the interpretations is retrievable in a larger grammatical or in the extra-linguistic context, and even in many cases the ambiguity can not be solved.

7.a. Kupih si sinyata roklya.
(I bought myself the blue dress. / I bought my blue dress.)

7.b. Kupih sinyata si roklya sled chervenata.
(I bought my blue dress after the red one.)

65Thus the following sequences (6) and (7) allow two pronominal interpretations – the reflexive personal pronoun or the reflexive possessive pronoun with singular verbs and respectively – three interpretations in plural (plus the reciprocal one) – (8) and (9). When the pronoun follows the verb form the question-forming particle li may intervene between the two.

(6) V+sg (li) si {PRO+REF : zt | PRO+PSS :zt} NP+def

(7) N+def si {PRO+REF : zt | PRO+PSS : zt} V+sg

(8) V+pl (li) si {PRO+REF : zt | PRO+PSS : zt | PRO+REC : zt} NP+def

(9) N+def si {PRO+REF : zt | PRO+PSS : zt | PRO+REC : zt} V+pl

66When the definite NP appears before the verb form the possessive clitic either projects NP internally, or if the NP does not have any other modifiers except the clitic it appears between the N and the verb form, thus giving rise to ambiguity (being an immediate NP constituent (possessive) or NP substitute (reflexive, reciprocal) – (7) and (9). Paths 10 to 18 at the Graph 2 below present the ambiguity cases in singular taking into account the word order dependencies in the immediate context – the place of the interrogative particle li, the position of the verb forms, auxiliary verbs, and past participles.

Graph 2. Possessive si disambiguation rules and ambiguity cases in singular

Graph 2. Possessive si disambiguation rules and ambiguity cases in singular

67When the verb form is in the singular and it is followed / or preceded by si and any category other than definite NP appears in the immediate context the possible interpretation of si is the reflexive personal pronoun (in plural the reciprocal reading is also admissible). The possessive reading of the clitic is not licensed because it can only be a constituent of definite NPs and any phrase intervening between the NP and si, blocks the reflexive possessive interpretation. Reversely stated, the pronominal interpretation of si with these verbs is reflexive pronoun, unless the verb form or si is immediately followed by an NPdef, or si is preceded by an Ndef – paths 1 to 9 at the Graph 2.

68The ambiguities with NP+def following the verb form may be restricted by considering the structure of the definite NP : if it has as a constituent a full form determined possessive pronoun (e.g. moyat (my), tvoyat (your), negoviyat (his), neiniyat (her)) or clitical (mi, ti, mu, ji) or a demonstrative pronoun denoting persons and things, it will block the possessive interpretation : the form si preceding the NP cannot belong to it, since another possessive pronoun is already projected in the NP or the demonstrative occupies the position reserved for it – this fact is also considered in paths 1-9 at the Graph 2.

4. Pronominal clitic se versus syntactic particles se

69The se-passive verbs are intransitive verbs with limited forms to third person singular and plural which correspond with NP transitive and ditransitive personal verbs with non-animate object (the generic usage is a special case).

8.a. Hudozhnikat risuva karinata (The artist is painting, )

8.b. Kartinata se rusuva (ot hidozhnik). (The picture is painted.)

9.a. Slanceto raztopi leda. (The sun thawed the ice.)

9.b. Ledat se raztopi ot slanceto. (The ice was thawed)

70The passive meaning is realized with the integration of a passive operator adjacent to the verb – a passive particle se in the complement position. The correspondence is exemplified as follows :

Initial predicate : Selective restrictions :
[NP1 – NP2 (PP)] NP1 = non-animate |
animate (agentive)
NP2 = non-animate

71Transformed se-passive predicate :

[[NP2] → [[V ← [se]] ← [(PP = NP1)]]]

72The passive constructions, corresponding with a transitive verb with an animate object, exclude the "se-morphology". Thus part of the ambiguity with the reflexive and reciprocal se-verbs is avoided :

10.a. Ivan bie Petar. (Ivan is beating Peter.)

10.b. Petar e nabit (beshe nabit) ot Ivan. (Peter was beaten by Ivan.)

10.c. *Petar se nabi ot Ivan. (Peter was beaten by Ivan.)

73An exception is the generic usage of passive voice, where both an animate subject and an object are allowed.

11. Dete se obicha cyal zhivot. (A child is loved for
a life time.)

74The impersonal se-passive corresponds to NP transitive and NP PP ditransitive imperfective verbs with animate subject and NP complement or PP transitive and PP PP ditransitive imperfective verbs with animate subject.

12.a. Ivan aranzhira cvetyata. (Ivan arranged the flowers.)

12.b. Aranzhira se s vkus. (It is tastefully arranged.)

13.a. Ivan blagodari na Petar. (Ivan tanks Petar.)

13.b. Za tova ne se blagodari. (This thing is not to be thanked.)

75The transformation can be represented with the following main requirements :

Initial predicate : Selective restrictions :
[NP1 – NP2 (PP)] NP1 = animate
[NP1 – ] NP2 = non-animate | animate
V = imperfect aspect

76Transformed impersonal se-passive predicate :

[V ← [se]]

77Through the transformation an impersonal passive operator – the syntactic particle se is introduced. On the contrary with the se-passive, the impersonal passive corresponds to verbs that are imperfective only and could take an animate object.

14. Tuk ne se chaka. (Waiting is not allowed here.)

Graph 3 : The interpretation of se with third person singular verbs

Graph 3 : The interpretation of se with third person singular verbs

78If we consider the competition between reflexive, reciprocal and passive meanings the following statements can be formulated : first and second person singular verbs can take only reflexive pronoun se ; third person singular perfective verbs can take either reflexive pronoun se or generic passive particle se ; third person singular imperfective NP (PP) verbs can be found with either reflexive pronoun se, generic passive particle se or impersonal particle se ; third person singular imperfective (PP) (PP) verbs can take only with impersonal particle se – paths 1-4 in Graph 3. In the plural the distribution is as follows : first and second person plural NP (PP) verbs can take both reflexive and reciprocal meanings ; first and second person plural NP PP verbs can take only reflexive meaning ; third person verbs can take reflexive pronoun se, reciprocal pronoun se or generic passive particle se.

79The optative verbs are derived from personal NP transitive or NP PP ditransitive imperfective verbs that take two arguments : agentive subject and non-animate complement.

15.a. Ivan pie mlyako. (Ivan drinks milk.)

15.b. Pie mu se mlyako. (He feels like drinking some milk.)

Initial predicate : Selective restrictions :
[NP1 – NP2 (PP)] NP1 = animate (agentive)
V = imperfective aspect NP2 = non-animate

80Transformed optative predicate :

[[NP2] → [[V ← [se]] ← [PP = NP1]]]

81The subject of transitive constructions becomes an indirect object in their optative counterparts. This indirect object is realized by a dative personal pronominal clitic which in fact expresses the category of person in the sentence. The object of transitive active verbs becomes subject in optatives – there are some similarities with the passive constructions. The optatives take only non-animate subjects projected as the NP complement of the source predicate. The transformation includes introduction of an optative operator – an optative particle se in the complement position of the initial predicate.

82The impersonal optatives correspond to transitive NP, transitive PP, ditransitive and intransitive verbs with animate agentive subject.

16.a. Ne hodya barzo. (I am not walking fast.)

16.b. Ne mi se hodi barzo. (I do not feel like walking fast.)

Initial predicate : Selective restrictions :
[NP1 – NP2 (PP)] NP1 = animate (agentive)
[NP1 – (PP) (PP)] NP2 = non-animate | animate
V = imperfective aspect

83Transformed impersonal optative predicate :

[[V [se]] ← [PP = NP1]]

84The constantly reciprocal verbs fulfill the optative requirements also, and since they are intransitive they form only impersonal optative diathesis.

85The agentive argument is transformed and is expressed by personal dative clitic bearing the person features. If the initial verb is NP transitive the argument corresponding to the direct object is absorbed. The initial predicate is transformed to impersonal optative with an impersonal optative operator – the impersonal optative particle se in the complement position.

86As the optative meaning is expressed only in the third person the additional word sense disambiguation rules are formulated considering only these cases. The dative clitic is obligatory with optative meaning, thus in sequences with explicit dative clitic adjacent to the NP transitive (singular and plural) verbs se can be only optative syntactic particle (third person {PC+OPT} or impersonal third person {PC+IMP+OPT}) – rules (10) to (13).

(10) {PRO+PER : zt} se {PC+OPT } | {PC+IMP+OPT } V – NP – sg

(11) V – NP – sg (li) se {PC+OPT} | {PC+IMP+OPT}{PRO+PER : zt}

(12) {PRO+PER : zt} se {PC+OPT} V – NP – pl

(13) V – NP – pl (li) se {PC+OPT }{PRO+PER : zt}

87The immediate explicitly expressed indefinite inanimate NP restricts readings to third person optative particle – (14) and (15) :

(14) NP-non-animate indef {PRO+PER : zt} se {PC+OPT} V – NP – sg

(15) V – NP – sg (li) {PRO+PER : zt} se {PC+OPT}NP-non-animate indef

88Third person singular imperfective (PP) (PP) verbs can be found only with impersonal optative particle se.

(16) {PRO+PER : zt} se {PC+IMP+OPT} V – (PP) – (PP) – sg

(17) V – (PP) – (PP) – sg (li) se {PC+IMP+OPT }{PRO+PER : zt}

89The statement that the NP PP verbs with both NP and PP complements can be used with the reflexive rather than reciprocal se pronominal interpretation in plural can be made stronger because those verbs do not allow optative interpretation either. The reasons are similar – the explicit PP blokes the optative meaning.

(18) {PRO+PER : zt} se (PRO+REF : za} V – NP – PP

(19) V – NP – PP (li) se (PRO+REF : za}{PRO+PER : zt}

5. Dative ethic

90It is difficult to delineate strictly the contexts where the dative ethic is found and the connotation it gives. It has been observed that the particle is used predominantly in colloquial speech and is rarely found otherwise. It is most often found with intransitive verbs such as lezha (lie), spya (sleep) (17.a.) that select only one argument, projected as the subject, as well as with se-reflexiva tantum, accusativa tantum, dativa tantum, reflexiva dativa tantum, intransitiva tantum and se-reciproca tantum (17.b.) verbs.

17.a. Deteto (si) spi. (The child is sleeping.)

17.b. Tya (si) se smee prosto taka. (She laughs just like that.)

91It is the only possible interpretation with NP-transitive verbs selecting an animate NP projected as a proper noun or animate kinship noun.

18.a. Mnogo (si) obicham mama. (I love my mother very much.)

92It is found with NP – PP-ditransitive verbs (with reflexive and / or reciprocal interpretation) used without an explicit internal NP complement (19.a.) and rarely with explicit NP complement, as well as with PP-transitive verbs. The cases where the dative ethic is in complementary distribution with other interpretations in contexts discussed so far, are accounted for in the graphs for the respective interpretations as {PC} ; illustrated below : reflexive personal (19.a.), possessive pronoun (19.b.) and as the reciprocal in the plural (19.c.) :

19.a. Momicheto si risuva.
(The girl is drawing. / The girl is drawing to herself.)

19.b. Tya si gi risuva.
(She draws them to herself. / She draws them (her drawings).)

19.c. Ostavi momchetata da si govoryat.
(Let the boys talk (to themselves/to each other). / Let the boys talk.)

93With the NP-transitive verbs and verbs behaving alike (affording only the possessive reflexive interpretation of si) the dative ethic occurs in the contexts with non-explicit NP (20.a.) or indefinite / definite NPs (20.b.).

20.a. Tya si reshe po cyal den. (She combs all day.)

20.b. Toy si brysne klienti po cyal den. (He shaves clients all day.)

94The competition between the dative ethic and the rest of the relevant interpretations for the different verb classes is not identifiable in a context narrower than the discourse, and the dative ethic is therefore considered where possible, although it has marginal usage.

VI. Future directions

95Outside the scope of the survey remain some cases the most important being : reflexive and reciprocal in the plural, general passive and reflexive in the 3rd person singular and plural, reflexive and possessive pronoun with adjacent definite NP which should be approached in a more elaborate classification of the verbs with respect to the their semantic properties and the properties of their arguments and are to be investigated and properly described.

96The clitics se / si that are in the focus of the investigation are very important not only with respect to the word sense disambiguation (therefore to the machine translation) but also to the comprehensive and exhaustive classification of Bulgarian verbs which is a major task already undertaken in the creation of the Frame Dictionary of the Bulgarian Language. The formulation of rules resolving the ambiguity existing in the Bulgarian pronominal paradigms as well as between pronominal and non-pronominal interpretations is a crucial point with respect to the development of frame lexicon, because the proper capturing and assignment of certain interpretations and the ruling out of others by means of the context rules (and their FST implementation) will facilitate the embracing of whole classes of verbs sharing certain semantic features (grammatically identifiable) for a further analysis and classification.



IDE N. and V´ERONIS (1998), “Word Sense Disambiguation : The State of the Art”, in Computational Linguistics, 24(1), p. 1-40.

KOEVA S. (1998), “Bulgarian Grammar Dictionary. Description of the linguistic data organization concept “, in Bulgarian language, 6, p. 49-58.

KOEVA S. (2004а), “Semantic and syntactic description of reflexive, reciprocal, optative and impersonal diatheses in Bulgarian”, in Bulgarian linguistics, vol. 4, Sofia, Academic press, p. 182- 231.

KOEVA S. (2004b), “Theoretical model for a formal representation of syntactic frames”, Scripta and e-Scripta, vol. 2, Sofia, Academic Press, p. 9-26.

RESNIK P. S. and YAROWSKY D. (2000), “Distinguishing systems and distinguishing senses : new evaluation methods for Word Sense Disambiguation”, Natural Language Engineering, 6 (3), p. 113-133.

SILBERZTEIN M. (1993), Dictionnaires électroniques et analyse automatique de textes. Le système INTEX, Paris, Masson.

Table des illustrations

Titre Table 1 : Classification the different senses of se / si clitics
Fichier image/jpeg, 42k
Titre Table 2 : Distribution of pronominal and syntactic se / si clitics according to the verb person and number
Fichier image/jpeg, 50k
Titre Table 3. Distribution of the word sense distinguishing features
Fichier image/jpeg, 39k
Titre Graph 1 : Bulgarian se pronominal clitic in plural
Fichier image/jpeg, 46k
Titre Graph 2. Possessive si disambiguation rules and ambiguity cases in singular
Fichier image/jpeg, 54k
Titre Graph 3 : The interpretation of se with third person singular verbs
Fichier image/jpeg, 38k

© Presses universitaires de Franche-Comté, 2007

Conditions d’utilisation :

Rechercher dans OpenEdition Search

Vous allez être redirigé vers OpenEdition Search