INTERPRETING BULGARIAN SOUND ALTERNATIONS OF

INFLECTIONAL MORPHOLOGY IN DATR

Velislava Stoykova

Institute for Bulgarian Language, Bulgarian Academy of Sciences, 52, Shipchensky proh. str., bl. 17, 1113 Soﬁa, Bulgaria

Keywords:

Knowledge representation, Semantic networks, Sound alternations, Inﬂectional morphology.

Abstract:

The paper presents an approach to interpret sound alternations of Bulgarian language for inﬂectional morphol-

ogy of deﬁnite article. The DATR language for lexical knowledge presentation is accepted as a framework,

and the analysis and examples of semantic network for different part-of-speech are presented. Finally, more

general conclusions for formal interpretation of sound alternations for inﬂectional morphology are deﬁned.

1 INTRODUCTION

Recent developments in AI offer various approaches

to knowledge presentation. Thus, natural language

processing applications use different techniques to

represent and differentiate between phonological,

morphological, and syntactic knowledge even all

these types represent the inherent language features.

The sound alternations inﬂuence the inﬂectional

morphology of almost all part-of-speech of standard

Bulgarian language and as a result they form irregular

word forms. In fact, we have a rather unsystemat-

ically formed variety of regular and irregular sound

alternations which is very difﬁcult to be interpreted

formally.

2 THE COMPLEXITY OF THE

FORMAL INTERPRETATION

The problem of interpreting sound alternations of in-

ﬂectional morphology is a key problem of any for-

mal presentation of standard Bulgarian language. In

our paper we present different types of sound alter-

nations, and analyse the DATR encoding of the fea-

ture of deﬁniteness in Bulgarian and the related in-

ﬂectional morphology interpretation. We propose a

formal account of sound alternations using both the

architecture of the application – the principle of lo-

cal and global inheritance and the lexical information

presentation scheme. We offer an interpretation of

both phonetic and morphological features in one con-

sistent formal representation.

2.1 Types of Sound Alternations

The phonetic alternations in Bulgarian are of various

types. They take place in different historical periods

of language development, and form different types of

irregular word forms. The sound alternations inﬂu-

ence both derivational and inﬂectional morphology.

Further we are going to analyse only those alterna-

tions which are signiﬁcant for the inﬂectional mor-

phology.

The sound alternations in standard Bulgarian are

divided into two general types: (i) phonologically

based sound alternations, and (ii) historically based

sound alternations (Grammar, 1982). The classiﬁca-

tion is accepted in the academic descriptive grammar

works, and it does not concern the formal way the

sound alternations could be interpreted. In further ex-

amples sound alternations are illustrated with respect

to the grammar features of number and deﬁniteness,

which in Bulgarian language are inﬂectional grammar

features.

2.2 Phonologically based Sound

Alternations

Phonologically based sound alternations concern

mostly vocal sounds. Those of them which reﬂect

nominal inﬂectional morphology are of three types.

The ﬁrst type of phonologically based sound alterna-

tions is the so-called epenthesis of [@]

. The example

Here we use SAM Phonetic Alphabet (SAMPA) to

mark sounds and Latin alphabet for the letters. Because of

mismatching between the two alphabets some of the Cyril-

486

Stoykova V..

INTERPRETING BULGARIAN SOUND ALTERNATIONS OF INFLECTIONAL MORPHOLOGY IN DATR.

DOI: 10.5220/0003293804860491

In Proceedings of the 3rd International Conference on Agents and Artiﬁcial Intelligence (ICAART-2011), pages 486-491

ISBN: 978-989-8425-40-9

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

is as follows:

[ideali"z@m] - [ideali"zm-@t]

*idealism (sg.undef.) - idealism-the (sg.def.)

The second type of phoneticallybased sound alter-

nations is the methathesis of [r@], [l@] : [@r], [@l].

The example is:

[gr@b] - [g@"rb-Ove]

*back (sg.) - back-s (pl.)

The third type phonologically based sound alter-

nation is the so-called vocal alternation of ja-slide.

The alternation takes place when [3] (ja) is in stressed

position and stands before non-palatal syllable. Un-

der these conditions the alternative word form has dif-

ferent vocal sound, namely [e]. The ja-slide divides

Bulgarian dialects into Eastern and Western, however,

in the standard Bulgarian both regular and irregular

forms are accepted. The example is as follows:

[gr’ax] - [gre"x-Ove]

*sin (sg.) - sin-s (pl.)

2.3 Historically based Sound

Alternations

Historically based sound alternations are mostly con-

sonant alternations. The ﬁrst type is the so-called II

palatalization. It could be explained as a result of the

more general process of palatalising the consonants,

which takes place during centuries and inﬂuences all

Slavic languages. This alternation changes the conso-

nants [g], [k], [x] into [z], [ts], [s] respectively when

they precede the vocals [3], [e] or [i]. The examples

are:

[pO"dlOg] - [pO"dlOz-i]

*subject (sg.) - subject-s (pl.)

[ve"stnik] - [ve"stnits-i]

*newspaper (sg.) - newspaper-s (pl.)

[mOna"x] - [mOna"s-i]

*monk (sg.) - monk-s (pl.)

3 THE INFLECTIONAL NATURE

OF DEFINITENESS IN

BULGARIAN

The standard Bulgarian language does not use cases

for syntactic representation but it has very rich inﬂec-

tional system - both for derivational and for inﬂec-

tional morphology, and it uses prepositions and a base

lic letters are presented by using two Latin letters.

word form instead a case declination. It is consid-

ered to be a language using relatively free word order,

so the subject can take every syntactic position in the

sentence (including the last one). Another important

grammar feature of Bulgarian is the feature of deﬁ-

nite article which is an ending morpheme (Grammar,

1983). The fact gives a priority to morphological in-

terpretations of deﬁniteness in spite of syntactic since,

and at the level of syntax, the deﬁnite article showsthe

subject (when it is not a proper name).

3.1 The Deﬁnite Article - Its Semantics

and Morphosyntactic Features

At syntactic level, the deﬁniteness in Bulgarian may

express various types of semantic relationships like a

case (to show subject), part-of-whole, deixis etc. The

deﬁnite article can assign an individual, or quantity

deﬁniteness, and it has a generic use as well.

3.2 The Formal Morphological Marker

of Deﬁniteness

The syntactic function of deﬁniteness in Bulgarian is

expressed by a formal morphological marker which

is an ending morpheme (Grammar, 1983). It is differ-

ent for the genders, however, for the masculine gen-

der two types of deﬁnite morphemes exist – to de-

termine a deﬁned in a different way entities, which

have two phonetic alternations, respectively. For the

feminine and for the neuter gender only one deﬁ-

nite morpheme exists, respectively. For the plural,

two deﬁnite morphemes are used depending on the

ending vocal of the main plural form. The follow-

ing part-of-speech in Bulgarian take the deﬁnite arti-

cle: nouns, adjectives, numerals (both cardinals and

ordinals), possessive pronouns (the full forms), and

reﬂexive-possessive pronoun (its full form).

The deﬁnite morphemes are the same for all part-

of-speech, however, in further description we are

going to analyze only some general types of rules

used for the interpretation of inﬂectional morphol-

ogy of deﬁniteness in Bulgarian given in Stoykova

(Stoykova, 2002), (Stoykova, 2004a), (Stoykova,

2010). Thus, our task is to present the related archi-

tecture of the application which suggests a subsequent

algorithm for the rule-based interpretation of the in-

ﬂectional morphology of deﬁnite article.

INTERPRETING BULGARIAN SOUND ALTERNATIONS OF INFLECTIONAL MORPHOLOGY IN DATR

487

4 THE COMPUTATIONAL

MORPHOLOGY APPROACHES

The standard computational approach to both deriva-

tional and inﬂectional morphology is to represent

words as a rule-based concatenation of morphemes,

and the main task is to construct the relevant rules for

their combinations. With respect to the number and

types of morphemes, different theories offer differ-

ent approaches depending on the variations of either

stems or sufﬁxes as follows (Fig.1):

(i) Conjugational solution offers an invariant stem

and variant sufﬁxes, and

(ii) Variant stem solution offers variant stems and

invariant sufﬁx.

Both these approaches are suitable for languages,

which use inﬂection rarely to express syntactic struc-

tures, whereas for those using rich inﬂection some

cases where phonological alternations appear both in

stem and in concatenating morpheme a ”mixed” ap-

proach is used to account for the complexity. Also,

some complicated cases where both preﬁxes and suf-

ﬁxes have to be processed require such approach.

Figure 1: Conjugational solution and variant stem solution.

We evaluate the ”mixed” approach as a most ap-

propriate for our task because it considers both stems

and sufﬁxes as variables and, also, can account for the

speciﬁc phonetic alternations. The additional require-

ment is that during the process of the inﬂection all

generated inﬂected rules (both using preﬁxes and suf-

ﬁxes) have to produce more than one type of inﬂected

forms.

Thus, the DATR language for lexical knowledge

presentation is a suitable formal framework for pre-

senting inﬂectional morphology of Bulgarian deﬁnite

article.

5 THE DATR LANGUAGE

The DATR language for lexical knowledge presenta-

tion offers syntagmatic operators (which can be used

to deﬁne the concatenation procedures) and paradig-

matic operators (which can be used to deﬁne the spe-

ciﬁc structure of the inﬂecting rules, and for further

development of the part-of-speech interpretations).

The DATR language is a non-monotonic lan-

guage for deﬁning the inheritance networks through

path/value equations (Evans and Gazdar, 1996). It has

both an explicit declarative semantics and an explicit

theory of inference allowing efﬁcient implementation,

and at the same time, it has the necessary expressive

power to encode the lexical entries presupposed by

the work in the uniﬁcation grammar tradition (Evans

and Gazdar, 1989a), (Evans and Gazdar, 1989b).

In DATR information is organized as a network

of nodes, where a node is a collection of related in-

formation. Each node has associated with it a set of

equations that deﬁne partial functions from paths to

values where paths and values are both sequences of

atoms. Atoms in paths are sometimes referred to as

attributes.

DATR is functional, it deﬁnes a mapping which

assigns unique values to node attribute-path pair, and

the recovery of this values is deterministic. With re-

spect to its universality, DATR’s formal properties and

techniques underlay both the rule-based inference and

non-monotonic inference by default, and allow to ac-

count for language phenomena such as regularity, ir-

regularity, and subregularity, by using deterministic

parsing.

The semantics of DATR uses non-monotonic

inference and default inheritance, and allows the

generalization-capturing representation of the inﬂec-

tional morphology. DATR has the expressive power

which is capable to encode and process both syntac-

tic and morphological rules and it allows represen-

tation of grammar knowledge by using the semantic

networks.

The DATR language has a lot of imple-

mentations, however, the analysed application

was made by using QDATR 2.0

(consult

http://www.cogs.susx.ac.uk/lab/nlp/datr

/datrnode49.html

for a related ﬁle

bul_det.dtr

This PROLOG encoding uses Sussex DATR notation

(DATR, 1997). DATR allows construction of various

types of language models (language theories), how-

ever, the analysed application presents a rule-based

formal grammar and a lexical database. The partic-

ular query to be evaluated is related inﬂected word

forms, and the implementation allows to process

words in Cyrillic alphabet.

6 THE ARCHITECTURE OF THE

MODEL

The analyzed application of inﬂectional morphology

of Bulgarian deﬁnite article is linguistically moti-

vated. In particular, the underlying basic idea is that

ICAART 2011 - 3rd International Conference on Agents and Artificial Intelligence

488

(i) definite

morphemes

(ii) plural

morphemes

(iv) lexical

database

(iii) nountype

hierrarchy

(query)

Inflectingforms

Figure 2: The general architecture of the model.

of a paradigm since morphemes are deﬁned to be of

semantic value and are considered as a realisation of

a speciﬁc morphosyntactic phenomenon. The words

are encoded by introducing different roots to account

for the related phonetic alternations, which are de-

ﬁned to be of semantic value as well.

Some other DATR applications which present

Slavonic inﬂectional morphology are available for

Polish, Russian, Czech, Slovene. The nominal inﬂec-

tional morphology interpretations deals mostly with

case morphology presentation (Czuba, 1994), (Cor-

bett and Fraser, 1993), however they do not offer a

systematic account of sound alternations. As for ver-

bal inﬂectional morphology presentation, the inter-

pretations (Skoumalova, 1996), (Brown, 1998), (Er-

javec, 1992) use as underlying idea that of a conjuga-

tion.

The architecture represents an inheritance net-

work consisting of various nodes which allows to ac-

count for all related inﬂected forms within the frame-

work of one grammar theory. Thus, the general archi-

tecture of the application is as follows (Fig. 2):

(i) All deﬁnite inﬂecting morphemes for all forms

of deﬁnite article attached to node

DET

and deﬁned by

their values by the paths

<masc>

<masc_1>

<femn>

<neut>

, and

<plur>

(ii) 12 inﬂecting morphemes for generating plural

forms deﬁned at node

Suff

(iii) The inﬂectional rules deﬁned as concatena-

tions of morphemes for generation of all possible in-

ﬂected forms attached to the related inﬂectional types

nodes.

(iv) The words are given as lexical database at-

tached to their inﬂectional type nodes. They are

deﬁned as lexical entries through paths

<root>

, and

, so to account for

the different phonological alternations.

The DATR logical representation framework uses

rule-based reasoning with non-monotonic inference

and default inheritance to represent the inﬂectional

rules in semantic network. It suggests the structure of

semantic network that can employ the generalization-

capturing rules in which the grammar knowledge is

encoded by the attachment of inﬂectional rules to the

Figure 3: The word structure according to the general lin-

guistic morphological theory.

related nodes. In principle, DATR permits multi-

ple default inheritance and prioritized inheritance en-

forced by orthogonal representation, and suggest the

lexicon being structured mostly by inheritance. This

technique allows to account for the grammar irreg-

ularities and to use the compilation rules which can

generate all possible inﬂected forms within one appli-

cation.

The general morphological theory offers a seg-

mentation of words (Fig. 3) which consists of root

to which preﬁxes, sufﬁxes or endings are attached.

In Bulgarian, all three types of morphemes are used

and a hierarchical structure of the lexical represen-

tation in which the feature of gender is a trigger to

change the values of the inﬂected forms is used. Dur-

ing the process of inﬂection, also, various phonetic

alternations are taking place. The phonetic alterna-

tions at the morpheme boundary are interpreted ei-

ther by deﬁning new grammar rules or new nodes, and

the phonetic alternations inside morphemes are inter-

preted by introducing different roots. It is possible,

also, to use the technique of ﬁnite state transducers

(Stoykova, 2004b).

The analysed application interprets, also, more

complicated cases of inﬂection, where both preﬁxes

and sufﬁxes can be processed by deﬁning new nodes

of the network.

7 THE INFLECTIONAL RULES

The DATR analysis of nouns (Stoykova, 2002) starts

with node

DET

which deﬁnes all inﬂecting morphemes

for the deﬁnite article and is as follows

DET:

<sing undef> ==

<sing def_2 masc> == _ja

<sing def_2 masc_1> == _a

<sing def_1 masc> == _jat

<sing def_1 masc_1> == _ut

<sing def_1 femn> == _ta

<sing def_1 neut> == _to

Here and elsewhere in the description we use Latin al-

phabet to present morphemes instead of Cyrillic used nor-

mally. Because of mismatching between both some of typ-

ically Bulgarian phonological alternations are assigned by

two letters, whereas in Cyrillic alphabet they are marked by

one.

INTERPRETING BULGARIAN SOUND ALTERNATIONS OF INFLECTIONAL MORPHOLOGY IN DATR

489

<plur undef> ==

<plur def_1> == _te.

The node

Suff

deﬁnes 12 inﬂecting morphemes for

generating the plural inﬂected forms.

Suff:

<suff_11> == _i

<suff_111> == _ovci

<suff_12> == _e

<suff_121> == _ove

<suff_122> == _eve

<suff_123> == _ovce

<suff_21> == _a

<suff_22> == _ja

<suff_211> == _ishta

<suff_212> == _ta

<suff_213> == _ena

<suff_214> == _esa.

The basic node of the nouns inﬂectional types is the

node

Noun

Noun:

<suff> == suff_11

<gender> == masc_1

<> == <stem> DET: <Idem "<gender>">

<stem sing> == "<root sing>"

<stem plur>=="<root plur>"Suff:<"<suff>">.

The example lexeme of word for traveller ’putnik’ is

deﬁned by

<root>

and

Putnik: <> == Noun

<root> == putnik

<root plur> == putnic.

The following inﬂected forms are generated accord-

ing to the deﬁned rules:

Putnik: <gender> == masc_1.

Putnik: <sing undef> == putnik.

Putnik: <plur undef> == putnic_i.

Putnik: <sing def_1> == putnik_ut.

Putnik: <sing def_2> == putnik a.

Putnik: <plur def_1> == putnic_i_te.

The adjectives inﬂectional types hierarchy (Stoykova,

2004a) uses deﬁnite morphemes of node

DET

and the

basic node of the hierarchy deﬁnes the grammar rules

for generating all inﬂected forms for the feature of

gender, number, and deﬁniteness as follows.

AdjG:

<sing undef masc> == "<root>"

<sing undef femn> == "<root gend>" _a

<sing undef neut> == "<root gend>" _o

<sing def_2 masc> =="<plur undef masc>" DET

<sing def_1 masc> =="<plur undef masc>" DET

<sing def_1> == "<sing undef>" DET

<plur undef> == "<root gend>" _i

<plur def_1> == "<plur undef>" DET.

Node

Adj

inherits all grammar rules of node

AdjG

but

employs also the grammar rules for generating forms

for comparison of degree which are produced by us-

ing prepositional morphemes.

Adj:

<> == AdjG

<compar> == po_ "<>"

<superl> == naj_ "<>".

Node

Adj_1

inherits the grammar rules of node

Adj

but changes the rules for generating singular forms

for feminine and for neuter gender to use the palatal

gender morphemes.

Adj_1:

<> == Adj

<sing undef femn> == "<root gend>" _ja

<sing undef neut> == "<root gend>" _jo.

Node

Adj_2

deﬁnes the inﬂectional rules of the ad-

jectives, which realise two types of phonetic alterna-

tions during the process of inﬂection. At this node

additional inﬂectional base form

is in-

troduced to account for the complexity.

Adj_2:

<> == Adj

<plur undef> == "<root plur>" _i.

The example lexeme for word short ’tesen’ is deﬁned

as follows (the generated inﬂected forms are given at

the Appendix):

Tesen: <> == Adj_2

<root> == tesen

<root gend> == tjasn

<root plur> == tesn.

The reﬂexive-possessive pronoun uses inﬂectional

rules (Stoykova, 2010) as follows:

Adj_5:

<> == Adj

<sing undef femn> == "<root gend>" _ja

<sing undef neut> == "<root gend>" _e

<sing def_1 masc> == "<root gend>" DET

<sing def_2 masc> == "<root gend>" DET.

The example lexeme of the reﬂexive-possessive pro-

noun self ’svoj’ is deﬁned as follows (the generated

inﬂected forms are given at the Appendix):

Svoj: <> == Adj_5

<root> == svoj

<root gend> == svo.

8 CONCLUSIONS

The proposed analysis uses linguistic motivation for

the encoding. The structure of semantic hierarchy

uses the grammar feature with a wider scope as a

trigger to change the value of inﬂected forms. The

interpretation do not offer a systematic account of

sound alternations but suggests some ideas for pre-

senting them by using both architecture of the appli-

cation – the principle of local and global inheritance,

and lexical information presentation scheme. Most

of inﬂectional type nodes are deﬁned to account for

ICAART 2011 - 3rd International Conference on Agents and Artificial Intelligence

490

different phonological alternations. The phonetic al-

ternations at the morpheme boundary are interpreted

either by deﬁning new grammar rules or new nodes,

and the phonetic alternations inside morphemes are

interpreted by introducing different roots.

The analyzed application of Bulgarian inﬂec-

tional morphology accounts for orthographic princi-

ples, phonetic alternations, and morphological depen-

dencies and is expected to be reencoded and imple-

mented in lexicography for producing different types

of dictionaries.

REFERENCES

Brown, D. (1998). Stem indexing and morphophonological

selection in the russian verb. In R. Fabri, A. Ortmann,

and T. Parodi (eds.) Models of Inﬂection. Tuebingen:

Niemeyer, 196-221.

Corbett, G. and Fraser, N. (1993). Network morphology: a

datr account of russian nominal inﬂection. In Journal

of Linguistics. 29, 113-142.

Czuba, K. (1994). The datr web pages at sussex. In

http://www.cogs.susx.ac.uk/lab/nlp/datr/ datrnode49,

ﬁle polish n.dtr

DATR (1997). The datr web pages at sussex. In http://www.

cogs.susx.ac.uk/lab/nlp/datr/.

Erjavec, T. (1992). Treatments of slovene verb morphology

in inheritance models. In MSc Thesis. Edinburgh.

Evans, R. and Gazdar, G. (1989a). Inference in datr. In

Fourth Conference of the ECACL. 66-71.

Evans, R. and Gazdar, G. (1989b). The semantics of datr.

In Anthony G. Cohn, ed. Proceedings of the Seventh

Conference of the Society for the Study of Artiﬁcial

Intelligence and Simulation of Behaviour. 79–87.

Evans, R. and Gazdar, G. (1996). Datr: A language for

lexical knowledge representation. In Computational

Linguistics. 22(2):167–216.

Grammar (1982). Gramatika na suvremennia bulgarski kni-

zoven ezik. In Fonetika, tom. 1.

Grammar (1983). Gramatika na suvremennia bulgarski kni-

zoven ezik. In Morphologia, tom. 2.

Skoumalova, H. (1996). Czech hierarchical lexicon. In

http://utkl.ff.cuni.cz/∼skoumal/czech verb.html.

Stoykova, V. (2002). Bulgarian noun – deﬁnite article in

datr. In D. Scott, ed. Artiﬁcial Intelligence: Methodol-

ogy, Systems, and Applications. LNAI 2443. Springer,

152–161.

Stoykova, V. (2004a). The deﬁnite article of bulgarian ad-

jectives and numerals in datr. In C. Bussler and D.

Fensel, eds. Artiﬁcial Intelligence: Methodology, Sys-

tems, and Applications. LNAI 3192. Springer, 256–

266.

Stoykova, V. (2004b). Modeling sound alternations of bul-

garian language in datr. In E. Buchberger ed. Proceed-

ings of KONVENS 2004, Shriftenreiche der Oesterre-

ichischen Gesellschaft fur Artiﬁcial Intelligence. Band

5, Wien, 201-204.

Stoykova, V. (2010). Bulgarian possessive and reﬂexive-

possessive pronouns in datr. In R. Trappl ed. Cyber-

netics and Systems 2010. 426-432.

APPENDIX

Tesen: <sing undef masc> == tesen.

Tesen: <sing undef femn> == tjasna.

Tesen: <sing undef neut> == tjasno.

Tesen: <plur undef> == tesni.

Tesen: <compar sing undef masc> == po-tesen.

Tesen:<compar sing undef femn> == po-tjasna.

Tesen:<compar sing undef neut> == po-tjasno.

Tesen: <compar plur undef> == po-tesni.

Tesen:<superl sing undef masc> == naj-tesen.

Tesen:<superl sing undef femn>== naj-tjasna.

Tesen:<superl sing undef neut>== naj-tjasno.

Tesen: <superl plur undef> == naj-tesni.

Tesen: <sing def_1 masc> == tesnijat.

Tesen: <sing def_2 masc> == tesnija.

Tesen: <sing def_1 femn> == tjasnata.

Tesen: <sing def_1 neut> == tjasnoto.

Tesen: <plur def_1> == tesnite.

Tesen: <compar sing def_1 masc>==po-tesnijat.

Tesen: <compar sing def_2 masc>==po-tesnija.

Tesen: <compar sing def_1 femn>==po-tjasnata.

Tesen: <compar sing def_1 neut>==po-tjasnoto.

Tesen: <compar plur def_1> == po-tesnite.

Tesen:<superl sing def_1 masc>==naj-tesnijat.

Tesen: <superl sing def_2 masc>==naj-tesnija.

Tesen:<superl sing def_1 femn>==naj-tjasnata.

Tesen:<superl sing def_1 neut>==naj-tjasnoto.

Svoj: <sing undef masc> == svoj.

Svoj: <sing undef femn> == svoja.

Svoj: <sing undef neut> == svoe.

Svoj: <plur undef> == svoi.

Svoj: <sing def_1 masc> == svojat.

Svoj: <sing def_2 masc> == svoja.

Svoj: <sing def_1 femn> == svojata.

Svoj: <sing def_1 neut> == svoeto.

Svoj: <plur def_1> == svoite.

INTERPRETING BULGARIAN SOUND ALTERNATIONS OF INFLECTIONAL MORPHOLOGY IN DATR

491