Formal Grammatical and Ontological Modeling of Corpus Data on

Tibetan Compounds

Aleksei Dobrov

1 a

, Anastasia Dobrova

2 b

, Maria Smirnova

1 c

and Nikolay Soms

2 d

Saint-Petersburg State University, Saint-Petersburg, Russia

LLC “AIIRE”, Saint-Petersburg, Russia

Keywords: Tibetan Language, Compounds, Computer Ontology, Tibetan Corpus, Natural Language Processing, Corpus

Linguistics, Immediate Constituents.

Abstract:

This article provides a consistent formal grammatical and ontological description of the model of the Tibetan

compounds system, developed and used for automatic syntactic and semantic analysis of Tibetan texts, on the

material of a hand-veriﬁed corpus. This model covers all types of Tibetan compounds, which were previously

introduced by other authors, and introduces a number of new classes of compounds, taking into account their

derivation, structure and semantics. The article describes the tools used for ontological modeling of Tibetan

compounds; special attention is paid to the problem of modeling the semantics of verbs and verbal compounds.

Nominal and verbal compounds are considered separately, it is noted that the importance of verbal compounds

for the Tibetan language system is not less than that of nominal compounds. The statistical data on the

absolute frequency distribution of the use of compounds of different types in the current version of the corpus

annotation and on the amounts of ontology concepts associated with each class of compounds are given.

1 INTRODUCTION

The research introduced by this paper is a continu-

ation of several research projects (“The Basic corpus

of the Tibetan Classical Language with Russian trans-

lation and lexical database”, “The Corpus of Indige-

nous Tibetan Grammar Treatises”, “Semantic inter-

preter of texts in the Tibetan language”), aimed at the

development of a full-scale natural language process-

ing and understanding engine based on a consistent

formal model of Tibetan vocabulary, grammar, and

semantics, veriﬁed by and developed on the basis of a

representative and hand-tested corpus of texts.

The Basic Corpus of the Tibetan Classical Lan-

guage (The Basic Corpus of the Tibetan Classical

Language, 2019) and the Corpus of Indigenous Ti-

betan Grammar Treatises (The Corpus of Indigenous

Tibetan Grammar Treatises, 2019) comprise 34,000

and 48,000 tokens, respectively. Tibetan texts are rep-

resented both in the Tibetan Unicode script and in

a standard Latin (Wylie) transliteration (Grokhovskii

et al., 2015). These corpora are developed, annotated

https://orcid.org/0000-0003-0245-5407

https://orcid.org/0000-0002-8419-1005

https://orcid.org/0000-0001-5429-2051

https://orcid.org/0000-0002-0546-5101

and tested manually by a team of professional tibetol-

ogists, and in this sense are unique today.

The ultimate goal of the current project is to cre-

ate a formal model (a grammar and a linguistic ontol-

ogy) of the Tibetan language, including morphosyn-

tax, syntax of phrases and hyperphrase unities, and se-

mantics that can produce a correct morpho-syntactic,

syntactic, and semantic annotation of the corpora

without any manual corrections.

This article discusses the results achieved cur-

rently in modeling Tibetan compounds, both from

syntactical and from semantical perspective.

In Tibetan, there is no clear boundary between

morphology and syntax; at least, there are no mate-

rially expressed boundaries of word forms and, from

the point of view of an automatic system, the analysis

of compounds is in no way different from the analysis

of free combinations of Tibetan morphemes like noun

phrases or sentences. Modeling compounds is one of

the most important tasks in the current research, not

only because the frequency of use of compounds in

Tibetan texts is high (at least, as compared with texts

in Indo-European languages), but also because with-

out a correct syntactic and semantic model of com-

pounds, a huge ambiguity of Tibetan text segmenta-

tion and parsing arises, which leads to a combinatorial

144

Dobrov, A., Dobrova, A., Smirnova, M. and Soms, N.

Formal Grammatical and Ontological Modeling of Corpus Data on Tibetan Compounds.

DOI: 10.5220/0008162401440153

In Proceedings of the 11th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2019), pages 144-153

ISBN: 978-989-758-382-7

explosion (Dobrov et al., 2018a, p. 345).

To date, there is no systematic description, even

no single classiﬁcation of Tibetan compounds in ti-

betological literature, which could be considered gen-

erally accepted. The classiﬁcation, which is given in

this article for the ﬁrst time, was developed on the ba-

sis of formal-grammatical and ontological modeling

of the phenomena observed in the above-mentioned

corpora, and does not pretend to be universal, how-

ever, it covers all types of Tibetan compounds, which

were introduced by the researchers earlier, and intro-

duces new classes which do not seem to be earlier

described.

The presented classiﬁcation also covers all types

of compounds of the corpora. The exceptions are

few speciﬁc cases found in the poetic tests. How-

ever, since Tibetan poetic texts are characterized by

a large number of omissions of grammatical mark-

ers and speciﬁc abbreviations, the status of these lex-

ical units remains to be determined. It should be

also noted that the corpora include only texts in the

classical Tibetan language. Modern Tibetan texts are

planned to be added to the corpus, which may lead to

the discovery of new types of compounds. In this case

the current version of grammar will be corrected.

Unlike many other languages, the Tibetan lan-

guage is characterized by wide use not only of nom-

inal, but also of verbal compounds derivational mod-

els. Modeling verbal compounds is more complicated

for some further-mentioned reasons than modeling

nominal compounds; therefore, this article considers

not only the results of this modeling, but also the tools

that were used to obtain them.

2 RELATED WORK

Linguistic ontologies in natural language understand-

ing (NLU) systems are used as analogues for the

semantic dictionaries that were used before (cf.

(Melcuk, 1995); (Mel’

cuk and Žolkovskij, 1984),

(Leont’eva, 2003) etc.); the main difference between

an ontology and a conceptual dictionary is that, in a

semantic dictionary, semantic valencies are, in fact,

postulated, whereas in ontologies, valencies are au-

tomatically computed by inference engine subsys-

tems; semantic restrictions are deﬁned not in terms

of word lists, but in terms of base classes of ontologi-

cal concepts (that is the idea behind the mechanism of

word-sense disambiguation in (Dobrov et al., 2016),

(Dobrov, 2014), (Matuszek et al., 2006), (Rubashkin

et al., 2012), etc.).

As a formal model, ontology has to predict per-

missible and exclude impermissible relations between

concepts. Despite the clearness and obviousness of

these two requirements, there is no generally accepted

deﬁnition of the term ‘ontology’ in the scientiﬁc liter-

ature, which would have reﬂected them. The most fa-

mous and widely cited general deﬁnition of the term

‘ontology’ is ‘an explicit speciﬁcation of a concep-

tualization’ by Gruber (Gruber, 1993). Many differ-

ent attempts were made to reﬁne it for particular pur-

poses. Without claiming for any changes to this de-

facto standard, we have to clarify that, as the major-

ity of researchers in natural language understanding,

we mean not just any ‘speciﬁcation of a conceptual-

ization’ by this term, but rather a computer ontology,

which we deﬁne as a database that consists of con-

cepts and relations between them.

Ontological concepts have attributes. Attributes

and relations are interconnected: participation of a

concept in a relation may be interpreted as its at-

tribute, and vice versa. Relations between concepts

are binary and directed.

They can be represented as logical formulae, de-

ﬁned in terms of a calculus, which provides the rules

of inference. Relations themselves can be modeled by

concepts.

There is a special type of ontologies - so called

linguistic ontologies ((Dobrov et al., 2016), (Dobrov,

2014), (Matuszek et al., 2006), (Rubashkin et al.,

2012), etc.), which are designed for automatic pro-

cessing of unstructured natural language texts. Units

of linguistic ontologies represent concepts behind

meanings of real natural language expressions. On-

tologies of this kind actually model linguistic picture

of the world that stands for language semantics, as

subject domain. Ontologies that are designed for nat-

ural language processing are supposed to include rela-

tions that allow to perform semantic analysis of texts

and to perform lexical and syntactic disambiguation.

The ontology, used for this research, was developed

according with the above mentioned principles (Do-

brov, 2014). Its structure is described in detail in the

articles (Dobrov et al., 2018a), (Dobrov et al., 2018b).

Totally within the framework of this research 4335

concepts that are meanings of 3943 Tibetan expres-

sions were modelled in the ontology.

The only attempt to classify Tibetan compounds

was made by Stephan V. Beyer. All Tibetan com-

pounds are created by the juxtaposition of two exist-

ing words. Compounds are virtually idiomatized con-

tractions of syntactic groups which have inner syn-

tactic relations frozen and are often characterized by

omission of grammatical morphemes (Beyer, 1992,

p. 102). E.g., phrase (1) is clipped to (2).

Depending on part of speech of compound and its

components Stephan V. Beyer identiﬁes several mod-

Formal Grammatical and Ontological Modeling of Corpus Data on Tibetan Compounds

145

(1) x'I,Ãxn, (2) x,Ãxn,

kha ’i rgyan kha rgyan

mouth GEN adornment mouth_adornment

’adornment of mouth’ ’moustache’

els of compound-building in the Tibetan language.

The following ﬁve models are original Tibetan: noun

+ noun → noun; noun + adjective → noun; adjective

+ noun → noun; adjective + adjective → noun; noun

+ verb → verb (Beyer, 1992, p. 103–105). Stephan

V. Beyer also notes that the Tibetan language uses ad-

ditional devices for compound-building, in part bor-

rowed by Tibetans from Sanskrit within the process

of translation: noun + verb → noun; intensiﬁer + verb

→ verb (Beyer, 1992, p. 108–110).

According to syntactic relation between the com-

ponents Tibetan compounds may be divided into

two main classes: compounds with subordinate re-

lations and compounds with coordinate relations

(Grokhovskii and Smirnova, 2017, p. 137).

3 THE SOFTWARE TOOLS FOR

PARSING AND FORMAL

GRAMMAR MODELING

This study was performed with use of and within the

framework of the AIIRE project (Dobrov et al., 2016).

AIIRE is a free open-source NLU system, which is

developed and distributed in terms of GNU General

Public License (http://svn.aiire.org/repos/tproc/trunk/

t/).

This framework implements the full-scale proce-

dure of natural language processing, beginning from

graphematics (Aho-Corasick algorithm had to be used

for the Tibetan language due to absence of word de-

limiters), continuing with morphological annotation,

going further with syntactic parsing, and ending with

semantic analysis.

The morphemic dictionaries developed for the

morphological annotation for the Tibetan Language

were described in (Dobrov et al., 2017) and are not

relevant to this paper.

Syntactic parsing is performed in terms of a com-

bined constituency and dependency grammar, which

consists of the so-called classes of immediate con-

stituents (hereinafter CICs). These classes are devel-

oped as python-classes, with the builtin inheritance

mechanism involved, and provide attributes that spec-

ify the following information:

• The template of semantic graph which represents

the meaning of this constituent;

• The list of possible head constituent classes;

• The list of possible subordinate constituent

classes;

• The dictionary of possible linear orders of the sub-

ordinate constituent in relation to the head and the

meanings of each order;

• The boolean ﬁeld for head ellipsis possibility;

• The boolean ﬁeld for subordinate constituent el-

lipsis possibility;

• The boolean ﬁeld for possibility of non-idiomatic

semantic interpretation.

Due to the absence of word delimiters and any

formal evidence of boundaries between morphology

and syntax, Tibetan texts have to be parsed by mor-

phemes instead of being parsed by wordforms, as it

can be done for Indo-European languages. Therefore,

the formal grammar contains CICs both for regular

syntactic models and for models which are usually

treated as word-formational, in particular some mod-

els of derivates (there only a few of them) and models

of compounds.

The grammar is developed in straight accordance

with semantics, in a way that the meanings of syn-

tactic and morphosyntactic constituents can be cor-

rectly evaluated in accordance with the Composition-

ality principle. Each constituent is provided with a

set of semantic interpretations on the stage of the se-

mantic analysis; if this set proves to be empty for

some versions of constituents, then these versions

are discarded; this is how syntactic disambiguation

is performed. The results of semantic analysis are

stored as semantic graphs, but, for idioms like com-

pounds, these graphs consist of single concepts, thus,

the structure of semantic graphs is not a matter of dis-

cussion in this article.

4 THE SOFTWARE TOOLS FOR

ONTOLOGICAL MODELING

The ontology is implemented within the frame-

work of AIIRE ontology editor software; this soft-

ware is free and open-source, it is distributed un-

der the terms of GNU General Public License

(http://svn.aiire.org/repos/ontology/, http://svn.aiire.

org/repos/ontohelper/), and the ontology itself is

available as a snapshot at http://svn.aiire.org/repos/

tibet/trunk/aiire/lang/ontology/concepts.xml and it is

also available for unathorized view or even for edit at

http://ontotibet.aiire.org (edit permissions can be ob-

tained by access request). The basic ontological edi-

tor is described with examples from the Tibetan ontol-

ogy in (Dobrov et al., 2018a), (Dobrov et al., 2018b),

KEOD 2019 - 11th International Conference on Knowledge Engineering and Ontology Development

146

(Grokhovskii and Smirnova, 2017).

4.1 Ontological Editor for Modeling

Verb Concepts in the Ontology

Modeling verb (or verbal compound) meanings in

the ontology is associated with a number of difﬁcul-

ties. First of all, the classiﬁcation of concepts denoted

by verbs should be made in accordance with sev-

eral classiﬁcation attributes in the same time, which

arise primarily due to the structure of the correspond-

ing classes of situations that determine the seman-

tic valencies of these verbs. These classiﬁcation at-

tributes are, in addition to the semantic properties

themselves (such as dynamic / static process), the

semantic classes of all potential actants and circum-

stants, each of which represents an independent clas-

siﬁcation attribute. With the simultaneous operation

of several classiﬁcation attributes, the ontology re-

quires classes for all possible combinations of these

attributes and their values in the general class hierar-

chy.

Special tools are used to speed up and partly au-

tomate verbal concepts modeling. AIIRE ontological

editor - Ontohelper is used to build the whole hierar-

chy of superclasses for any verb meaning in the on-

tology.

The logic behind this tool is also based on the di-

vision of verbs into dynamic (terminative and non-

terminative) and static ones (Maslov, 1998). Dynamic

verbs express actions, events and processes associated

with different changes. Static verbs express states, re-

lations or qualities (GED, , p. 105). A terminative

verb denotes an action which has a limit in its de-

velopment. A non-terminative verb denotes an action

which doesn’t admit of any limit in its development

(activity).

When using the Ontohelper editor, it is necessary

to determine whether the verb being modeled denotes

action, state or activity. Terminative, non-terminative

and static verb meanings correspond to subclasses of

concepts ‘to perform an action’, ‘to perform an activ-

ity’ and ‘to be in a state’ in the ontology, respectively.

The basic class for subjects of the verb to be modeled

is indicated, as well as the basic class of direct objects

for transitive verbs and the class of indirect dative ob-

jects for verbs denoting addressed actions. It is also

possible to specify classes of circumstances, i.e., ob-

jects with special case government (e.g., for verbs that

govern the associative case).

When all necessary attributes of a verb mean-

ing are speciﬁed, the Ontohelper editor builds the

whole ontological classes hierarchy from scratch for

this particular combination of attributes, and if some

classes are already present in the ontology, they are

not built again, but tested in terms of consistency with

the current actant / circumstant relations model. This

allows not only to boost the speed of semantic valen-

cies ﬁne-tuning for verb classes, but also to rebuild the

whole hierarchy in cases when new actant / circum-

stant relation or class has to be established according

to some new observations on the corpus phenomena.

5 CLASSIFICATION OF TIBETAN

COMPOUNDS AND THEIR

MODELING IN THE FORMAL

GRAMMAR AND THE

COMPUTER ONTOLOGY

Depending on the part of speech classiﬁcation, nom-

inal and verbal compounds can be distinguished. Ini-

tially, the ontology allowed binding concepts to ex-

pressions with marking the expression as an idiom

and establishing a separate type of token, common for

nominal compounds. Since a large number of combi-

natorial explosions were caused by the incorrect ver-

sions of compounds parsing (the same sequence of

morphemes can be parsed as compounds of differ-

ent types) and their interpretation as noun phrases of

different types, it was decided to expand the number

of token types in the ontology according to identiﬁed

types of nominal and verbal compounds (see below).

As all Tibetan compounds are idioms, in AIIRE

ontology, in addition to the meanings of a compound,

meanings of its components are also modeled, so that

they could be interpreted in their literal meanings too.

This is necessary, because AIIRE natural language

processor is designed to perform natural language un-

derstanding according with the compositionality prin-

ciple (Pelletier, 1994), and idiomaticity is treated not

merely as a property of a linguistic unit, but rather as

a property of its meaning, namely, as a conventional

substitution of a complex (literal) meaning with a sin-

gle holistic (idiomatic) concept (Dobrov et al., 2018b,

p. 78–79).

5.1 Nominal Compounds Modeling in

the Formal Grammar and the

Computer Ontology

Depending on the syntactic model of the com-

pound derivation, the following types were distin-

guished for nominal compounds: compound noun

root group (CompoundNRootGroup); compound at-

tribute group (CompoundAttrGroup); noun phrase

Formal Grammatical and Ontological Modeling of Corpus Data on Tibetan Compounds

147

with genitive compound (NPGenCompound); com-

pound class noun phrase (CompoundClassNP); ad-

junct compound (AdjunctCompound); named entity

compound (NamedEntityCompound).

Compound noun root group (3) consists of NRoot

(nominal root), being the head class, and Compound-

NRootGroupArg, attached as a subordinate con-

stituent. The linear order of the subordinate con-

stituent in relation to the head is right. Compound-

NRootGroupArg stands for compound argument that

consists of a noun root attached with an intersyl-

labic delimiter - upper dots (Tib. tshegs), like in

(3). NForeign (foreign noun) is allowed to be head in

both CICs - CompoundNRootGroup and Compound-

NRootGroupArg. The linear order of the Compound-

NRootGroupArg subordinate constituent in relation

to the head is left. Heads and arguments of all Ti-

betan compounds can not be ellipsed. For all com-

pounds the setting ‘only_idiom=True’ was also made.

According to this setting any non-idiomatic interpre-

tations of a compound are excluded.

This type of compounds does not require estab-

lishment of any semantic relations in the computer

ontology for each compound. It is enough that the

meaning of the compound and its components are

modeled in the ontology, and that the general co-

ordination mechanism is also modeled in the mod-

ule for syntactic semantics (the meaning of a coor-

dinate phrase is calculated as an instance of ‘group’

concept which involves ‘include’ relations to its ele-

ments). Compound attribute groups (4) also belong

to this semantic type. It is a group of superﬁcially

homogeneous attributes within a compound. This

way of derivation is quite frequent for Tibetan per-

sonal names (the name consists of a set of epithets

(attributes), without any explicit noun) and for words,

denoting size (e.g. (4)). CompositeAttrGroup con-

sists of CompositeAtomicAttributeTopic and Com-

positeAttrCoord, where the ﬁrst part is the ﬁrst at-

tribute or group of attributes and the second part is the

last attribute attached as a subordinate constituent. If

there are more than two attributes, they are attached in

exactly the same way with CompositeAttrGroup self-

embedding. The morphosyntactic structure of com-

pounds of this type is described in detail in (Dobrov

et al., 2017).

(3) góUg,lxg, (4) rI²,ï²,

gtsug-lag ring-thung

crown_of_head_hand long_short

’basket’ ’length’

NPGenCompound is another frequent class of

Tibetan nominal compounds. These compounds

are derived from noun phrases with genitive ar-

guments by omission of the genitive case marker.

In accordance with the current grammar version,

the head constituents of NPGenCompound can be

CompoundAtomicNP, NForeign, PlaceNameForeign,

LetterCnt (countable letter), PDefRoot, PIntRoot,

NPGenCompound; the subordinate constituent class

can only be NPGenCompoundArg. The linear order

of the subordinate constituent in relation to the head

is right.

CompoundAtomicNP means atomic nominal

phrase within a compound. PlaceNameForeign (for-

eign place name) was allowed to be the head in

NPGenCompound for such cases as e.g. (5). PDe-

fRoot (deﬁnitive pronoun) and PIntRoot (indeﬁnite

pronoun) were included into possible head classes for

such combinations as e.g. (6).

(5) sI,On,mI,rIgs, (6) gúxn,dbx²,

si-khron-mi-rigs gzhan-dbang

Sichuan-nationality other-power

’Sichuan people’ ’dependent connector’

In some cases one of the components of a com-

pound is itself a compound. For example, in the

NPGenCompound (7) the head class can also be

NPGenCompound (8). This class of immediate con-

stituents is not the only case when a compound is

among heads or arguments of another compound. In

the poetic texts, even more complex structures were

discovered, the status of which is still to be clariﬁed.

(7) dpE,mOd,x², (8) dpE,mOd,

dpe-mdzod-khang dpe-mdzod

book-repository-house book-store

’library’ ’book repository’

The CIC NPGenCompoundArg stands for a

genitive compound argument that consists of the

head immediate constituent, attached with the

intersyllabic delimiter (argument immediate con-

stituent) on the left. Head classes of NPGen-

CompoundArg include: CompoundAtomicVN-

NoTenseNoMood, IndepNRoot, OnlyCompoundN-

Root, NForeign, PersNameForeign (personal name

foreign), NPGenCompound, CompoundAtomicVN.

CompoundAtomicVN stands for an atomic nomi-

nalized verb within a compound (the nominalizer

in compounds is always omitted, thus, the nomi-

nalized verb form superﬁcially comprises the verb

root only). CompoundAtomicVNoTenseNoMood is

a CompoundAtomicVN which does not have neither

mood, nor tense. As in this case, Tibetan verb roots

often do not have different allomorphs for different

moods and tenses. IndepNRoot (independent noun

root) is a noun root (allomorph of a noun root), which

KEOD 2019 - 11th International Conference on Knowledge Engineering and Ontology Development

148

can be used both within a compound and in free com-

binations. OnlyCompoundNRoot stands for a noun

root (allomorph of a noun root), which can be used

only within a compound (such noun roots are often

single-syllabic contractions of multisyllabic roots).

The relation between NPGenCompound compo-

nents is subordinate genitive relation. When model-

ing compounds of this type in the computer ontol-

ogy, it is necessary to establish speciﬁc subclasses of

the general genitive relation ’to have any object or

process (about any object or process)’ between ba-

sic classes of compound components. For example,

NPGenCompound (9) was formed from the genitive

nominal group (10). Thus, the concept ’geographical

object’ (the basic class for the ﬁrst component of the

compound (9) - bod ’Tibet’) had to be connected with

the concept skad ’language’, which is a basic class it-

self, with a relation ’to have a language (about any

geographical object)’, which is a subclass of the gen-

eral genitive relation.

(9) bOd,Ýxd, (10) bOd,I,Ýxd,

bod-skad bod gyi skad

Tibet_language Tibet GEN language

’the Tibetan language’ ’the language of Tibet’

Another frequent class of Tibetan nominal com-

pounds is CompoundClassNP. Compounds of this

type are derived from regular noun phrases with

adjectival or, more often, quasi-participial (there are

no participles in the Tibetan language, but rather

nominalized verbs that can act both as participles and

as processual nouns) attributes. Possible head classes

for the CIC CompoundClassNP are IndepNRoot,

OnlyCompoundNRoot, NForeign, PersNameFor-

eign (foreign personal name), PlaceNameForeign,

LetterCnt. Itsmodiﬁer class can be CompoundAtom-

icAttribute, CompoundAtomicAttributeNoTense,

CompoundAtomicAttributeNoTenseNoMood. Com-

poundAtomicAttribute consists of a state verb,

denoting an object feature (the head class), attached

by the intersyllabic delimiter (argument class) on the

left. For example, CompoundClassNP (11) has the

head class IndepNRoot srog ’breath’ and the modiﬁer

class CompoundAtomicAttributeNoTenseNoMood,

consisting of the intersyllabic delimiter attached on

the left to the verb chen ’be big’ which does not have

neither mood, nor tense.

The only requirement for modeling compounds of

this type in the ontology is that the basic class of a

nominal component in a compound must be a sub-

class of the speciﬁed verb subject (for the verbal com-

ponent of the compound). I.e., the verb, from which

the attribute is derived, must allow this subject by its

valencies.

Another class of Tibetan nominal compounds is

Adjunct compound (e.g. (12)). Compounds of this

class are derived from regular noun phrases with ad-

juncts. Thus, the head class for this CIC is NRoot;

and the argument class is CompoundRightNRootArg,

consisting of a noun root and the intersyllabic delim-

iter. It is necessary that the components of the com-

pound belong to the same basic class in the ontology,

or that there is no limitation on their equivalence re-

lations (the classes of the concepts should not be dis-

joint).

(11) ëOg,En, (12) lE,ôxn,

srog-chen le-tshan

breath_be_big section_section

’aspiration’ ’section’

Three upper-mentioned classes of nominal com-

pounds have exactly the same surface struc-

tures: NRootGroupCompound, NPGenCompound,

and AndjuctCompound. Compounds of all three

classes look like combinations of two bare noun roots,

but they have completely different syntactic struc-

tures and completely different semantic models. As

all compounds are modeled as idioms, when binding

ontology concepts, that they denote, to Tibetan lan-

guage units, it is necessary to specify the syntactic

class (type of token) for each concept and to make the

natural language processing engine exclude all other

possible parses thereof.

Finally, there is also such class of nominal com-

pounds as NamedEntityCompund. This class was in-

troduced for combinations of letters or exponents of

arbitrary Tibetan morphemes with NRoot like in (13).

It was decided that the Letter or Exponent is the head

component of NamedEntityCompound.

(13) lx,×x,

la-sgra

la_marker

’grammatical marker la’

The NamedEntityCompound CIC is a subclass of

named-entity nomination, where the name of the en-

tity is a letter or an exponent of any Tibetan mor-

pheme. Thus, semantic restrictions are imposed on

the possible classes of the subordinate constituent

concepts, due to the fact that only linguistic units can

have such names according with the ontology (there is

a corresponding relation between the ‘linguistic unit’

and ‘linguistic unit exponent’ concepts).

Formal Grammatical and Ontological Modeling of Corpus Data on Tibetan Compounds

149

5.2 Verbal Compounds Modeling in the

Formal Grammar and the

Computer Ontology

Depending on the syntactic model of the com-

pound derivation, the following types were dis-

tinguished for verbal compounds: verb coordinate

compound (VerbCoordCompound); compound tran-

sitive verb phrase (CompoundTransitiveVP); com-

pound atomic verbal phrase with circumstance (Com-

poundAtomicVPWithCirc) and compound associa-

tive verb phrase (CompoundAssociativeVP).

In fact, each of these types of verbal compounds

is represented by three types in the current grammar

version - verbal compound, which varies tense and

mood (e.g. CompoundTransitiveVP); verbal com-

pound, which varies only in mood (e.g. Compound-

TransitiveVPNoTense); and verbal compound, which

doesn’t vary in tense and mood (e.g. CompoundTran-

sitiveVPNoTenseNoMood). Verbal compounds like

other verbs are processed using the Ontohelper editor.

VerbCoordCompound (e.g. (14)) consists of

VRoot (verbal root, being the head of VerbCoord-

Compound) and VerbCompoundCoord that stands for

the second verb (VRoot being the head of VerbCom-

poundCoord) with the intersyllabic delimiter. Mod-

eling a verb coordinate compound meaning does not

require establishing any special semantic relations in

the computer ontology, because the upper-mentioned

general coordination meaning evaluation is involved.

These compounds are contractions of regular coordi-

nate verb phrases with conjunctions omitted.

In compound transitive verb phrase (15), the ﬁrst

nominal component is a direct object of the second

verbal component. The head class thisCompound-

TransitiveVP can be VRoot or CompoundTransfor-

mativeVP. The arguments include CompoundInstan-

ceNPArg and CompoundAtomicVNArg. The linear

order of the subordinate constituent in relation to the

head is left.

CompoundTransformativeVP is a contraction of

a regular Tibetan transformative verb phrase, i.e., a

verb phrase with terminative object. As the cor-

pus shows, compound transformative verb phrases

can themselves be parts of compound transitive verb

phrases, i.e., the complete compound can be a con-

traction of a verb phrase both with terminative and

absolutive objects.

Heads of CompoundInstanceNPArg, that is a noun

phrase argument within a compound, are IndepN-

Root, OnlyCompoundNRoot, PIndRoot and PInt-

Root. The argument class is represented by the in-

tersyllabic delimiter.

CompoundAtomicVNArg stands for a

compound argument that consists of Com-

poundAtomicVN, CompoundAtomicVNNoTense,

CompoundAtomicVNNoTenseNoMood being head

classes, and intersyllabic delimiter argument.

To ensure the correct analysis of compounds of

this type, it is necessary that the concept of the nom-

inal component of the compound be a subclass of the

basic class speciﬁed as a direct object class for the

concept of the verbal component of the compound.

E.g., the literal meaning of the compound (15) is ’to

fasten help’. The class ’any object or process’, which

includes the concept phan-pa ’help’, was speciﬁed as

a direct object for the verb ’dogs ’to fasten’.

(14) sx²s,Ãxs, (15) ¸xn,'dOgs,

sangs-rgyas phan-’dogs

be_puriﬁed_be_broaden help_fasten

’awaken and broaden’ ’assist’

The CIC CompoundAtomicVPWithCirc was

made for a combination of CompoundAtomicVP

(verbal phrase within a compound represented by a

single verb root morpheme – the head class) and the

modiﬁer – CompoundCircumstance, attached on the

left. CompoundCircumstance stands for a terminative

noun phrase within a compound, consisting of on

atom (CompoundAtomicTerminativeNP) and the

intersyllabic delimiter (the terminative case marker is

omitted as usually in compounds).

The basic class of the nominal component of

CompoundAtomicVPWithCirc should be connected

by the relation ’to be a relationship object’ with the

relation ’to have a manner of action or state’ - the ter-

minative case meaning. Thus, for the compound (16),

this relation was established on the basic class of its

nominal component rnam-pa ’type’ – ’any category’.

The texts, as a rule, use the idiomatized nominal-

ized forms of verbal compounds with the omission of

the syllabic formative –pa (a nominalizer). Thus, the

nominalized form of the verbal compound rnam-dbye

denotes a grammatical term ’case’. In this regard, in

addition to the verbal compound (16), its full nominal

form (17) is also processed in the computer ontology.

(16) Êxm,dE, (17) Êxm,pxr,dE,bx,

rnam-dbye rnam-pa r dbye-ba

type_divide type LOC divide-

NMLZ

’divide into classes’ ’case’

CompoundAssociativeVP is another class of Ti-

betan verbal compounds which was introduced for

contractions of regular associative verb phrases. It

consists of the associative verb (the head class) and

KEOD 2019 - 11th International Conference on Knowledge Engineering and Ontology Development

150

its indirect object (possible arguments being Com-

poundInstanceNPArg, CompoundAtomicVNArg).

Thus, the ﬁrst component of the compound (18)

lhag-ma ’remainder’ should belong to the class of as-

sociative objects speciﬁed for the verb bcas ’to pos-

sess’ in the Ontohelper editor. Full idiomatized nomi-

nal form of this compound (19) is also modeled in the

computer ontology.

lhag-bcas lhag-ma dang bcas-pa

remainder_possess remainder ASS

possess-NMLZ

’have a continuation’ ’continuative’

In most cases, the direct hypernym of verbal com-

pounds is the concept expressed by their verbal com-

ponent. For example, verbal compounds (20) and (21)

have the same hypernym, that is their verbal compo-

nent ’chad ’explain’.

(20) gO²,bÙxd, (21) Êxm,bÙxd,

gong-bshad rnam-bshad

top_explain type_explain

’explain above’ ’explain completely’

In other cases, there is no class-superclass relation

between the meaning of the verbal compound and the

verb from which it is derived. However, their type and

valency are always the same.

Moreover, it was revealed that such grammati-

cal features of Tibetan verb compounds as transitiv-

ity, transformativity, dativity, and associativity are al-

ways inherited from the main verb, even when the

corresponding syntactic valency seems to be fulﬁlled

within the compound.

6 CURRENT CORPUS

ANNOTATION AND

ONTOLOGICAL STATISTICS

Absolute frequencies of compounds use in current

corpora annotation are represented in Table 1. This

data shows that the most frequent Tibetan compound

class is NPGenCompound, which is a nominal com-

pound, but already the second most frequent com-

pound class is CompoundAtomicVPWithCirc, which

is verbal. That means that verbal compounds are not

less important for the Tibetan language system than

nominal ones.

The current amounts of ontological concepts for

each compound class are represented in Table 2 for

Table 1: Statistics on compound use in current corpora.

Compound class Absolute

frequency

NPGenCompound 1581

CompoundAtomicVPWithCirc 718

CompoundNRootGroup 465

AdjunctCompound 333

CompoundClassNP 180

NamedEntityCompound 145

CompoundTransitiveVP 70

CompoundAssociativeVPNoTense 46

CompoundAtomicVPWithCirc-

NoTenseNoMood

CompoundAttrGroup 24

VerbCoordCompoundNoTense-

NoMood

CompoundAssociativeObject 12

VerbCoordCompoundNoTense 3

CompoundAssociativeVPNoTense-

NoMood

CompoundAssociativeVP 2

VerbCoordCompound 1

CompoundTransitiveVPNoTense-

NoMood

CompoundAtomicVPWithCirc-

NoTense

Table 2: Statistics on nominal compounds in the ontology.

Compound class Absolute

frequency

NPGenCompound 248

CompoundClassNP 26

CompoundNRootGroup 24

AdjunctCompound 15

NamedEntityCompound 8

CompoundAttrGroup 4

VerbCoordCompound 2

VerbCoordCompoundNoTense 1

nominal compounds and in Table 3 for verbal com-

pounds. This data shows the similar distribution

among compound classes and, in some respect, re-

ﬂects the productivity of the compound derivational

models.

Table 3: Statistics on verbal compounds in the ontology.

Compound class Absolute

frequency

CompoundVPWithCirc 44

CompoundTransitiveVP 19

VerbCoordCompoundNoTense-

NoMood

CompoundAssociativeVPNoTense 3

CompoundAssociativeVP 1

Formal Grammatical and Ontological Modeling of Corpus Data on Tibetan Compounds

151

7 CONCLUSIONS AND FURTHER

WORK

The current results of the formal grammatical and on-

tological modeling of Tibetan compounds presented

in this article represent the ﬁrst of its kind consistent

systematic formal description of this material, which

is conﬁrmed by corpus data. This description does not

claim to be universal for the entire Tibetan language,

but it not only covers all types of Tibetan compounds

that researchers have introduced before, but also in-

cludes models of classes of compounds that have not

been previously described. Moreover, this model is

part of the Tibetan language module of a working

automatic text processing system, and it is veriﬁed

by analyzing the results of the automatic syntactic

and semantic annotation of the corpus of texts. This

model still does not cover all cases of compounds

use in the corpus, namely, some types of contractions

found in poetic works. An exhaustive modeling of

such phenomena is planned to be performed within

the framework of this study.

ACKNOWLEDGMENT

This work was supported by the Russian Foundation

for Basic Research, Grant No. 19-012-00616 Seman-

tic interpreter of texts in the Tibetan language.

REFERENCES

The Basic Corpus of the Tibetan Classical Language [On-

line]. 2019. Available at: http://corpora.spbu.ru/

bonito/index_gram.html. Accessed at: 19 May 2019.

The Corpus of Indigenous Tibetan Grammar Treatises [On-

line]. 2019. Available at: http://corpora.spbu.ru/

bonito/index.html. Accessed at: 19 May 2019.

Great Encyclopedical Dictionary [Bolshoy entsiklopedich-

eskiy slovar]. Linguistics [YAzyikoznanie]. Scien-

tiﬁc Publishing House “Great Russian Encyclopedia”

[Nauchnoe izdatelstvo «Bolshaya Rossiyskaya entsik-

lopediya»], Moscow, 1998, 2nd (reprint) of linguistic

encyclopedic dictionary edition.

Beyer, S. (1992). The Classical Tibetan Language. State

University of New York, New York.

Dobrov, A. (2014). Semantic and ontological relations in

aiire natural language processor. In Semantic and on-

tological relations in AIIRE natural language proces-

sor, pages 215–222, Rzeszow-Soﬁa. ITHEA.

Dobrov, A., Dobrova, A., Grokhovskiy, P., Smirnova, M.,

and Soms, N. (2018a). Computer ontology of tibetan

for morphosyntactic disambiguation. In Alexandrov,

D. A., Boukhanovsky, A. V., Chugunov, A. V., Ka-

banov, Y., and Koltsova, O., editors, Digital Trans-

formation and Global Society, pages 336–349, Cham.

Springer International Publishing. https://doi.org/10.

1007/978-3-030-02846-6_27.

Dobrov, A., Dobrova, A., Grokhovskiy, P., Smirnova, M.,

and Soms, N. (2018b). Modeling in a computer on-

tology as a morphosyntactic disambiguation strategy.

In Sojka, P., Horák, A., Kope

cek, I., and K., P., edi-

tors, Text, Speech, and Dialogue. TSD 2018. Lecture

Notes in Computer Science, volume 11107, pages 76–

83, Cham. Springer International Publishing. https:

//doi.org/10.1007/978-3-030-00794-2_8.

Dobrov, A., Dobrova, A., Grokhovskiy, P., and Soms,

N. (2017). Morphosyntactic parser and textual cor-

pora: Processing uncommon phenomena of tibetan

language. In Proceedings of the International Con-

ference IMS, pages 143–153, Saint-Petersburg. https:

//doi.org/10.1145/3143699.3143719.

Dobrov, A., Dobrova, A., Grokhovskiy, P., Soms, N., and

Zakharov, V. (2016). Morphosyntactic analyzer for

the tibetan language: aspects of structural ambigu-

ity. In International Conference on Text, Speech, and

Dialogue, pages 215–222. https://doi.org/10.1007/

978-3-319-45510-5_25.

Grokhovskii, P. and Smirnova, M. (2017). Principles

of tibetan compounds processing in lexical database.

In Proceedings of the International Conference IMS,

pages 135–142. SCITEPRESS. ISBN: 978-1-4503-

5437-0 DOI 10.1145/3143699.3143718.

Grokhovskii, P., Zakharov, V., Smirnova, M., and

Khokhlova, M. (2015). The corpus of tibetan gram-

matical works. Automatic documentation and mathe-

matical linguistics, 49(5):182–191. https://doi.org/10.

3103/S0005105515050064.

Gruber, T. (1993). A Translation Approach to Portable On-

tology Speciﬁcations, volume 5 of Knowledge Acqui-

sition. Stanford University. Computer Science De-

partment. Knowledge Systems Laboratory. https://doi.

org/10.1006/knac.1993.1008.

Leont’eva, N. (2003). Ruslan semantic dictionary as

a tool for computer understanding [semanticheskij

slovar ruslan kak instrument kompyuternogo poni-

maniya]. In Understanding in communication. Pro-

ceedings of the scientiﬁc practical conference [Pon-

imanie v kommunikacii. Materialy nauchnopraktich-

eskoj konferencii], pages 41–46, Moscow.

Maslov, Y. S. (1998). Verb [glagol]. In Yartsev, V.

and N.D, A., editors, Great Encyclopedical Dic-

tionary [Bolshoy entsiklopedicheskiy slovar]: Lin-

guistics [YAzyikoznanie]. Great Russian Encyclopedia

[Bolshaya Rossiyskaya Entsiklopediya].

Matuszek, C., Cabral, J., Witbrock, M. J., and DeOliveira,

J. (2006). An introduction to the syntax and content

of cyc. In AAAI Spring Symposium: Formalizing and

Compiling Back-ground Knowledge and Its Applica-

tions to Knowledge Representation and Question An-

swering, pages 44–49.

Melcuk, I. (1995). Phrasemes in language and phraseology

in linguistics. Lawrence Erlbaum, New Jersey.

KEOD 2019 - 11th International Conference on Knowledge Engineering and Ontology Development

152

Mel’

cuk, I. and Žolkovskij, A. (1984). Explanatory Combi-

natorial Dictionary of Modern Russian. Sonderband.

Ges. zur Förderung Slawistischer Studien.

Pelletier, F. (1994). The principle of semantic composition-

ality. Topoi, 13(11).

Rubashkin, V. S., Fadeeva, M. V., and Chuprin, B. Y.

(2012). The technology of importing fragments from

owl and kif-ontologies [tekhnologiya importa frag-

mentov iz owl i kif-ontologij]. In Proceedings of the

conference "Internet and modern society [Materialy

nauchnoj konferencii "Internet i sovremennoe obshch-

estvo"], pages 217–230.

Formal Grammatical and Ontological Modeling of Corpus Data on Tibetan Compounds

153