Method of Reconstruction of Semantic Relations using Translingual
Information
Viktor Osika, Sergey Klimenkov, Evgenij Tsopa, Alexey Pismak,
Vladimir Nikolaev and Alexander Yarkeev
Computer Science Department, ITMO University, Saint Petersburg, Russia
Keywords: Semantic Analysis, Semantic Network, Semantic Web, Natural Language Processing, Wiktionary, Russian
Language.
Abstract: The article is devoted to the problem of developing a method of restoring semantic relations using
translingual information. Wiktionary article may contain a translation section (a translingual section) – an
important element that allows to link a sense from one language to a sense from another language that is
expressed with a reference to the lexeme of that sense. One of our tasks is the design and implementation of
Wiktionary-based ontology for the semantic analysis. In this article we present a set of rules that can be used
to establish the coincidence of the same senses in different languages, and consequently map links restored
from Russian Wiktionary to English nodes. The algorithm for restoring inter-sense references includes the
selection of candidate links across senses of different language sections and set of rules to accept the
candidate to be included in the list of senses’ links. As a result, 69,309 potential links-transfers (excluding
duplicated links) were selected. More than 16,000 links between the nodes of semantic senses of the
Russian and English sections of Wiktionary were confirmed that allowed the creation of a generalized
ontology.
1 INTRODUCTION
The ability of applications to perform semantic
analysis of natural language texts becomes
especially important in modern information systems
that process unstructured and semi-structured data.
One of the most important parts of such systems is
the graph of meanings or concepts, related to each
other by semantic links. In the field of information
processing such graph is called ontology or
thesaurus. The most famous thesaurus for the
English language is WordNet(Miller, George.,
1995), which has been created by professional
linguists. For the Russian language several similar
thesauri have been constructed, including the
development of St. Petersburg University RussNet
(Azarowa, 2008) and RuThes (Loukachevitch, and
Dobrov, 2014), which is a part of the commercially
used ontology converted to WordNet. In addition, a
partial semantic translation of WordNet into Russian
has been made in (Balkova, Suhonogov, Yablonsky,
2008).
An important feature of the lexicon is its
variability. Although the thesauri described above
represent the core of concepts, they can not take into
account continuous changes that occur with
language. One of authors’ tasks is the development
and usage in the semantic analysis of the ontology,
based on Russian Wiktionary (Klimenkov, Tsopa,
Pismak., Yarkeev., 2016), (Pismak, Kharitonova,
Tsopa, Klimenkov, 2016). Russian Wiktionary is a
platform for thousands of enthusiasts that fill it with
new information and change the dictionary every
day. It should be noted that Wiktionary contains all
the information that is necessary for the computer
ontology construction: lexemes, word forms based
on the classification of A.A. Zaliznyak (Zaliznyak,
2008), important morphological and lexicographic
information that includes meanings (senses) of
words and their semantic links. In their previous
works authors transformed Wiktionary into a
machine-readable structure, organized its automatic
synchronization with the online resource of
Wiktionary, and restored explicit and implicit
semantic links between the senses of words
(Klimenkov, Tsopa, Pismak, Yarkeev., 2016).
Osika V., Klimenkov S., Tsopa E., Pismak A., Nikolaev V. and Yarkeev A.
Method of Reconstruction of Semantic Relations using Translingual Information.
DOI: 10.5220/0006516602390245
In Proceedings of the 9th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (KEOD 2017), pages 239-245
ISBN: 978-989-758-272-1
Copyright
c
2017 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
This study represents the evolution of the earlier
described approach that reconstructs links using
translingual information from the dictionary and
verifies the correctness of created links with English
and German versions of Wiktionary.
2 STRUCTURE OF THE
ONTOLOGY AND
REPRESENTATION OF
SEMANTIC LINKS
The structure of articles in Wiktionary for Russian
language (and also for some other Slavic languages)
is fundamentally different from the majority of other
Wiktionaries. In Russian Wiktionary lexeme can
contain several senses that it expresses, and each of
these senses may contain one or more semantic links
to other lexemes. In English Wiktionary it is
impossible to distinguish a specific meaning, the
semantic link exists only between lexemes. This
property of the Russian dictionary allows to create a
semantic network using explicit and implicit
structural dependencies.
In addition to the sense-to-lexeme links in
Wiktionary, there is a translation section that links
Russian Wiktionary to other languages. Such section
is called translingual. Translations are important
elements that allow to link a sense from one
language to sense from another, expressed through
the link to the lexeme of some sense. In this article
we describe a set of rules that can be used to
establish the coincidence of identical senses in
different languages, and, therefore, map links
reconstructed from Russian Wiktionary to senses of
English Wiktionary.
To reconstruct semantic links it is required to
convert the content of Wiktionary articles into a set
of semantic nodes with a certain structure. Taking
into account the feature of the conducted research, it
is necessary to provide storage in the semantic graph
of translational information.
The ontology was implemented in the form of
Neo4j graph database (Van Bruggen, 2014). Its
scheme is described by the UML-model shown in
Diagram 1. Nodes of types Gloss, MorphItem and
Lexeme are translingual.
3 ALGORITHM FOR LINKS
RECONSTRUCTION
The block of translingual links in Wiktionary is
allocated in a separate section. In such sections for
Russian and some Slavic languages the
correspondence of sense and its translations to other
languages is explicit and shown through the
successive numbering of senses (Fig. 1). In other
Diagram 1: The structure of nodes and links of the semantic graph.
Figure 1: The structure of translingual link.
sections the correspondence of the sense to its
translation can be determined probabilistically based
on the method described in (Pismak., Kharitonova,
Tsopa, Klimenkov, 2016). The translation block
contains lexemes with links to lexemes from another
language section of Wiktionary.
The developed method allows to establish links
between senses of two (and more) language sections
based on the developed set of rules. As initial data
we will use the results of the study (Klimenkov,
Tsopa, Pismak., Yarkeev., 2016), which contains the
reconstruction of inter-sense links for the Russian
section, and a dictionary for the third language.
Based on the number of lexemes and senses, as well
as the presence of advanced programming interface
for data extraction (Miller, George, 1990), the
German section was chosen.
The general algorithm for the reconstruction of
inter-sense links in the developed method is as
follows:
1. Candidate links between the senses of
different language sections are chosen
based on the developed rules;
2. The candidate-link is verified and included
in the general list of links between the
senses using another set of rules.
Below we present the rules for the search and
reconstruction of the inter-sense links.
3.1 Bilateral Links
If there are symmetrical cross-links between
lexemes and senses of two dictionaries, that is, the
sense of Russian Wiktionary links to the lexeme of
English Wiktionary, and one of the senses of English
language, in turn, points back to the Russian lexeme
that expresses this sense, then we can identify a
potential candidate for the link reconstruction.
∀
∈
∈
.
(1)
∃
,
&∃
,
→
,
(2)
Figure 2: The usage of bilateral communication.
Example (Figure 2):
LR
1
(словарь) → LR
1
S
1
(список, перечень,
собрание языковых единиц одного уровня,
обычно упорядоченных по алфавиту) → LE
1
(dictionary) → LE
1
S
1
(A reference work with a list
of words from one or more languages, normally
ordered alphabetically) → LE
1
(словарь).
3.2 Lexemes with a Single Sense
If the sense of the lexeme of the Russian dictionary
links to the lexeme of the English dictionary and this
lexeme contains only one sense, then we can identify
a potential candidate for the reconstruction of the
link (Figure 3).
∀
L
R
1
∈G
RU
LE
1
∈G
EN
|
LE
1
|
1
.
(3)
∃
,
→
,
(4)
Figure 3: Lexeme with a single sense.
For example, if one of the senses of the Russian
lexeme "шельф" links to the only sense of the
English lexeme "shelf", these two senses are
translations of each other: LR1 (шельф) → LR
1
S
1
(прибрежная зона океана) → LE
1
(shelf) → LE
1
S
1
(A reef, shoal or sandbar) → LR
1
(шельф).
It does not matter which of the lexemes (English
or Russian) has a single sense - the developed
principle is symmetric. For example, if one of the
senses of the English lexeme "trivially" is linked
with the single sense of the Russian lexeme
"тривиально", these two meanings are translations
of each other: LE
1
(trivially) → LE
1
S
1
(In a trivial
manner) → LR
1
(тривиально) → LR
1
S1 (банально,
неоригинально) → LE
1
(trivially).
3.3 The Usage of the Third Dictionary
For the search of potential links we can use the third
language dictionary (Figure 4). The German section
of Wiktionary has been used in the work.
Figure 4: The usage of the third dictionary.
If the sense of the lexeme of the Russian
dictionary has a link to the lexeme of the German
dictionary, that in turn has the sense that contains a
link to the lexeme of the English dictionary, and one
of the senses of this lexeme links back to the
Russian lexeme, then we can identify a potential
candidate for the reconstruction of the link.
∀
∈
∈
∈
.
(5)
∃
,
&∃
,
&∃
,
→
,
(6)
That is, if the sense LR1S1 in one Wiktionary
language section contains a translingual link to the
lexeme LD
1
in another language section and the
sense LD
1
S
1
contains a translingual link to the
lexeme LE
1
of the third language section, and the
sense LE
1
S
1
contains a translingual link to the
lexeme LR
1
, then you can reconstruct a biliteral link
of the type "Translation" between the sense LR
1
S
1
and the sense LE
1
S
1
.
Similarly to 3.2, this rule is also symmetric - the
original lexeme can be found in Russian Wiktionary
or in English Wiktionary:
LR
1
(смазать) → LR
1
S
1
(нанести смазку) → LD
1
(schmieren) → LD
1
S
1
(etwas mit etwas bestreichen)
→ LE
1
(oil) → LE
1
S
1
(to lubricate with oil) → LR
1
(смазать).
LE
1
(owner) → LE
1
S
1
(one who owns something) →
LD
1
(Besitzer) → LD
1
S
1
(die Person, die tatsächli-
che Herrschaft über eine Sache ausüben) → LR
1
(хозяйка) → LR
1
S
1
(собственница) → LE
1
(owner).
3.4 Verification with the Usage of
Linked Lexemes
After the analysis of the obtained links, we found
that there are incorrect links in the list of candidates.
To check and eliminate incorrect links we developed
a set of correction rules (Figure 5).
Figure 5: The usage of common linked lexemes.
If the sense-candidate of the Russian dictionary
has one of the semantic links known in the
dictionary (synonym, antonym, hyperonim, etc.)
with a lexeme, and one of the senses of this lexeme
has a link to the lexeme of the English dictionary, to
which the sense-candidate from the same dictionary
is connected with the same type of the link, then we
can confirm candidates as a reconstructed link.
∀
∈
∈
∈
.
(7)
∃
,
&∃
,
→
,
(8)
For example, if as a result of applying the
proposed rules the link with the type “translation”
has been restored between senses of lexemes "close"
and "завершать", then we can verify its correctness
in the following way. One of the senses of the
English-speaking lexeme "close" is the antonym of
the lexeme "begin", one of its senses, in turn, linked
in Wiktionary by the link of type "translation" with
the Russian lexeme “начинать”: LE
1
(close) →
LE
1
S
1
(To finish) → LE
2
(begin) → LE
2
S
1
(To start)
→ LR2 (начинать).
Due to the fact that the lexeme "начинать" in
Russian section of Wiktionary is an antonym of one
of the senses of the original lexeme "завершать", we
can conclude that the link was reconstructed
correctly: LR
1
(завершать) → LR
1
S
1
(оканчивать)
→ LR
2
(начинать).
3.5 Verification using Common
Translations into a Third Language
Figure 6: Verification of common linked lexemes that
have a translation into a third language.
∀
∈
∈
∈
∈
.
(9)
∃
,
&∃
,
&∃
,
→
,
(10)
If the sense-candidate of the Russian dictionary
is connected with a semantic link to a lexeme that
has a sense with a link to the lexeme of the German
dictionary, and the sense-candidate of the English
dictionary is connected with a link of the same type
with a lexeme that has sense that is linked to the
same lexeme of the German dictionary, then
candidates should have a reconstructed link.
For example, the correctness of the
reconstruction of the link between senses of the
lexemes "important" and "важный" can be verified
as follows. One of the senses of the English-
speaking lexeme "important" is an antonym of the
lexeme "petty", one of which senses, in turn, refers
to the German-speaking lexeme "unbedeutend": LE
1
(important) → LE
1
S
1
(Having relevant and crucial
value.) → LE
2
(petty) → LE
2
S
1
(Little, small,
secondary in rank or importance) → LD
2
(unbedeutend).
At the same time, one of the senses of the
Russian-speaking lexeme "важный" is the antonym
of the lexeme “незначительный”, which similarly
refers to the German-speaking lexeme
"unbedeutend": LR1 (важный) → LR
1
S
1
(имеющий большое значение) → LR
2
(незначительный) → LR
2
S
1
(небольшой по
количеству, размерам, силе проявления) → LD
2
(unbedeutend).
Hence, we conclude that the link between the
senses of the lexemes "important" and "важный"
was correctly defined (Figure 6).
3.6 Verification using a Sequence of
Translations
Figure 7: Sequence of translations.
If the sense-candidate from English dictionary
has a link to the lexeme of the German dictionary,
that in turn has a link to sense, containing a link to
the lexeme of the Russian dictionary that contains
the sense-candidate, then the link between
candidates is a restored link (Figure 7).
∀
∈
∈
∈
∈
∈
.
(11)
∃
,
&∃
,
&
∃
,
&∃
,
→
,
(12)
For example, the verification of the correctness
for the reconstruction of the link between the senses
of the English lexeme “rivalry” and the Russian
lexeme “соперничество” can be done as follows.
The original sense of the lexeme "rivalry" has a link-
translation to the lexeme "Rivalität" (from German
Wiktionary), one of the senses of which, in turn, has
a similar link to one of the senses of the Russian
lexeme "соперничество": LE
1
(rivalry) → LE
1
S
1
(The relationship between two or more rivals who
regularly compete with each other) → LD
1
(Rivalität) → LD
1
S
1
(das Verhältnis von
Rivalenzueinander) → LR
1
(соперничество) →
LR
1
S
1
(ситуация, при которой кто-либо
стремится превзойти, победить кого-либо
другого в чём-либо). Thus, the correctness of the
reconstruction of the link is verified.
4 RESULTS
The experiments are based on Wiktionary dump
from 17.04.2017. The Russian version of Wiktionary
contained 569,120 dictionary entries (lexemes),
1,298,654 individual senses and 305,024 links of
type "sense-lexeme".
The algorithm for the reconstruction of links
with the usage of translingual links has been
implemented as follows. The candidates for links
were reconstructed (result of applying rules from
Sections 3.1-3.3). Links in the list of candidates
were checked for correctness (applying rules from
Sections 3.4-3.6). If none of the rules from previous
step were applied to the candidate, such link was not
included in the dictionary.
As a result of applying the rule from Section 3.1
we get 25,400 candidate links. The rule from Section
3.2 restored 23,851 links (when searching for
translations from English Wiktionary); the same rule
applied for translations from Russian Wiktionary
restored 18,023 links. The rule from 3.3 is also
symmetric. It was applied twice: searching for
translations from English Wiktionary allowed to
reconstruct 193 links, and similar search for Russian
Wiktionary allowed to restore 2,737 links. The result
of the usage of rules from Sections 3.1-3.3 is 69,309
potential links-transfers that were reconstructed
(excluding duplicated links).
Using rules from Sections 3.4-3.6 we verified the
correctness of reconstructed links from candidate
list. The rule from Section 3.4 allowed to confirm
the correctness of 4,448 links: 1,732 links using
translations of Russian articles (1,224 synonyms,
450 antonyms, 41 hyponyms and 17 hyperonims)
and 2,716 - using translations of English articles
(1,736 synonyms, 618 antonyms, 258 hyponyms and
104 hyperonims). The rule from point 3.5 confirmed
the validity of 737 links (501 synonyms, 187
antonyms, 15 hyponyms and 34 hyperonims). The
rule from Section 3.6 allowed to confirm the
correctness of 17,027 reconstructed links – 5,807
translations from Russian to English and 11,220
from English to Russian.
Consistent application of the rules from Sections
3.4-3.6 allowed to confirm the correctness of the
reconstruction of 16,664 links obtained in the
previous stage.
5 CONCLUSIONS
After applying the developed method, more than
16,000 links between the nodes of semantic senses
of Russian and English sections of Wiktionary were
reconstructed. These links allowed to create a
generalized ontology. Inter-semantic links expand
the connections of the sense with a set of synonyms
in synset (Miller, George, 1990), (Gross, and Miller,
1990), (Fellbaum, Christiane, 1990) for the lexeme.
As a result we can merge the English semantic node
of ontology with Russian semantic node. In the
implementation of presented approach we decided to
keep two semantic nodes separated, but we
connected them with a special type of the link called
"synonym-translation".
For further research we plan to combine the
developed method with results of our previous
researches to reconstruct more sense-to-sense links.
Another direction of our further researches is
focused on study the ability to apply the information
from temporal content of articles in Wiktionary in
context of link reconstruction, namely the history of
article changes.
REFERENCES
Azarowa, I., 2008. RussNet as a computer lexicon for
Russian. Proceedings of the Intelligent Information
systems IIS-2008.
Balkova, V., Suhonogov, A., Yablonsky, S., 2008. Some
issues in the construction of a Russian wordnet grid.
Proceedings of the Forth International WordNet
Conference, Szeged, Hungary.
Bruggen, R., 2014. Learning Neo4j. Packt Publishing Ltd.
Fellbaum, C., 1990. English verbs as a semantic net.
International Journal of Lexicography.
Gross, D., Miller, K., 1990. Adjectives in wordnet.
International Journal of lexicography.
Klimenkov, S., Tsopa, E., Pismak, A., Yarkeev, A., 2016.
Reconstruction of Implied Semantic Relations in
Russian Wiktionary. Proceedings of the 8th
International Joint Conference on Knowledge
Discovery, Knowledge Engineering and Knowledge
Management (KDIR).
Loukachevitch, N., Dobrov, B., 2014. RuThes linguistic
ontology vs. Russian wordnets. Proceedings of Global
WordNet Conference GWC-2014.
Miller, George A., 1995. WordNet: a lexical database for
English. Communications of the ACM.
Miller, George A., 1990. Nouns in WordNet: a lexical
inheritance system. International journal of
Lexicography.
Pismak, A., Kharitonova, A., Tsopa, E., Klimenkov, S.,
2016. Method of automatic construction of semantic
network from weakly structured sources. Software
products and systems.
Torsten, Z., Müller, C., Gurevych, I., 2008. Extracting
Lexical Semantic Knowledge from Wikipedia and
Wiktionary. LREC.
Zaliznyak, A., 2008. Grammatical dictionary of the
Russian language M .: AST-Press.
