The study of collocability is an important task and is still highly relevant in linguistics. The present paper discusses the issue of collocability and collocations in a number of Russian dictionaries (the Dictionary of the Russian Language, Dictionary of Set Verb-Noun Phrases, the Dictionary of Russian Idiomatics and the Dictionary of Collocations). These dictionaries were analyzed and used to create a Russian collocations database which includes both verified or dictionary collocations, as well as data from text corpora. At present, the database includes about 18,500 collocations.
Although the term “collocation” appeared in the Russian language long ago [Akhmanova, 1966], it is not generally recognized by Russian linguists and even is absent in the Russian Linguistic Dictionary [Yartseva, 1990]. For a long time there was no agreement among scholars how to call such lexical units; cf. “set verbalnoun expressions” [Deribas, 1983], “analytic lexical phrases” [Teliya, 1996], etc. It can be seen as a reason why collocations dictionaries are not so well-known compared to English ones. The majority of modern authors understand under a collocation a statistically set phrase. Collocations can be put between free phrases and idioms on a scale of phrases. The monograph [Borisova, 1995a] has proved to be the first work in Russian linguistics that officially “introduced” this term in Russian studies. According to the author one of the key properties of a collocation is “the impossibility of prediction of such combinations on the basis of meanings of their components” [Borisova, 1995a: 13]. In the “Meaning–Text” model [Mel’chuk, 1974] collocations are considered as a subclass of more extensive class of set phrases, or phrasemes.
Following Testelets’s approach [Testelets, 2001] we interpret collocability of a lexical unit extremely widely and understand it as the ability to connect with other lexical units. Thus, we will consider phrases of different degrees of stability (from collocations to idioms and phraseological units with non-compositionality). For example, according to Telia’s [Telia, 1996] classification: idioms (rabochaja loshad’ ‘working horse’), phraseological units (teljachij vostorg ‘foolish enthusiasm’), fixed expressions (vsegokhoroshego ‘all the best’), cliche (minutu vnimanija ‘minute of attention’). Also here we add terms (kontrol’naja palata ‘control chamber’) and set phrases (podvergnut’sja deformatsii ‘undergo deformation’).
We came to the conclusion that it could be a certain problem to find an appropriate dictionary that describes collocations and is large enough in order to both list high frequency word combinations and give a sufficient number of examples for them. The dictionaries differ in their representation of information on collocability and in coverage of the examples. Data on collocability can be found in various dictionaries and sources and in different parts of the dictionary entries, e.g. quotes, speeches or specialized zones. The latter can list set expressions representing a certain lexical unit. Dictionary entries should be structured in order to make possible further collocation extraction. And here lies a certain problem as the majority of Russian dictionaries (if can ever be found in digital form) represent scanned copies of printed versions without OCR. Their recognition is then followed by further division of entries into a structured format that takes into account possible grammatical information, examples, quotations etc. 3
For our project we made a survey of various dictionaries that can be used as sources of collocations. In the present paper we will dwell on those of them that have been already processed during the project.
The explanatory dictionaries implement various ways to represent the information on combinatorial restrictions but they share much in common. They use different fonts and labels in their entries. Set phrases, multiword expressions and collocations can be described not only in special sections of the entries but also in the examples, sayings and quotations.
The “Dictionary of the Russian Language” [DRL, 1981–1984] followed the “Dictionary of Contemporary Literary Russian Language” [DCLRL, 1948–1965] (one of the most important projects in Soviet lexicography started before the Second World War but its first volume appeared only in 1950) and comprises more than 80,000 lexical items resulting in 4 volumes. Its first edition was published in 1957–1961, the revised version came out in 1981–1984 and soon became popular among linguists and other scholars.
Compared to [DCLRL, 1948-1965] the given dictionary is more concise and we can see it on the excerpt from its online version (Fig. 1). Here collocations do not have a special mark-up (they are given as examples in italics) but the phraseological units do.
Traditionally dictionaries that describe collocability are intended at language learners and teachers. Below we will discuss Russian dictionaries that represent collocability and can be to a certain degree thus called collocations dictionaries. Also we can name several glossaries that were compiled for language learners, some of them printed in a small number of copies in the republics of the Soviet Union and hence are now rare books (for example, [Anisimova et al., 1975]).
The “Dictionary of Set Verb-Noun Phrases in Russian” [Deribas, 1983] was intended for students of Russian and comprises 5,197 collocations for 744 verbs and 1,345 nounsin total. The authors put collocations between free phrases and phraseological units: opravdyvat’ (opravdat’) ozhidaniya ‘to confirm expectations’; pitat’ uvazheniye ’to respect’; chitat’ lektsiyu ‘to hold a lecture’ etc. The majority of phrases consist of bigrams including verbs and nouns as direct or indirect objects. The authors emphasize the dictionary is focused on language learners and does not provide any definitions or explanations merely listing the collocations of literary
At the moment, there is no such a system for the Russian language that would allow researchers to obtain not only corpus data, but also “reference” (verified) information on the vocabulary and behavior of lexical units, combined due to different characteristics. At the same time, there is a number of unique and valuable lexicographic projects that describe collocability of units, although in different ways.
There are online resources (for example, slovari.ru or gramota.ru), which provide an opportunity to learn the meaning of a word and view dictionary entries containing information about collocability. But at the same time, users may have difficulty reading the entries, since collocability can be presented both in different parts of the entries and directly in the quotes themselves. When it comes to phrases, a complete dictionary entry is not always necessary, but there is a need in examples of real data, properly described and accompanied by an assessment of the correctness or frequency of use in speech.
The Russian National Corpus [RNC, 2003-2019] has a number of tools (n-grams search with statistical evaluation, lists of set phrases and fixed expressions, and lexical graphs) and dictionaries based on it. Here we can name the Dictionary of Russian Abstract Nouns’ Verbal Collocability [Biryuk, Gusev and Kalinina, 2008]. It presents information for over 10,000 phrases of the following structures: 1) noun+verb; 2) verb+noun; 3) verb+adjective+noun. The dictionary uses lexical functions for describing and classifying collocations and their senses. The collocation candidates were initially extracted from the parsed subcorpus of RNC. The nouns occupied one of the following syntactic positions: 1) direct object of a transitive verb; 2) indirect object of a transitive verb; 3) subject of an intransitive verb. The Dictionary of Russian Idiomatic Expressions [Kustova, 2008] presents word combinations with the meaning of a high intensity. The author describes lexical items with restricted collocability, i.e. idiomatic expressions that should be learned by native and non-native speakers. The initial word list was based on RNC and printed dictionaries. One can find phraseological units (kruglyj sirota ‘orphan’), collocations (plakat’ navzryd ‘to sob violently’), idiomatic expressions (glubokoo gorchen ‘deeply disappointed’) and semantically motivated free phrases (chrezvychajno ogorchen ‘extremely disappointed’). 5
The dictionaries mentioned in the previous section were used as a source for the database. Below we present the pipeline for data processing and discuss the principles of the database.
The analysis of the dictionaries suggests that there is a need in a unified format that can be used for describing data. An entry has the following characteristics: a collocation; a syntactic model; a dictionary index (if applicable); references to the dictionaries (if applicable); frequencies in corpora (in ipm); references to the dictionaries; values of the association measures; visualization.
The database includes two kinds of collocations, i.e. dictionary and statistical ones. The former present in various lexicographic sources whereas the latter can be extracted automatically from corpora. In order to obtain data on co-occurrences in the Russian language, we process AraneumRussicum Maximum corpus (about 15 billion words), which was created automatically and is based on web texts of different genres being one of the largest collection of Russian texts [Benko andZakharov, 2016]. We use a statistical approach for automatic extraction of word combinations from corpora that implied several association measures (t-score, MI, log-likelihood).
At present the data from the described dictionaries are available in the database. We analyzed dictionary entries and extracted collocation candidates either from phraseological sections or as separate items written with a special type. Altogether the list comprises about 18,500 collocations. The initial merged lists were longer but a certain part of the dictionaries word lists overlapped. These collocations received the corresponding index, i.e. the number of the dictionaries they were presented in. The index indicates the given items are highly reproducible in speech and can be used by language learners.
Preprocessing involved also morphological analysis in order to present collocations in their canonical form but grammatical information was also preserved. 5.1
Altogether we extracted 11,210 phraseological units that were marked with a special diamond ‘ 0symbol:T hetotalamountof theheadwordsthathadphraseologicalunitsintheirentrieswas5; 955(about7:4%f romtheenti grams(makovoj rosinki v rot ne brat0`starving0):Inthedatabasewef ocusedonthef ollowingmodelsf romthedictionary : adjective + noun.
Examples: bolotnyj gaz ‘marsh gas’, morskay amilja ‘nautical mile’, gomericheskij khokhot ‘Homeric laughter’, zolotaj asvad’ba ‘golden wedding’, zhivoj um ‘nimble mind’.
verb + noun / verb + preposition + noun.
Examples: bit’ kartu ‘to cover a card’, boltat’ jazykom ‘to jabber away’, nesti otvetstvennost ’ ‘to be responsible’, oblechd overijem ‘to trust’, stavit’ tochku ‘to end’.
Among other models found in the candidates list we can name the following ones: 1) noun + noun (kniga pocheta‘ book of honorable guests’); 2) preposition + noun (bez umolku ‘nonstop’); 3) pronoun + noun (nechego skazat’ ‘nothing to say’).
The phraseological units found in the dictionary often enumerate certain semantic groups lexical items of which collocate with a keyword. Examples: v pylu (srazhen’ja, bitvy, sporaetc) ‘heat of the (fight, battle, debate)’, v rassrochku (kupit’, prodat’) ‘in instalments (to buy, to sell)’, v storonu (skazat’, proiznesti) ‘aside (to say, to utter)’. We considered each unit with an appropriate word as a separate phrase. Such an approach enabled us to enrich the data and also in future we plan to use special semantic tags to make the information clear for language learners.
Collocations can be also embedded into longer phraseological units and thus were treated separately. The dictionary entries do not represent the synonymic expressions separately but by commas or conjunctions. E.g. v khristianskij vid privesti or khristianskij vid pridat ‘ironic, to put in order’. In the database thus we have two different records with verbs and one record with noun phrase (khristianskij vid ). 5.2
The initial preprocessing involved OCR procedure and extraction of the collocations written in capital letters. This resulted in a total sum of 3,058 collocation candidates. We also extracted collocations from quotations that were not highlighted in the entries and marked them with the asterisk ‘*’ symbol. These phrases vary in their fixedness but they can be an important source of information. Hence additional 232 collocations were added to the database. Examples: ozhivlennajabeseda ‘lively talk’, vypusknojekzamen ‘final examination’, proizvoditel’nost’ truda ‘workforce productivity’. Polysemic headwords marked in the dictionary with digits were preserved in the database as separate entries. Examples: otvlekat’ vnimanije 1 ‘to distract attention’ (meaning: ‘showing interest’) vs okazyvat’ vnimanije 2 ‘to give attention’ (meaning ‘care’).
In the dictionary 2,044 verb pairs (imperfective and perfective aspects) are given via slash (e.g. vnosit’ /vnesti jasnost’ ‘to clarify something’). They were separated in order to make two records resulting in total 4,088 verb phrases.
The dictionary lists a large number of collocations that slightly differs in their meaning and can be called synonyms to a certain degree. They vary either in wordforms (case or number) or in prepositions. Examples: nakhodit’sja vo vlasti vs nakhodit’sja pod vlast’ju‘ to be in smb’s power’, ostavit’ pamjat’ o sebevs ostavit’ pamjat’ po sebe ”to leave memories’. 5.3
The preprocessing of the data extracted from the given electronic dictionary was not so elaborated as it was the case with other dictionaries. The following models are present in the dictionary: adjective + noun.
adverb + verb.
Examples: glubokaja blagodarnost’ ‘deep gratitude’, krupnyj uspekh ‘enormous success’, nadezhnaja opora ‘firm support’, ogromnyje perspektivy ‘exciting prospects’, jaryj storonnik ‘stickler’.
Examples: aktivno obsuzhdat’ ‘to discuss actively’, bezgranichno verit’ ‘to trust implicitly’, vovsju moch nestis’ ‘to rush headlong’.
Examples: bezumno interesno ‘extremely interesting’, nesterpimo dolgo ‘unbearably long’, sovsem nedavno ‘just recently’.
Examples: absoljutno ochevidno ‘absolutely obvious’, zhutko prijatno ‘extremely pleasant’, iskljuchitel’no vazhno ‘exceptionally important’.
Examples: gluboko porjadochnyj ‘totally honest’, zhutko bogatyj ‘awfully rich’, otchajanno smelyj ‘extraordinarily courageous’, khorosho znakomyj ‘well-known’, jarkovyrazhennyj ‘strongly marked’. adverb + adverb. adverb + predicative. adverb + adjective. particle + noun.
Example: prjamochudo ‘really badly’.
We confined ourselves to noun phrases and lemmatized them. The total amount was about 7,000 collocations. 5.4
The dictionary comprises more than 3,770 collocations with perfective and imperfective verb pairs and 383 with only imperfective verb forms. After processing the dictionary data, excluding prepositional phrases and representing phrases with perfective and imperfective verbal forms as different pairs we had a list of 7,923 items. 5.5
The volume of the verified (dictionary) collocations depends on the volume of the dictionaries that are used. As it was mentioned above dictionaries’ volume is not sufficient enough to describe vaster groups of lexis and hence to give a broader coverage that could be comparable to a word list of an explanatory dictionary (it often counts several thousand).
Table 1 demonstrates the number of collocation candidates (noun and verb phrases) from four dictionaries. It can be noticed that an electronic dictionary based on corpus data has larger volume compared to a printed dictionary.
The Dictionary of the Russian Language [
This work was supported by the grant of the Russian Science Foundation (Project No. 19-78-00091).
We made an overview of the database that includes data on collocability from various lexicographic sources. It is already available online (http://collocations.spbu.ru).
Even though Russian cannot be counted as a less resourced language there is a certain need in online tools that can be used both in research and language learning. The low overlap between the dictionaries (about 15%) suggests that they describe different lexical units. The Dictionary of the Russian Language [DRL, 19811984] puts emphasis on phraseological units whereas other lexicographic sources pay attention to collocations. There are also other syntactic models that should be taken into account as well as examples from quotations. There is a need in further processing of other dictionaries.
It is planned to open free access to the system for a possible assessment by the users of the degree of stability of the structures found in corpora. This will allow getting an interpretation of the data by the speakers of the language and can be used in further improving the resource and other tasks, for example, when creating a specialized dictionary or in systems using machine learning. Also, the amount of illustrative material and information on collocability presented in lexicographic sources (and, accordingly, in the “gold standard”) may indicate the frequency of use of the unit and correlate with it. This may be required when developing teaching and learning materials for students of the Russian language. [Khokhlova, 2018] Khokhlova M. (2018) Building a Gold Standard for a Russian Collocations Database. In Proceedings of the XVIII EURALEX International Congress: Lexicography in Global Contexts. Ljubljana.
Pp. 863–869.