TERMINOLOGY AND THE PRESENT

Rūta MARCINKEVIČIENĖ

Vytauto Didžiojo universitetas

TERMINOGRAPHY AND CORPORA

Corpus linguistics, which emerged with the first corpus more than three decades ago, revolutionized general lexicography. Today, explanatory dictionaries are compiled to a greater or lesser extent on the basis of corpora. In this respect, terminography remains neglected even in countries that are more advanced in corpus linguistics—let alone newcomers such as us. This article will attempt to show how important, indeed indispensable, a corpus is in compiling terminology dictionaries.

The era of computer technologies, and with it computer corpora, clearly separated traditional terminology and terminography from the new. Terminologists’ views began to differ considerably on several matters: the very concept of a term, sources of terms, criteria for distinguishing terms from non-terms, terminologists’ qualifications, the standardization of terms, and some other issues. As the fields of the specialized sciences expanded and became increasingly intertwined, the workload of terminologists grew substantially, since traditional methods made it scarcely possible to keep pace with scientific and technological progress. This is precisely why terms came to be viewed differently: usage, rather than the norm, is now emphasized.

Traditional terminology, which emerged at the beginning of this century in response to the enormous technological breakthrough, was chiefly concerned with naming concepts, organizing the great variety of terms that had appeared in a short time, and systematizing and standardizing the terminology of individual branches and fields of science. From the very beginnings of terminology as an independent science, terms were viewed differently from other words. The founder of terminology, Wüster (cited in Pearson 1998: 10), argued that terms should be treated differently from words in general language, since terminology begins with the analysis of concepts. And concepts, as we know, exis-

5

t independently not only of the terms that name them, but also of different languages. Terms are nothing other than names for concepts, ideally interacting with the concepts themselves in a one-concept–one-term relationship. Terms were defined primarily with regard to the place of the concepts they named within a hierarchical system of concepts and their relations to other concepts. The basis of terminology consisted of terms created and/or standardized and approved by specialists in the relevant field, and endowed with a clearly defined, established—and therefore protected—meaning. It is therefore no surprise that traditional terminologists adhering to prescriptive principles (cf. Wüster, for whom Sein-Norm was far more important than Ist-Norm) were wholly uninterested in the actual state of the language of individual scientific disciplines and the authentic use of terms.

When proceeding from a concept to a term and thereby compiling the terminology of each individual branch according to the onomasiological principle, terminologists of that time did not—and could not—face the problem of identifying terms, only that of creating them. Moreover, at the time, individual branches of science were still fairly autonomous. Terminologists had also divided up the fields more clearly and were familiar with the terminology of relatively isolated disciplines. As common ground emerged, their representatives began to take an interest in terms; in addition, new people entered terminology, not necessarily specialists in a particular scientific field and/or linguists—that is, programmers, creators of terminology databases and electronic dictionaries, and others. Since actual language use and the number of new terms had long since outstripped the available lists of approved terms, terminography in rapidly changing scientific fields, already lagging far behind due to the traditional card-index method, no longer enabled newcomers to rely on terms registered earlier. They began collecting them from texts; the text itself became the starting point of the terminologist’s work (Rogers et al. 1994: 843). In addition, their actual usage is examined, and this often differs greatly from the ideal sought by language standardizers. This is where it becomes unclear what should be considered a term, when a word becomes terminologized, which uses of a word count as terminological, and how to distinguish a term from a non-term. Thus, terminology is no longer such a clearly defined and isolated part of the lexicon. It forms a group of words that is fairly difficult to distinguish, with prototypical members at its center, always perceived as

6

After the central category come those words which acquire the status of terms only under certain circumstances, in a particular context. At the periphery there are also words that do not seem like terms at first glance, but should nevertheless be regarded as such. This applies, for example, to common-language vocabulary found in texts intended to popularize science. For this and other reasons, it is very important for terminographers to identify terms, extract them from authentic texts, and describe them. This is precisely why corpora and the software that facilitates working with them are becoming so important. Attention to actual usage leads terminologists toward description rather than prescription, while the divide between prescriptivism and descriptivism separates terminography into traditional and modern approaches.

Why a corpus

As a source of terms, a corpus is useful to terminographers for many reasons. Authors who advocate its creation and use point to the main ones. Perhaps most importantly, a single corpus immediately provides three things terminographers need: the terms themselves, their definitions—that is, domain-specific information about the concepts themselves—and examples of usage, and thus information about frequency of use, term combinations, and so on (Sager 1990, Meyer et al. 1996, Pearson 1998). Of course, researchers who already have a preliminary list of terms can also use a corpus. In that case, they obtain domain-specific and linguistic information from it. However, terms obtained from a corpus have one major advantage: they are systematic and interconnected, since individual texts present sets of terms used by individual authors, each of which has been carefully considered and whose terms fit harmoniously together. It is precisely for this reason—that the terms obtained should reflect distinct systems—that entire texts, rather than extracts or samples, must be included in the corpus (Meyer et al. 1996: 268). A corpus consisting of complete texts is also essential for domain-specific information, which would be lost if one relied on portions of texts.

Today, the corpora created by terminographers are still used mainly to gather information about terms already known. Researchers check

7

whether they are still in use and, if so, how; whether their usage patterns have changed and, along with them, their meanings; and whether the scope of the concepts they designate has changed with the emergence of new terms. However, corpora are much more necessary for obtaining new terms, which is perhaps why term identification and their automatic or semi-automatic extraction are now among the most popular topics in electronic terminography. In this respect, the terminographer differs from the lexicographer. The latter does not ask what a word is, only which words to include in which dictionaries; the terminographer, by contrast, quite often asks what a term is. New terms extracted from a corpus and stored and managed in special databases have another characteristic: their sources are precisely identified and dated, making it possible to trace the emergence, spread, and disappearance of terms.

If we conceive of the terminographer’s final product as an electronic database (DB), or a printed dictionary prepared on its basis, then the corpus is important at every stage of preparing that DB. Computers encompass the terminographer’s entire work cycle, from computer corpora, through specialized term search and identification programs, to the creation of terminological databases, the description of terms, and its distribution in electronic form or as a traditional book (Ahmad et al. 1994: 267).

The terminographer’s work cycle proceeds as follows. At the first stage, a corpus is compiled from an electronic text archive according to certain principles; terms are then extracted from it and transferred to a DB, where they are described. That is, separate DB fields contain examples of their typical usage taken from concordances, as well as definitions obtained from the same source. Subsequently, on the basis of the linguistic and domain-specific information obtained from the corpus, words related to the term being described are identified: synonyms, antonyms, hyponyms, and hypernyms; cross-references to their entries are provided, along with the term’s domain affiliation and place in the general conceptual system. Grammatical and pragmatic features, restrictions on combinations or other usage restrictions, information about whether the term is recommended, and, if intended, translation equivalents are also provided (for more on the characteristics of terminographic description, see Sager 1990: 142–163). Thus, a corpus is indispensable from the very beginning, that is, from becoming acquainted with the subject field itself

8

to the final stage—the completed dictionary entry. Today’s terminologists are highly skeptical about computerizing existing printed dictionaries that were not compiled on the basis of corpora, that is, transferring them to databases (Sager 1990: 141).

Another reason for preparing special corpora for terminographers is that authentic language examples are even more important to terminography than to lexicography. They are much more important than introspection because, unlike a lexicographer, a terminographer is not a specialist and does not create texts in the language variety being described. When faced with a problematic case, they cannot ask, “How would I use this word?” but only, “What does this word mean to an expert in the field, and how would a scientist, technologist, or expert use it?” Thus, because their own linguistic intuition is of limited use, terminographers depend much more on corpora than lexicographers do. Atkins et al. divide all corpus users into three groups: (a) those interested in the language of texts, (b) those interested in the content of texts, and (c) those interested in texts as texts, that is, their structure, creation, and so on (cited in Meyer et al. 1996: 269). Lexicographers belong to the first group, while terminographers belong to the first two, since both linguistic and subject-matter knowledge are important to them. Once the importance of corpora to terminology and the opportunities they offer have been understood, it is necessary to define what a terminographer’s corpus should be like.

What kind of corpus

As mentioned, a terminographer needs two equally important things from a corpus: linguistic and subject-field information; it is therefore important that both can be obtained from it. From the linguistic point of view, the corpus must be such that terms are presented clearly and correctly, preferably with definitions. It is desirable for there to be as many terms as possible, and as varied as possible; ideally, all those created and used by a particular linguistic community. It is also important that terms be used in rich and diverse contexts. From the subject-field point of view, the corpus should include texts that bring out the content of the concepts denoted by the terms, as well as their interrelationships and dependencies. It is very important that the terms be defined in those texts.

9

For the reasons given, experts recommend including in corpora the full range of possible texts, considered from the communicative and pragmatic perspective, from the simplest to the most complex. Pearson divides texts in specialized language into four groups according to the likelihood of finding terms in them: a) those in which the expert’s addressee is a partner with equivalent knowledge—a fellow expert, that is, articles in peer-reviewed scholarly journals and monographs; the density of terms is highest in these texts; b) texts addressed by an expert to a specialist in the same field but at a lower level than the expert, for example, various instructions and textbooks; these contain somewhat fewer terms; c) popular science texts written for people unfamiliar with the specialized field. Such texts deliberately avoid complex or any terms, preferring to explain, paraphrase, and use analogies and comparisons; thus, if the author so wishes, they can do entirely without terms; d) texts corresponding to teacher–student communication: textbooks and teaching materials. These contain terms, but they are used somewhat differently. An important feature of these texts is that terms are explained (Pearson 1998: 36–40).

Describing the texts in a terminographic corpus solely from the communicative and pragmatic perspective is not enough. Other textual parameters and proportions are just as important for the quality and representativeness of a corpus. When asked how many texts need to be collected in order to rely on the data of the corpus compiled from them, experts answer: “Unlike general-purpose corpora, in this case quality is far more important than quantity” (Meyer et al. 1996: 268). For understandable reasons, specialized corpora are generally smaller than general-purpose ones. Yet, although smaller, they are made up of complete, unabridged texts. Moreover, they must cover the entire field being represented, including all its subfields, as selected in advance by the terminographers. It is often difficult to draw a clear line between separate scientific fields, especially when a field is connected with broader subjects and corpus compilers are tempted to go deeper—that is, to include more texts written by experts for experts that undoubtedly belong to that field. Terminologists are therefore advised to consult experts in the field being described, so that they can recommend which texts to include in order to compile a corpus that is representative in every respect, and so that concepts of systemic importance and the terms denoting them are not overlooked along with the excluded texts

10

that are systematically important, along with the terms denoting them, and, most importantly, so that the priorities of a single compiler do not dominate (Ahmad 1993: 60). In the case of a terminographic corpus, the principle sometimes applied by lexicographers to general-purpose corpora—“What I have, I include”—is entirely inappropriate.

Taking into account corpus typology and the principal corpus types in terms of continuity—finite, continuous, and monitor—a terminographic corpus is recommended to be compiled as a collection of texts of the second, or better still, the latter type. The corpus would then be continuously supplemented with new texts, while older ones would either be retained (in the case of a continuous corpus) or moved to an archive if the corpus is a monitor corpus. A continuous corpus would keep growing, and its preselected structural principles would not necessarily be maintained, since changing publishing trends could cause newly added texts to tip the balance in one direction or another. The proportions of a monitor corpus would remain unchanged, and new texts would occupy only the amount of space originally allocated to texts of the corresponding type. In any case, the corpus must be continuously updated, since terms become obsolete much faster than everyday words.

Having considered the general questions of corpus size, nature, and subject-field boundaries, and having made decisions on them, the next step is to determine which specific texts to include so that the corpus is representative and balanced. In each particular case, the genre, date of writing, author, and linguistic status of a text must be taken into account, with the aim of achieving the greatest possible diversity. For it is genre diversity and a broad range of texts, from serious to popular, that make it possible to select both scholarly, formal terms, among which authorial neologisms abound, and popular terms, which are much more widespread and established in everyday speech.

When assessing the age of texts, preference is given to more recent ones; however, older texts may also prove valuable, especially if considered from the subject-field perspective. Authorship should also be highly varied, so that the stylistic features of one or another heavily represented author do not tip the balance. Although the stylistic diversity of scholarly texts is not very great compared, for example, with that of fiction, an author’s style is important not only in terms of the use of linguistic units but also in their selection. It is therefore unsurprising that different scholars, especially if they belong to different schools of scientific thought, use var-

11

such terms. And for the purposes of standardizing and normalizing terminology, as well as searching for neologisms, the greatest possible diversity of terms in the corpus is very important. The last, but no less important, consideration is also the linguistic status of the texts. Preference is given to original rather than translated texts, written by native speakers, and subsequently reviewed and edited.

Corpus analysis tools for terminographers

Over the past decade, various software tools have been developed to make it easier for lexicographers to search for words and perform other corpus analyses. Although the work of lexicographers and terminographers differs in some respects, they use almost the same software capabilities, namely: a) word-frequency lists; b) concordances, that is, all occurrences of the word under study in a context of a certain size, usually one line; and c) statistics on collocational partners (Meyer et al. 1996: 274). These tools help analyze the corpus from a linguistic perspective.

Alongside the linguistic software tools generally used by lexicographers, specialized programs suitable only for terminographers are gradually emerging. They differ from general-purpose programs not in linguistic but in subject-matter terms, since they are designed to analyze the information contained in a corpus. In this case, the corpus is treated not as a collection of contexts in which terms are used, but as a source of specialized knowledge in a particular field or branch of science. Several groups of such tools can be identified: a) programs for analyzing a text in terms of the information it contains—for example, searching for meaningful words and information-rich passages that indicate relations and dependencies among concepts; b) programs that select corpus passages needed by a terminographer, such as definitions of terms; c) programs that help analyze concepts designated by terms and facilitate the organization of knowledge obtained from the corpus (knowledge acquisition systems). Although this last group is entirely new and highly promising, it has so far received too little attention in specialized literature. Much more has been written about the possibility of semi-automatically extracting definitions of terms found in a corpus (Pearson 1998: 121–204).

12

A morphologically annotated corpus—that is, one in which all words are tagged according to their parts of speech—is essential for collecting both definitions and multiword terms. The initial stage of the work must be carried out by the lexicographer, who reads the texts and collects defined terms. On the basis of this information, the lexicographer later constructs morphological patterns for definitions—all the combinations of part-of-speech sequences that occur in definitions. The computer then searches the entire corpus for such word combinations. The result is not yet a list of terms and their definitions, but a list of candidates for defined terms. This is then given to a subject expert to review and remove irrelevant sentences. To reduce the number of unnecessary morphological combinations that occur by chance, the computer is supplied with lexical items that occur most often in definitions—frequent definition words. Multiword terms can be obtained from a corpus in a similar way; this is the principle behind the first-generation automatic term-identification programs LEXTER (Bourigault 1995) and TERMINO (Lauriston 1994).

Naturally, processing a corpus in this way raises certain problems. First of all, by no means are all the selected word combinations terms. This problem is addressed by compiling and entering into the computer a list of unwanted words (stop-list), which are excluded from lists of candidate terms. In addition, it often remains unclear which words belong to the multiword term itself and which are merely part of its surrounding context. But these issues are not always clear when terms are collected manually, either.

Semi-automatic method of term selection

To illustrate the possibilities that even the simplest software tools offer terminographers, we selected the electronic version of V. Labutis’s “Syntax” (T.I, V., 1994) for analysis and examined the text using the package of tools for linguists developed by Mike Scott, WordSmith Tools, version 2.0 (Oxford, 1997). The purpose of the statistical analysis was to obtain several lists of words or word combinations from which candidates for terms, and the terms themselves, could later be selected.

The statistical indicators for the text analyzed are as follows: the entire text consists of 37,319 words and word forms; after duplicates are removed, what remains is

13

9,249 words and their forms. The type/token ratio is 24.78. The entire text contains 2,364 sentences, with an average sentence length of 15.78 words.

First, the simplest operation was carried out: a frequency list of all the words in the text was compiled, allowing us to expect terms among the most frequently used words. As can be seen from the excerpt of the list presented, the text consists of words and word forms arranged in order of frequency. Since the text has no grammatical tags indicating the part of speech to which each word form belongs, the list is not lemmatized; thus, different grammatical forms of the same word appear separately. For the sake of brevity, both uninflected words and the different inflected forms of words will henceforth be referred to simply as words.

Table 1. The 20 most frequent words and word forms in the text

No.

Word and word form

Frequency in the text under study

Share of the text occupied (%)

1.

IR

1,319

3.53

2.

SU

479

1.28

3.

AR

390

1.05

4.

SENTENCE

288

0.77

5.

THIS

284

0.76

6.

WORDS

258

0.69

7.

CAN

243

0.65

8.

O

229

0.61

9.

IŠ

206

0.55

10.

SENTENCES

196

0.53

11.

Į

194

0.52

12.

IS

194

0.52

13.

HOW

187

0.50

14.

NE

182

0.49

15.

ONLY

171

0.46

16.

ACCORDING TO

170

0.46

17.

TO BE

155

0.42

18.

JŲ

143

0.38

19.

SENTENCES

143

0.38

20.

PHRASES

141

0.38

14

As usual, a text consists of a relatively small number of words used many times and a large number of words used rarely. Words used once account for 62% of all word occurrences. They usually make up half the text; after lemmatizing the list, apparently that proportion would remain, since derived word forms would be grouped under their base forms. Hapax words and rare words, that is, those occurring from 2 to 10 times, are of no interest to a terminologist, since the probability of finding terms among them is low. What remains is a relatively short part of the list, just 6% of frequently used words in the text, from which it is already worthwhile to collect single-word terms.

Diagram

Distribution of text words and word forms by frequency of occurrence (number of occurrences)

Words and their forms occurring more than 10 times: 6%

Words and their forms occurring from 2 to 10 times: 32%

Words and their forms occurring once: 62%

If these usage data were presented as a curve rather than a diagram, it would become clear that 10 occurrences constitute a very distinct boundary, a threshold beyond which words are used much more frequently than those below it. A more thorough analysis showed that, of the 599 words occurring 10 or more times, as many as 239 are single-word terms or parts of multi-word terms—in other words, words important to the terminographer. These include the names of parts of speech, cases, linguistic phenomena, units, and branches of linguistics.

The advantage of a general frequency word list is the very clear usage-frequency threshold mentioned above, while it has several drawbacks. One of them is that the list consists of individual words, and, as is well known, there are not that many single-word terms. On the other hand, most of the new terms that have not made it into dictionaries and are therefore of particular interest to terminologists are precisely multi-word—

15

words. The second drawback is that a large part of the list is occupied by very frequent function words, which are also uninformative for terminologists. This drawback can be remedied by compiling a somewhat different list of content words. Content words are extracted from a text in a slightly more complex way, by comparing the text under study with a much larger corpus—in our case, the 55-million-word computerized corpus of Vytautas Magnus University. If a word occurs in the text under study relatively more often than it does on average in the large corpus, it is considered a content word of that text. Put more simply, if the text is about cinema, this word will occur in it much more often and will therefore appear on the list of content words. As the table shows, there are far fewer function words here, since their frequency in the text and the corpus is roughly the same. Thus, more distinctive words in the text under study—terms, in the case of a scholarly text—come to the fore. Unfortunately, as in the first list, they are given individually rather than in phrases.

Table 2. The 20 most frequent content words and word forms

No. | Word and word form | Frequency in the text under study | Frequency in the corpus | Content-word coefficient 1. | SENTENCE | 288 | 707 | 2,974.8; 2. | SENTENCES | 196 | 370 | 2,108.4; 3. | WORDS | 258 | 6,504 | 1,554.3; 4. | SENTENCES | 143 | 386 | 1,453.6; 5. | PHRASES | 141 | 405 | 1,418.4; 6. | CLAUSES | 103 | 130 | 1,171.5; 7. | PHRASES | 103 | 591 | 909.2; 8. | CLAUSE | 77 | 124 | 847.4; 9. | IN THE SENTENCE | 85 | 310 | 819.6; 10. | CLAUSES | 73 | 106 | 815.0; 11. | OF THE CLAUSE | 71 | 84 | 814.2; 12. | SENTENCE | 91 | 528 | 801.4; 13. | BY A CLAUSE | 69 | 78 | 795.8; 14. | SYNTAX | 68 | 163 | 704.7; 15. | SENTENCE | 73 | 634 | 588.1; 16. | SENTENCES | 61 | 315 | 550.1; 17. | OF THE PREDICATE | 44 | 60 | 495.3; 18. | BY SENTENCES | 49 | 205 | 460.5; 19. | OF A COMPOUND | 42 | 75 | 455.4; 20. | BY A PREDICATE | 36 | 36 | 421.4

16

The third capability of software based on statistical methods is to extract from the text under study those word combinations that occur in the text in an identical form two or more times. The program can detect sequences as long as eight words. Naturally, there are not many formations of this length in a text. Fixed combinations of four, three, and especially two words are much more numerous. In V. Labutis’s text under study, there are 2 eight-word combinations used at least twice, 6 seven-word combinations, 23 six-word combinations, 65 five-word combinations, 211 four-word combinations, 756 three-word combinations, and as many as 2,695 two-word combinations. Sequences of words of different lengths are presented below; combinations whose components could be considered compound terms are highlighted in bold, and the number of occurrences of each sequence in the text is given alongside.

EIGHT WORDS

IT CONSISTS OF NO FEWER THAN TWO INDEPENDENT WORDS 2

CONSISTS OF NO FEWER THAN TWO INDEPENDENT WORD FORMS 2

SEVEN WORDS

IT CONSISTS OF NO FEWER THAN TWO INDEPENDENT 2

NO FEWER THAN TWO INDEPENDENT WORD FORMS 2

CONSISTS OF NO FEWER THAN TWO INDEPENDENT WORDS 2

USE OF WORD FORMS WITH OTHER WORD FORMS 2

SIX WORDS

NO FEWER THAN TWO INDEPENDENT WORDS 4

FORMS OF VERBS AND WORDS FUNCTIONING AS PREDICATES 3

IN THE ACADEMIC GRAMMAR OF THE LITHUANIAN LANGUAGE LKG III 2

TWO OR SEVERAL FORMS OF INDEPENDENT WORDS 2

APPLICATION OF FORMS TO OTHER WORD FORMS 2

AND WITH SOME PREPOSITIONAL CONSTRUCTIONS 2

A CLAUSE OF A SENTENCE OR A CLAUSE OF A COMPLEX SENTENCE 2

WITH OTHER CASES AND PREPOSITIONAL CONSTRUCTIONS 2

WITH CASES AND PREPOSITIONAL CONSTRUCTIONS 2

A WORD FORM APPLIED IN SOME WAY TO 2

APPLICATION OF WORD FORMS TO OTHER WORDS 2

FIVE WORDS

FEWER THAN TWO INDEPENDENT WORDS 4

VERBS AND WORDS USED AS PREDICATES 4

17

CASES AND PREPOSITIONAL CONSTRUCTIONS 3

PREDICATIVE STRUCTURES FORMING A COMPLEX SENTENCE 3

OR SEVERAL FORMS OF INDEPENDENT WORDS 2

A SEPARATE MESSAGE ABOUT A REAL-WORLD SITUATION 2

A CLAUSE OR A COMPONENT OF A COMPLEX SENTENCE 2

SERVES AS A DEPENDENT COMPONENT OF A NOUN PHRASE 2

USE OF FORMS WITH OTHER WORDS 2

CAN ALSO BE PART OF A COMPLEX SENTENCE 2

AND BETWEEN THE PREDICATIVE STRUCTURES OF A COMPLEX SENTENCE 2

AND FORMS OF WORDS USED AS SUBJECTS 2

AS FORMS OF TWO INDEPENDENT WORDS 2

IN OTHER CASES AND IN PREPOSITIONAL CONSTRUCTIONS 2

OF A CASE OR A CASE WITH A PREPOSITION 2

IN RELATION TO THE GRAMMATICAL CATEGORY OF PERSON 2

SYNTAGMATIC RELATIONS AND WORD COMBINATIONS 2

THE CLASSIFICATION HAS OFTEN BEEN CRITICIZED 2

WITH A SECOND-DEGREE SUBORDINATE CLAUSE 2

WITH WHICH PREPOSITIONAL CONSTRUCTIONS 2

WITH OTHER CASES AND PREPOSITIONAL CONSTRUCTIONS 2

WITH THE DATIVE AND WITH THE INSTRUMENTAL 2

WITH A SUBORDINATE CLAUSE OF THE WHOLE SENTENCE 2

THESE ARE MOST OFTEN VERBAL COMBINATIONS WITH 2

APPLICATION TO OTHER WORD FORMS 2

SUBORDINATE CLAUSES OF THE SAME DEGREE 2

HAVE A SHARED SECONDARY SENTENCE ELEMENT 2

ROLE WITH RESPECT TO THE ENTIRE COMPLEX SENTENCE 2

ALTHOUGH A WORD FORM IS APPLIED 2

APPLICATION OF WORD FORMS TO OTHERS 2

WORD COMBINATION OR OTHER CONSTRUCTION 2

FOUR WORDS

AND WITH PREPOSITIONAL CONSTRUCTIONS 5

AND WORDS FUNCTIONING AS PREDICATES 4

WHOSE MAIN COMPONENT IS 4

LESS THAN TWO INDEPENDENT 4

NO FEWER THAN TWO 4

A WORD FORM FUNCTIONING AS THE SUBJECT 4

FUNCTIONING AS THE SUBJECT AND PREDICATE 4

FORMS OF TWO INDEPENDENT WORDS 3

COORDINATING AND SUBORDINATING 3

CASES AND WITH PREPOSITIONAL 3

OF THE WORD FUNCTIONING AS A MAIN COMPONENT 3

PREDICATIVE STRUCTURES FORMING A SENTENCE 3

WITH THE GRAMMATICAL CATEGORY OF PERSON 3

A WAY OF EXPRESSING WORD RELATIONS 3

A SECOND-DEGREE SUBORDINATE CLAUSE 2

SEE THESE CONCEPTS 2

18

ABOUT THIS IN THE EXERCISES 2

OR THE CLAUSE OF A COMPLEX SENTENCE 2

REVEALS ANOTHER PREDICATIVE COMPONENT 2

A SHARED SECONDARY CLAUSE ELEMENT 2

OF MANNER, QUALITY AND QUANTITY 2

AN ADJECTIVE FORMS A PHRASE WITH 2

AS A DEPENDENT COMPONENT OF A NOUN PHRASE 2

THREE WORDS

MAY BE 14

WITH PREPOSITIONAL CONSTRUCTIONS 11

MORE OR LESS 10

LITHUANIAN SYNTAX 9

SENTENCES MAY BE 9

VERBAL PHRASES WITH 8

COMPONENTS MAY BE 7

WHOSE MAIN COMPONENT 7

NOMINAL PHRASES WITH 6

AT LEAST 6

COORDINATION AND SUBORDINATION 6

SENTENCES MAY BE 6

FORMS OF INDEPENDENT WORDS 6

BY VERB AND PREDICATE 6

OF TWO INDEPENDENT WORDS 5

TWO-WAY RELATIONSHIP 5

TWO WORDS

MAY BE 118

AND OTHERS 41

LITHUANIAN LANGUAGE 36

WORD FORMS 35

ALSO 34

COMPLEX SENTENCE 32

SENTENCE PARTS 31

WORD FORMS 31

NOT ONLY 29

WORD COMBINATIONS 26

COMBINATIONS WITH 25

PREDICATIVE CONSTITUENTS 24

MAY HAVE 23

NEVERTHELESS 23

OF THE MAIN COMPONENTS 21

PREDICATIVE CENTER 20

PREDICATIVE COMPONENTS 20

AS IF 18

COMPLEX SENTENCES 18

19

CAN FORM 17

MAIN COMPONENT 17

WITH PREPOSITIONAL CONSTRUCTIONS 17

WITH THE GENITIVE CASE 17

AS WELL AS 16

LANGUAGE SYSTEMS 16

OF THE DEPENDENT COMPONENT 16

OF COMPLEX SENTENCES 16

IN ADDITION 15

AND ACCORDING TO 15

OF A WORD COMBINATION 15

As can be seen from the examples given, longer word combinations are more like expressions or phrases, while shorter ones are syntactic word combinations. They are not uniform

in their degree of fixedness either, as shown by the number of occurrences: for example, some two-word combinations occur 118 times even in a text this short. Moreover, some word chains coincide entirely with compound terms, e.g.: second-degree subordinate clause, bidirectional relation, language system, while in other cases the terms have to be extracted from them, e.g.: the other predicative constituent reveals, a word form is in some way applied to a word functioning as the main constituent, etc. Although they need to be shortened, they are informative in their own way, since predicative words (omitted in the examples), which do not belong to the term itself but to its context, provide important information about combinability and usage patterns.

Like the preceding lists, the latter list is not lemmatized, so the same multiword term recurs in a different grammatical form each time. Among other things, it also recurs in chains of word groups of different lengths—from the longest to the shortest: consist of at least two independent word forms, at least two independent word forms, independent word forms, word forms. Be that as it may, frequent repetitions do not hinder but help identify the most important, most commonly used compound terms. Even if the semi-automatic software tools mentioned above do not extract all the terms in the text under study, they nevertheless make it possible to identify the most important ones; and the fact that a terminographer reads and processes the lists compiled by the computer guarantees the quality of the final result.

20

Conclusion

Modern terminology, which seeks to collect, describe, and standardize the terms of various branches of science, cannot do without a computerized corpus and the other stages of preparing a computerized terminology bank associated with it: general-purpose or specialized software, a lexical database, etc. The corpus must be prepared so that it encompasses as wide a variety of texts as possible and thus represents the language of the branch of science being described. Even by using general-purpose software intended for lexicographers, one can automatically generate lists of words and word combinations that are candidates for terms, from which established and new terms are selected.

Received 1999 06 21

Literatūra

1. A h m a d 1994 – Ahmad, K., Davies, A., Fulford, H., Rogers, M. What is a term? The semi-automatic extraction of terms from text. In: M. Snell-Hornby, F. Pochacker, K. Kaindl (eds.). Translation Studies: An Interdiscipline. Amsterdam – Philadelphia, 1994, P. 267–278.

2. A h m a d 1993 – Ahmad, K. Pragmatics of Specialist Terms: The Acquisition and Representation of Terminology. In: Petra Steffens (ed.). Machine Translation and the Lexicon. Thirds International EAMT Workshop. Heidelberg, Germany, April 26–28. 1993. Proceedings. Berlin – Heidelberg, 1995, P. 51–76.

3. B o u r i g a u l t 1996 – Bourigault, D., Gonzalez-Mullier, I., Gros, C. LEXTER, a Natural Language Processing Tool for Terminology Extraction. In: Euralex ’96 Proceedings II, Goteborg, 1996, P. 771 – 780.

4. L a u r i s t o n 1994 – Lauriston, A. Automatic Recognition of Complex terms: Problems and the TERMINO solution. In: Terminology. Vol 1(1), 1994, P. 147–170.

5. M e y e r 1996 – Meyer, I., Mackintosh, K. The Corpus from a Terminographer’s Viewpoint. In: International Journal of Corpus Linguistics. Vol. 1(2), 1996, P. 257–285.

6. P e a r s o n 1998 – Pearson, J. Terms in Context. Amsterdam – Philadelphia, 1998.

7. R o g e r s 1994 – Rogers, M., Ahmad, K. Computerised Terminology for Translators: the role of text. In: M. Brekke, O. Andersen, T. Dahl, J. Myking. Applications and Implications of Current LSP Research. Proceedings of the 9^th European Symposium on LSP, Bergen, Aug. 2–6, 1993. Vol. II. Bergen, 1994. P. 840–851.

21

8. Sager 1990 – Sager, J.C. A Practical Course in Terminology Processing. Amsterdam – Philadelphia, 1990.

TERMINOGRAPHY AND CORPUS

Summary

The paper deals with role of a corpus in terminology in general and terminology in particular. It argues for the necessity to use not only corpus but also general purpose and specific tools for the extraction of terminology and compilation of terminological databases. Peculiarities of the terminographer’s corpus such as its contents, size and specificity are also discussed. In addition, a wide range of possible general purpose and specific tools for term extraction are presented here. Some of them, based on statistical analyses of a text, were applied for one Lithuanian text on syntax in order to show how different tools can produce different semi-automatic lists of terms.

22