Content uploaded by Antonio San Martín
Author content
All content in this area was uploaded by Antonio San Martín on Jul 16, 2018
Content may be subject to copyright.
893Lexicography in gLobaL contexts
The EcoLexicon English Corpus as an Open Corpus in Sketch
Engine
Pilar León-Araúz1, Antonio San Martín2, Arianne Reimerink1
1Department of Translation and Interpreting, University of Granada
2Department of Modern Languages and Translation, University of Quebec in Trois-Rivières
E-mail: pleon@ugr.es, antonio.san.martin.pizarro@uqtr.ca, arianne@ugr.es
Abstract
The EcoLexicon English Corpus (EEC) is a 23.1-million-word corpus of contemporary environmental texts. It
was compiled by the LexiCon research group for the development of EcoLexicon (Faber, León-Araúz & Reimer-
ink 2016; San Martín et al. 2017), a terminological knowledge base on the environment. It is available as an open
corpus in the well-known corpus query system Sketch Engine (Kilgarri et al. 2014), which means that any user,
even without a subscription, can freely access and query the corpus. In this paper, the EEC is introduced by de-
scribing how it was built and compiled and how it can be queried and exploited, based both on the functionalities
provided by Sketch Engine and on the parameters in which the texts in the EEC are classied.
Keywords: specialized open corpus, terminology, corpus exploitation
1 Introduction
Corpora have become a key element of almost all language studies, as any assertion about language
requires verication through real linguistic data to be deemed credible (Teubert 2005: 1). Having
access to general and specialized corpora is thus essential for anyone involved in research or any
professional activity related to language. However, many of these professionals do not have time to
compile large corpora. The EcoLexicon English Corpus (EEC) is a 23.1-million-word specialized
corpus of contemporary environmental texts. It was compiled by the LexiCon research group for the
development of EcoLexicon (Faber et al. 2016; San Martín et al. 2017), a terminological knowledge
base on the environment.1 In EcoLexicon, the EEC and its Spanish counterpart (together over 50 mil-
lion words) can be queried with pragmatic restrictions such as author, date of publication, target
reader, contextual domain, and keywords. However, its search engine does not provide all the func-
tionalities of the well-known corpus tool Sketch Engine (Kilgarri et al. 2014). This is why the EEC
was made available as an open corpus in Sketch Engine, which means that any user, even without a
subscription, can freely access and query the corpus.2 One very interesting module provided by the
query system is information extraction through word sketches, which are automatic corpus-derived
summaries of a word’s grammatical and collocational behavior (Kilgarri et al. 2010). Apart from the
built-in word sketches, Sketch Engine allows users to customize sketches for their specic needs. In
the case of the EEC, this has enhanced the extraction of semantic information.
In this paper, the EEC is introduced by describing how the corpus was built and compiled (Section 2),
and how it can be queried and exploited (Section 3), based on the functionalities provided by Sketch
Engine, the parameters in which the texts in the EEC are classied and the word sketches exclusively
created for the EEC. Finally, Section 4 oers some concluding remarks.
1 EcoLexicon is freely available at <http://ecolexicon.ugr.es>.
2 Certain advanced functionalities are only available for subscribed users.
894 Proceedings of the XViii eUrALeX internAtionAL congress
2 Creating the EcoLexicon English Corpus
The EEC is a 23.1-million-word corpus of contemporary environmental texts. It was rst created as
an internal tool for knowledge extraction while building EcoLexicon. However, it was made public-
ly available because it evolved to be a tool in itself that terminologists, translators or even experts
could exploit for dierent purposes (i.e. modeling, comprehension and production tasks) within the
specialized domain of the environment. As Sinclair (1991: 24) pointed out, we should not expect a
general reference corpus like the British National Corpus to adequately document specialized genres
and domains. It follows that we need more specialized corpora, compiled with enough texts and text
types to represent a knowledge domain, as they are more likely to document the conventions of the
genre and the concepts and terms of the domain.
Each text in the EEC is tagged according to a set of XML-based metadata, some of which are based
on the Dublin Core Schema, while others have been included to meet the needs of the research group.
Corpus metadata permit users to constrain corpus queries based on pragmatic factors, such as envi-
ronmental domains and target reader. Thus, for instance, the use of the same term in dierent contexts
can be compared. Tags are based on the following main parameters:
• Domain: the EEC encompasses all the domains and subdomains of environmental studies (e.g.,
Biology, Meteorology, Ecology, Environmental Engineering, Environmental Law, etc.).
• User: the corpus includes texts for three types of user, depending on level of expertise (i.e., ex-
pert, semi-expert, general public).
• Geographical variant: it comprises American, British, and Euro English.
• Genre: it covers a wide variety of text genres (e.g., journal articles, books, websites, lexicograph-
ical material, etc.).
• Editor: it distinguishes texts edited by scholars/researchers, businesses, government bodies, etc.
• Year: it includes texts from 1973 to 2016.
• Country: the texts are tagged according to the country of publication.
The EEC was processed and compiled in an internal application of the research group. Then it was
recompiled within Sketch Engine with the Penn Treebank tagset (TreeTagger version 3.3) and with
the EcoLexicon Semantic Sketch Grammar (ESSG) (León-Araúz & San Martín 2018; León-Araúz,
San Martín & Faber 2016), a CQL-based (Corpus Query Language) (Jakubíček et al. 2010) custom-
ized sketch grammar separate from the default sketch grammar. The ESSG was developed for the
extraction of semantic word sketches based on some of the most common semantic relations in termi-
nology: generic-specic, part-whole, location, cause, and function.
When a corpus is compiled with a collection of dierent pattern-based grammar rules such as the
above, new word sketches can be queried within the Sketch Engine (see Section 3.2). The ESSG
thus has three aims: (1) extracting semantic relations for building EcoLexicon; (2) oering seman-
tic word sketches in the EEC; and (3) providing other users with the possibility of reusing them in
their own corpora.3
3 Exploiting the EcoLexicon English Corpus
The combination of pragmatic, syntactic and semantic information that can be extracted from the cor-
pus makes the EEC an adequate resource for all kinds of end users with an interest in environmental
science, such as domain experts, professional writers, translators, terminologists, ESP researchers,
3 The latest version of the ESSG can be downloaded from <http://ecolexicon.ugr.es/essg/>.
895Lexicography in gLobaL contexts
etc., as stated above. Thanks to Sketch Engine’s automation capabilities, users are able to analyze and
extract a sizable quantity of linguistic data that would have been unmanageable in the past (Kosem
et al. 2014: 362). In the following sections, dierent queries will be provided by combining the main
functionalities of Sketch Engine with the parameters according to which the EEC is tagged.4
3.1 Search and Text Types
The feature Search is the main way to access concordances in Sketch Engine. Dierent types of que-
ries are possible (simple, lemma, phrase, word, character and CQL), and they can be combined with
the contextual lter, which allows the user to limit the lemmas that should appear around the word
or words of the query. Additionally, in the case of the EEC, any query performed through the Search
feature can be ltered according to text type based on the tagging of the EEC (domain, genre, editor,
etc.) (Figure 1).
The ltering by text type can be chosen manually for each query. However, the user can also create
subcorpora based on text types. For instance, a user may want to create a simple subcorpus for the
domains of Hydrology or Renewable Energy, or complex subcorpora ,such as one containing only
articles and books in British English from the domain of Biology for experts in the eld. Additionally,
the EEC comes with several subcorpora created by default (i.e. American English, British English,
Year 1973–1999, Year 2000–2009 and Year 2010–2016).
Figure 1: Sketch Engine’s Search and EEC Text types.
All these possibilities of query customization allow the user to retrieve, for instance, all the concord-
ances where recycle is a verb in texts addressed to the general public (lemma search ltered by user)
or where climate change occurs in Environmental Law texts (phrase search ltered by domain). Addi-
tionally, the Context option can be combined with any search, permitting the user to nd, for example,
all the concordances in Oceanography academic articles where the lemma wind appears in a window
of ±15 tokens of the lemma wave.
However, given that the EEC was recompiled with TreeTagger, it is possible to perform more ne-
grained queries in CQL, allowing for the formalization of grammar patterns in the form of regu-
lar expressions combined with POS-tags. CQL queries used together with text-type ltering are a
powerful tool to research the workings of environmental English. An example of a CQL query is
([tag=”N.*”] [lemma=”amount” & tag=”N.*”]) | ([lemma=”amount” & tag=”N.*”] [word=”of”]
[tag=”N.*”]), which nds concordances of the lemma amount either preceded by any noun or fol-
lowed by of and any noun. Figure 2 shows a sample of the resulting concordances limited to the
Meteorology subdomain.
4 Due to space restrictions, no instructions are provided. However, interested readers can consult the user-friendly Sketch Engine
manual at: < http://sketchengine.co.uk/user-guide/>
896 Proceedings of the XViii eUrALeX internAtionAL congress
Figure 2: Sample of the results for the CQL query amount preceded by a noun or followed by of and any
noun in the Meteorology subdomain.
With CQL queries, a user can also compare the frequency of dierent variants of multiword expres-
sions. For example, in the term geologic time scale, geologic can be replaced by geological and time
scale can be written as a single word. With the CQL query [lemma=”geologic.*”] ([lemma=”ti-
mescale”]|([lemma=”time”] [lemma=”scale”])) we can retrieve all the concordances where all the
variants appear, and with the Frequency – Node forms feature we can see which form is more frequent
(Figure 3).
Figure 3: Frequency of variants of geologic time scale in the EEC.
Another feature of Sketch Engine that permits users to fully exploit the EEC is Frequency – Text type.
With this feature, users can observe how language expression changes across dierent levels of ex-
pertise in the environmental domain. For instance, when searching for the verb liquefy, concordances
can be ltered according to the user type parameter. Not surprisingly, the verb appears more often in
expert-related texts than in texts addressed to the general public (Figure 4).
Figure 4: Frequency of liquefy in the EEC according to user type.
897Lexicography in gLobaL contexts
With this feature the frequency of terms in dierent domains can also be observed, thus verifying if
a term is more specic to one domain or another. For instance, by searching the lemma photovoltaic
and looking up its frequency according to domain, the results show that it is a term mainly linked to
the domain of Renewable Energy, although it also occurs, but with much lower frequency, in Clima-
tology and Air Quality Management (Figure 5).
Figure 5: Frequency of photovoltaic in the EEC according to environmental subdomain.
3.2 Word Sketch and Sketch Di
The EEC employs both the default sketch grammar for English underlying the word sketches in the
tool in combination with the ESSG. Users can benet from Sketch Engine’s default word sketches
when searching for the collocations that are used more often in specialized discourse in combination
with a certain term. For instance, Figure 6 shows the modiers of methane, the nouns modied by
methane and the verbs that collocate with methane both as object and subject.
Figure 6: Word sketches of methane extracted from the EEC.
Thanks to the ESSG, users can access ready-made semantic word sketches such as those shown in
Figure 7, where search terms may appear related to their hyponyms (i.e. microorganism), the whole
they are part of (i.e. oxygen), their underlying causes (i.e. tsunami), etc.
Figure 7: Semantic word sketches of methane extracted from the EEC.
The word sketch queries can be complemented with the text type lters provided by the tags of the
EEC (or subcorpora based on them). In this sense, users can also observe how concepts can change
their relational behavior across dierent environmental subdomains. For example, Figure 8 shows
how nitrogen is mainly categorized as a type of pollutant in the domain of Air Quality Management
and as a type of nutrient in that of Biology.
898 Proceedings of the XViii eUrALeX internAtionAL congress
Figure 8: Nitrogen generic-specic semantic word sketches in Air Quality Management (left)
and Biology (right) subcorpora.
Additionally, if users access the concordances extracted with the ESSG, they can extract knowl-
edge-rich contexts (i.e. contexts containing domain knowledge potentially useful for conceptual anal-
ysis (Meyer 2001)) like the ones in Table 1.
Table 1: Sample of knowledge-rich contexts extracted from the EEC with the aid of the ESSG.
generic-
specic
A hydrograph is a graph that reects the discharge of a river over a period of time.
The astronomical tide refers to the regular oscillations of the sea or ocean surface[…].
part-whole Sand grains usually consist of quartz but may also be fragments of feldspar, mica, and, […].
Seawater contains sodium chloride and other salts in concentrations three times greater
[…].
location Lagoons commonly form on coastlines that are subsiding, or where sea level is rising.
Most ozone is found in the stratosphere at elevations between 10 and 50 kilometers […].
cause […] the human costs of malaria outweigh the environmental damage caused by the use of
DDT.
Logging may also contribute to deforestation by making it easier for agriculture to […].
function Membrane-assisted BAC is used for the removal of priority pollutants from secondary […].
Liquid-in-glass thermometers are often used for measuring surface air temperature
because […].
Another word-sketch based feature that can be especially exploited with the EEC is Sketch di. It
allows the user to compare either the word sketches of two lemmas, the word sketches on the same
lemma in two subcorpora, or two dierent word forms of the same lemma. Figure 9 shows an exam-
ple of each type. At the left, the modiers of risk (in green) and hazard (in red) in the whole EEC are
contrasted. As it can be observed, these two semantically related terms tend to co-occur with dierent
modiers, although they also share some of them (in white). At the center, there is a sketch di that
shows how water takes dierent verbs as an object in Hydrology (in green) and Water Treatment and
Figure 9: Sample of sketch dis extracted from EEC.
899Lexicography in gLobaL contexts
Supply (in red), as well as a considerable number of shared results. Finally, the sketch di at the right
outlines the verbs that tend to have gas as subject in singular (in green) and in plural (in red) in the
whole EEC.
3.3 Word List
The Word list feature can be used to extract frequency lists with many dierent settings including
n-gram extraction, ltering based on regular expressions or keyword extraction with the aid of a us-
er-chosen reference corpus. This feature can be used in combination with an EEC subcorpus, which
allows the user to generate very specic frequency lists. Some examples of frequency lists that could
be useful to generate from the EEC are: nouns specic to Energy Engineering academic texts using
the British National Corpus as a reference; most common 4-grams in Zoology texts; adjectives con-
taining -friendly in the whole EEC; or the most common verbs in Geology texts (Figure 10).
Figure 10: Frequency list of verbs in Geology texts.
4 Conclusion
In this paper, we have shown how the EEC was built and compiled and how it can be queried and
exploited in Sketch Engine. The EEC’s metadata, the default sketch grammar and the ESSG make
the EEC a useful resource for any user interested in environmental science. As future work, we will
rene, improve and update the ESSG and develop new rules for Spanish. Furthermore, in the short
term, we plan to upload an improved version of the EEC (with more words and some minor codi-
cation issues solved) and a rst version of the Spanish counterpart. In the long run, we will enhance
the EEC with a new annotated version, where dierent semantic tags will be added to improve its
querying potential. These semantic tags will include semantic categories and argument structure.
900 Proceedings of the XViii eUrALeX internAtionAL congress
Sketch Engine’s API also allows for the exploitation of the EEC from external applications. An
example of this is EcoLexiCAT, a terminology-enhanced computer assisted translation (CAT) tool
that provides easy access to domain-specic terminological knowledge in context (León-Araúz &
Reimerink, 2018; León-Araúz, Reimerink & Faber, 2017). EcoLexiCAT integrates dierent features
of the professional translation workow in a stand-alone interface where a source text is interactively
enriched with terminological information (i.e., denitions, translations, images, compound terms,
corpus access, etc.) from EcoLexicon, BabelNet, IATE, and Sketch Engine. In the Sketch Engine
module of EcoLexiCAT’s interface, terms from both the source and target segments can be selected
and direct access is given to concordances, CQL queries and word sketches of the selected terms. For
a more detailed analysis, the output of the queries can be opened in a new tab that sends users to the
website of the Sketch Engine Open Corpora.
References
Faber, P., León-Araúz, P. & Reimerink, A. (2016). EcoLexicon : New Features and Challenges. In GLOBALEX
2016: Lexicographic Resources for Human Language Technology in conjunction with LREC 2016, pp. 73–80.
Jakubíček, M., Kilgarri, A., McCarthy, D. & Rychlý, P. (2010). Fast syntactic searching in very large corpora for
many languages. Proceedings of the PACLIC 24, pp. 741–747.
Kilgarri, A., Baisa, V., Bušta, J., Jakubíček, M., Kovář, V., Michelfeit, J., Rychlý, P. & Suchomel, V. (2014). The
Sketch Engine: ten years on. Lexicography, 1(1), pp. 7–36. http://doi.org/10.1007/s40607-014-0009-9
Kilgarri, A., Kovár, V., Krek, S., Srdanovic, I. & Tiberius, C. (2010). A Quantitative Evaluation of Word Sketch-
es. In A. Dykstra & T. Schoonheim (Eds.), Proceedings of the 14th EURALEX International Congress. pp.
372–379. Leeuwarden/Ljouwert, The Netherlands: Fryske Akademy.
Kosem, I., Gantar, P., Logar, N. & Krek, S. (2014). Automation of Lexicographic Work Using General and Special-
ized Corpora: Two Case Studies. In A. Abel, C. Vettori & N. Ralli (Eds.), Proceedings of the XVI EURALEX
International Congress: The User in Focus, pp. 355–364. Bolzano/Bozen: Institute for Specialised Commu-
nication and Multilingualism.
León-Araúz, P. & Reimerink, A. (2018). Evaluating EcoLexiCAT: a Terminology-Enhanced CAT Tool. In Pro-
ceedings of the 11th International Language Resources and Evaluation Conference (LREC2018). Miyazaki:
ELRA.
León-Araúz, P., Reimerink, A. & Faber, P. (2017). EcoLexiCAT: a Terminology-enhanced Translation Tool for
Texts on the Environment. In I. Kosem, C. Tiberius, M. Jakubíček, J. Kallas, S. Krek & V. Baisa (Eds.), Elec-
tronic lexicography in the 21st century. Proceedings of eLex 2017 conference, pp. 321–341. Leiden: Lexical
Computing.
León-Araúz, P. & San Martín, A. (2018). The EcoLexicon Semantic Sketch Grammar: from Knowledge Patterns to
Word Sketches. In I. Kerneman & S. Krek (Eds.), Proceedings of the LREC 2018 Workshop “Globalex 2018
– Lexicography & WordNets”, pp. 94–99. Miyazaki: Globalex.
León-Araúz, P., San Martín, A. & Faber, P. (2016). Pattern-based Word Sketches for the Extraction of Semantic Re-
lations. In Proceedings of the 5th International Workshop on Computational Terminology, pp. 73–82. Osaka.
Meyer, I. (2001). Extracting knowledge-rich contexts for terminography. In D. Bourigault, C. Jacquemin & M.-C.
L’Homme (Eds.), Recent advances in computational terminology, pp. 279–302. Amsterdam/Philadelphia:
John Benjamins.
San Martín, A., Cabezas-García, M., Buendía Castro, M., Sánchez Cárdenas, B., León-Araúz, P. & Faber, P. (2017).
Recent Advances in EcoLexicon. Dictionaries: Journal of the Dictionary Society of North America, 38(1),
pp. 96–115.
Sinclair, J. (1991). Corpus, Concordance, Collocation. Oxford: Oxford University Press.
Teubert, W. (2005). My version of corpus linguistics. International Journal of Corpus Linguistics, 10(1), 1–13.
http://doi.org/10.1075/ijcl.10.1.01teu
901Lexicography in gLobaL contexts
Acknowledgements
This research was carried out as part of projects FF2014-52740-P, Cognitive and Neurological Bas-
es for Terminology-enhanced Translation (CONTENT), and FFI2017-89127-P, Translation-oriented
Terminology Tools for Environmental Texts (TOTEM), funded by the Spanish Ministry of Economy
and Competitiveness.
