Navigation – Plan du site
Hétérogénéité de et dans les petits corpus

Discourse on climate and energy justice: a comparative study of Do It Yourself and Bootstrapped corpora

Camille Biros, Caroline Rossi et Inesa Sahakyan

Résumés

Cet article décrit et analyse les différentes étapes de constitution d’un corpus représentatif des questions de justice climatique et énergétique. Le corpus contient cinq millions de mots en tout et rassemble des rapports, des lettres d’information et pages web traitant des solutions équitables à faible empreinte carbone pour limiter le changement climatique. Il est divisé en six sous-corpus selon les types de communautés de discours et de méthodes de constitution. Nous commençons par la présentation du petit corpus fait maison que nous utilisons comme point de départ. Trois communautés de discours ont été sélectionnées afin d’observer d’éventuelles variations dans leur traitement de ces questions : Organisations Non Gouvernementales, institutions onusiennes et organisations du secteur de l’énergie renouvelable. Les sources ont été sélectionnées en fonction des auteurs, dates et mots clés présents dans les titres. Grâce aux logiciels de concordance AntConc et WMatrix, nous avons testé la comparabilité de ces corpus du point de vue de leur contenu thématique, de leur terminologie et de la classification de leurs unités lexicales. Nos premiers résultats nous permettent de confirmer l’existence de variations entre communautés de discours. Le caractère chronophage de notre démarche de constitution d’un corpus « maison », ainsi que le déséquilibre entre le nombre de mots obtenus pour chaque sous-corpus nous conduisent à utiliser BootCat afin de constituer un corpus plus fourni. L’outil utilise des mots clés comme « semences » pour la récupération et le téléchargement automatiques de pages web. Nous pouvons ainsi comparer une méthodologie traditionnelle de constitution de corpus à une méthodologie qui utilise le web en tant que corpus. Nos résultats BootCat sont confrontés à ceux du corpus maison pour voir s’ils révèlent aussi bien les spécificités des sous-corpus. Cette démarche aboutit à des conclusions sur les possibles utilisations de corpus relativement petits, et d’en souligner la pertinence pour l’étude de discours spécialisés.

Haut de page

Texte intégral

1The concept of environmental justice was coined in the United-States to refer to racial minorities being impacted unfairly by environmental degradation (Bullard 1990). Recently, two hyponyms of this term created by analogy have appeared in environmental debates: climate justice and energy justice. Like the term they are derived from, they are used to reflect on inequalities between different populations when facing environmental problems. They signify a will to build an energy transition to a low-carbon future that is inclusive of minorities and fair in its outcome. However, because these terms have been coined relatively recently, because they refer to complex ethical questions, and because they are being taken up across disciplines, a certain amount of controversy exists surrounding their use and definitions. It seems essential to clarify the policy objectives these terms suggest (Heffron & McCauley 2017: 7). Our aim is to shed light on how climate and energy justice are elaborated by different discourse communities. Our approach is variationist as we contrast the treatment of these questions in three sub-corpora representing Non-Governmental Organisations, United-Nations and the Renewable Energy Sector (RES) i.e. companies and representative institutions. We seek to characterise the specificities of the treatment of this issue by each of these discourse communities, according to their different objectives. This can only be done through the constitution of corpora which are specific to each discourse community. We began by constituting a DIY corpus, gathering reports and newsletters identified through a web-search by organisations representative of each of these communities. Further details on the methodology that guided our endeavour are given in what follows. The corpora obtained are relatively small corpora, based on three distinct discourse communities. The limitation in size is linked with the time-consuming nature of our methodology and the small length of the documents published by two of our discourse communities. We decided to extend the corpus by adding the content of automatically-retrieved, relevant web-pages. We used BootCat to create three sub-corpora from the web, and we hypothesised that by using distinctive keywords as seeds, they would be representative of our three discourse communities. Beyond the characterisation of the specialised lexis of each of these discourses, this work enables us to compare the results between a DIY corpus and a corpus bootstrapped from the Web. The overarching question is whether the characterisation of linguistic specificities in each discourse community under review is possible through a web-based automatically generated corpus or whether small DIY corpora offer more reliable tools for our objectives.

1. The DIY Corpus

1.1. Methodology

2The first step in the constitution of our corpus was to identify different discourse communities concerned by the concepts of “climate justice” and “energy justice”. We base our understanding of a discourse community on Swales’ text-based concept, broadly defined as communities with “a broadly agreed set of common public goals […] mechanisms of intercommunication among its members” and resorting to “one or more genres in the communicative furtherance of its aims” as well as “some specific lexis” (Swales 1990: 24-27). The concepts of “climate justice” and “energy justice” come from the activist community. The first Climate Justice Summit took place in The Hague in 2000. It coincided with the Conference of the Parties 6 (COP6). The Energy Justice Network was created in 1999 to bring together NGOs with a particular focus on how the energy transition could be acquired with more concern for distribution of burdens and benefits (Heffron & McCauley 2017: 2). Because of the activist origin of the terms, the first discourse community we focused upon was that of Non-Governmental Organisations. We consider their discourse through reports in which they deal with this issue.

3The second discourse community that seemed to us particularly relevant to consider these issues was that of UN institutions, because of their role in negotiating a binding international agreement to fight climate change. The fact that the term “climate justice” appears in the Paris Agreement, tended to confirm the interest of this discourse community for the question. UN organisations also seemed a good option to show contrasting views with NGOs. During the COP 15 conference in particular, the high level of activist mobilisation and the weak outcome of official negotiations led to a clash between the two with lasting effects in terms of opposition in communication strategies and objectives (Dahan & Aykut 2015: Chapter 7). We consider this discourse community through reports published by different United-Nation organisations on the theme of climate change and its impact on human welfare.

4The third discourse community we identified as having an essential role to play to define and work on these concepts is that of the Renewable Energy Sector (RES). Companies, and institutions representing them, may use climate and energy justice as arguments to promote their products by contrasting them with conventional energy sources that produce a large amount of carbon dioxide. They also may show a willingness to differentiate themselves from big energy companies who have little regard for social impacts by showing an interest in developing fair sources of energy.

5To constitute each sub-corpus representative of each discourse community, we began by using our prior knowledge of the issue, based on “horizontal reading” (Rühlemann & Aijmer 2014) of sources that we did not include in the corpus but kept as secondary sources, to focus on the discourse of organisations that were most likely to communicate on these issues. A simple Internet query with the keywords and the organisations thus identified enabled us to gather a first series of documents. However, our aim was to gather not only documents that used these terms but also those that dealt with the issue without using the terms. Indeed, as these terms come from the activist community, it seemed likely that the UN and RES would be less willing to use them. To extend the list of keywords for our query we used a list of definitions extracted from the documents gathered in the first stage, but this time we read them “vertically” (ibid.) using the concordance Antconc (Anthony 2014). From this sub-corpus of definitions, we were able to find new keywords to extend our corpus. We completed this work with defining extracts from academic publications (Rhaman 2016, Heffron et. al. 2015, Nicholson & Chong 2011, Heffron & McCauley 2017). In each definition, a semantic analysis enabled us to single out associated terms. Here are two examples of defining extracts and their use.

(1) “to deepen the understanding of climate justice which, at the nexus of climate change, development and human rights, seeks to ensure that climate action is fair and people-centred” (Mary Robinson Foundation Climate Justice Annual Report 2014: 3).

(2) “a range of groups are even now beginning to strategically utilize human right institutions, practices and discourses under the umbrella of “climate justice”, in debates about climate change” (Nicholson & Chong 2011: 122).

6In definition (1), the words “fair and people-centered” associated to “climate action” are presented as close synonyms of “climate justice”. In both definitions (1) and (2), “human rights” are associated to “climate justice”: in definition (1) there is a centre/periphery relation while in definition (2) human rights are associated to climate justice as a means to an end. This semantic analysis of definitions enabled us to identify the following terms as linked to climate and energy justice: human rights, fairness, equality, equity, accessibility, safety, distribution, people-centered, representation. The corpus we gathered with this method presents the following characteristics.

Table 1. DIY Corpus created through vertical and horizontal methodologies

Table 1. DIY Corpus created through vertical and horizontal methodologies

7To constitute each of these three corpora, the documents we gathered were mostly reports. Many of these were published around the Conference of the Parties in Paris in 2015. The discrepancy in sizes is explained by the fact that the UN publishes long reports whereas the documents in the RES and NGO corpora are much shorter. It was more difficult to find company publications on the topic. Although the RES corpus is quite small in terms of number of words, it seems representative as it includes 43 documents published by nine different organisations. We were attentive to maintain a certain generic uniformity and included only official material published by an identified organisation and dated. The problem of the discrepancy in sizes of the three corpora is somewhat toned down by the fact that most of the tools we adopt for our analysis use statistic tests (Log Likelihood Feature on Antconc, UCREL Semantic Analysis System on WMatrix) rather than raw counts. Besides, we added information on words per million to raw frequencies, so that comparison is possible across corpora of differing sizes.

1.2. Keyword analysis

  • 1 Log-Likelihood ratios are used within statistical tests to assess differences. Here the comparison (...)

8Our object is to better understand how each discourse community conceptualises issues of climate and energy justice. With this in mind, we built the corpus using thematic and institutional criteria to identify meaningful documents. Although a more refined qualitative approach to the texts would be necessary to fulfil our objectives, in a first stage, we wanted to question the pertinence of our corpus. Keyword analysis (Baker 2004) offers great means for this. Using automatic software, keywords indeed make it possible to characterise the corpus broadly and check its thematic homogeneity. To identify keywords in the three sub-corpora we loaded a sample of the Corpus of Contemporary American English (Davies 2010) as a reference corpus in the software Antconc and used the Log-Likelihood (LL) feature1. We compared the fifty first keywords in each corpora to determine similarities and differences. The twenty most significant keywords for each are presented in Table 2 below.

Table 2. Keywords in each sub-corpus

Table 2. Keywords in each sub-corpus

9The colour code enables a quick comparison between the content of the list of the first fifty keywords for each corpus as they appear differently according to whether you find them in only one corpus, in all three, or in a combination of two. They appear in blue if present in the three corpora, in green if present in the NGO and RES corpora, in red if present in the NGO and UN corpora, in yellow if present in UN and RES corpora and in black if specific to one corpus. Five keywords are common to the three corpora: climate, energy, emissions, global and development, i.e. which one can easily identify as pertaining to the field of climate change.

10If you compare the colours in each list, the first striking feature is the proximity between the NGO and RES corpora as compared to the UN corpus. Especially if you take the first twenty keywords, you can see that a majority of keywords from the UN appear in black, while only six appear in black in the NGO corpus and five in the RES corpus. Furthermore, nine appear in green in the NGO corpus and twelve in the RES corpus. Those in green tend to be lexical items linked to the energy sector and the production of electricity.

11To delve deeper into the lexical fields that are present in each corpus, it may be useful to turn to semantic differences. We used the semantic tagging feature available in the WMatrix (Rayson 2008). Based on the UCREL Semantic Analysis System2 (USAS), this software enables researchers to identify key semantic domains in a corpus by comparing it to a corpus of reference3.

1.3. Semantic Categories and Lexical Units

1.3.1. Overview

12The results obtained concerning the first twenty-five semantic categories in each corpus are presented in Table 3 below. As in the previous part, the colour code enables a comparison of the content of the three corpora.

Table 3. Semantic categories in each sub-corpus

Table 3. Semantic categories in each sub-corpus

13One can start by stating that one fifth of these lists is common to all three corpora, i.e. the following five categories: “Weather”, “Change”, “Place”, “Green issues” and “Science and Technology”. This tends to confirm the focus of our corpora on climate change, which is obviously related to the weather, describes a type of change, considers effects of change in different places, and can be labelled as a green issue. Science and technology are central to discourse on climate change both to better its understanding and to find solutions to fight it. A phenomenon which was already apparent in the comparison of keywords appears with much more clarity in the comparison of semantic categories: the proximity of the NGO corpus to the other two and the strong variation between the UN and RES corpora. The NGO corpus shares nearly all of its semantic categories with at least one other corpus, with only four semantic categories that are specific to this corpus. It could be seen as a combination of categories essential to the UN corpus and of categories essential to the RES corpus.

14The high number of lexical items linked to the energy sector in the NGO and RES corpus is confirmed when looking at the categories in green. Among the first are “Interested / excited / energetic” and “electricity and electrical equipment”. Of course the polysemy of the lexical unit energy clearly appears in the first category where it is associated with lexical units like interest, active, enthusiasm and keen. However, one can readily check by looking at the list of lexical units in this category that its importance is mainly due to the presence of an important lexical field of the energy sector. In our corpora, the lexical unit energy is mainly used as a reference to production of electricity rather than as a state of mind. A focus on economic matters also seems prevalent in these two corpora with the categories “Money and Pay” and “Money Generally”. The categories “Giving” and “Using” suggest a pragmatic approach. Overall, the impression that these two corpora offer a concrete focus on the energy transition dominates.

15Conversely, the categories shared by the NGO and UN corpora suggest a descriptive approach of the effects of climate change with “Geographical Terms” and “Geographical Names”, “Temperature”, “Quantities” and “Numbers”. The unmatched category appearing right at the beginning reveals a high number of technical terms, neologisms and references to institutions and organisations. With respect to categories specific to the UN corpus, the focus on description of problems is confirmed with the categories “Danger” and “Weak”. The abstract character of this corpus appears with the categories “Cause&Effect / Connection” and “Investigate, Examine, Test, Search”. The category “Confident” is also significant in the UN corpus. Indeed, our prior knowledge of the corpus enables us to assert that the expression “degrees of confidence” is used to qualify the certainty and uncertainty of statements on the possible effects of climate change, namely in the International Panel on Climate Change publications. A quick search in Antconc retrieves 54 concordances and confirms that the term is present throughout all four IPCC5 reports.

16To compare the content of the three corpora using the semantic tags, one can also go into more detail and look at the lexical units that compose the semantic categories in each corpus to see if they are related. In the next sections, we will compare the lexical units found in the five categories shared by each sub-corpus.

1.3.2. Lexical Units in categories shared by the three sub-corpora

17The first semantic category shared by the three sub-corpora is that of “Weather”. The list of the thirty most used lexical units in this category appears in Table 4, which gives raw counts of word frequency (w.f.) with words per million (wpm) in brackets.

Table 4. The most used lexical items in the category "Weather"

Table 4. The most used lexical items in the category "Weather"

18What is striking with this list is the uniformity between the different corpora when treating this theme. The fifteen first lexical units of each appear in the three corpora, with only two exceptions. From sixteen onwards, there is more diversity. The table also confirms the links already observed between the NGO corpus and the two others and the greater discrepancy between the UN and RES corpora. The bigger focus on energy markets in the NGO and RES corpora is confirmed when you see that the two lexical units specific to these two are sunny, probably as a reference to the use of solar energy, and wind market. There are even more lexical units specific to the energy market in the RES corpus: sunnier, wind companies, wind farm, wind mills and wind sector. The lexical units that are specific to the UN corpus suggest more focus on the negative outcomes of climate change as many are evocative of extreme events: monsoon, hurricane(s), snowfall and avalanches. In the UN corpus we significantly find technical terms like climate factor and weather conditions.

19Another category that shows a strong degree of uniformity between the three corpora is that of “Green Issues”, presented in table 5.

Table 5. The most used lexical items in the "Green Issues" category

Table 5. The most used lexical items in the "Green Issues" category

20In this list, seventeen lexical units are present in the three corpora. The link between the NGO corpus and the two others is confirmed. Surprisingly, the semantically-linked lexical units desertification and soil erosion do not appear in the NGO corpus as they do in the two others.

21There is also a strong degree of uniformity in the lexical units included in the category “Places”, presented in Table 6.

Table 6. The most used lexical items in the "Places" category

Table 6. The most used lexical items in the "Places" category

22Here, sixteen lexical units out of thirty are common to the three corpora. It confirms the greater proximity of the NGO corpus to the two others. municipal and city, common to the NGO and RES corpora, could be linked to an interest for the local scale where decisions about energy are or could be taken. Base and site could refer to sites of energy production. The lexical unit indigenous, common to the NGO and UN corpora, confirms an interest for negative impacts of climate change as indigenous people are often quoted as victims, in particular in the context of deforestation.

23Although the lexical field of “Change” is central in the three corpora, there is more diversity in ways of expressing it as one can observe in the list of lexical units presented in Table 7.

Table 7. The most used lexical items in the "Change" category

Table 7. The most used lexical items in the "Change" category

24The link between the NGO corpus and the two others is confirmed with six lexical units common to the NGO and RES corpora and four common to the NGO and UN corpora.

25The category “Science and Technology” also presents contrasting lists of lexical units with only ten common to the three corpora as can be seen in Table 8.

Table 8. The most used lexical items in the "Science and Technology" category

Table 8. The most used lexical items in the "Science and Technology" category
  • 4 Pollutant associated to conventional energy production.

26The proximity between the NGO and RES corpora is confirmed. The UN appears more specific here with the strong degree of abstraction confirmed by the presence of twelve lexical units referring to disciplinary fields. It suggests a stronger focus on science whereas the two others are dominated by references to technology (technology transfer, tech, technically), in particular technology linked to the production of energy (selenium4, nuclear).

1.4. Lexical Units on Justice

27Although our starting point for the constitution of the corpus were the terms “climate justice” and “energy justice”, the semantic tagging feature in WMatrix does not identify this issue as key in our corpora. The thematic prevalence of climate change is clear but our interest for equity issues in relation to climate change, which led to our selection of sources, is not reflected in the main semantic categories. In the UN corpus, we find no significant semantic category linked to ethics or law. In the NGO corpus we find two. The category Lawful (G2.1), appears 69th in terms of Log Likelihood, and the category “Ethical” appears 119th. In the RES corpus, the category “Ethical” appears 110th. The WMatrix software does not really help us to compare the importance of ethical issues in our three corpora. This is the reason why we decided to use the Antconc word count feature to find out the number of occurrences of our lexical units of interest in the corpus. Starting from the list of keywords identified through the semantic analysis of definitions in our corpus (Section 1.1), we established frequency counts with Antconc: Table 9 gives raw counts, with words per million in brackets (wpm).

Table 9. Frequency counts with Antconc

Table 9. Frequency counts with Antconc

28The first striking feature is the very small word frequencies of our three main terms of interest in the UN and RES sub-corpora. The use of climate justice is high in the NGO corpus, suggesting a term that is specific to this activist discourse community. The second important feature is the high use of the alternative terms equity and human rights in the UN corpus, terms which are also present in the NGO corpus. In the RES corpus, these terms only remain marginal but one may notice that equity and equality are the most used. Of course, this simple word count does not enable us to conclude that in all cases where equity is being used, issues linked to justice are being discussed. It is possible that the term is also being used in its economic meaning in our corpora. However, we were able to check that a significant number of occurrences referred to its ethical meaning. There is a high number of occurrences of distribution in the three corpora. Here too, the polysemy of the term tones down the interest of this simple word count. However, issues of distribution of burdens and benefits are central to climate and energy justice and the word count suggests the importance of this term in our corpora, whose collocations could be interesting to study in future research.

1.5. Significant Compounds and Clusters

29When analysing the differences between semantic categories in the three different corpora, we were particularly interested by the fact that in the UN corpus, the category “Weak” and “Danger” featured among the first twenty-five. This tends to suggest a focus on the description of the negative consequences of climate change. The category “Weak” seems particularly significant in dealing with the issue of justice. Most of the lexical units included refer to victims of climate change. The lexical units that are most important in the “Weak” category in the UN corpus according to WMatrix are the following: vulnerability (2142), vulnerable (480), vulnerabilities (391), weak (52), susceptible (40) and fragile (10). Compared to the two other corpora, where these lexical units do not appear as central, it seems that the UN tends to focus on describing victims of climate change and explaining why they are victims. Vulnerable and its morphological derivatives are most used for this purpose. To further understand the differentiated use of these lexical items in our three corpora, we used the N-Gram feature in Antconc to identify significant clusters and compare them in each corpus. Our hypothesis was that the two other corpora, with their pragmatic approach, would present more of an interest for identifying people responsible for climate change rather than identifying its victims. To consider this issue, we identified the adjective responsible and its derivatives in each sub-corpora following the same method. In the following tables, we consider the ten most significant N-grams in each corpus, with a minimum frequency of three.

Table 10. N-Grams starting with Vulnerab*

Table 10. N-Grams starting with Vulnerab*

30This table confirms that the lexeme vulnerable is much more represented in the UN corpus, with many clusters starting with the adjective or noun. In the NGO corpus, we have some occurrences also but in the RES corpus hardly any. A specificity of the items in the UN corpus is that several refer to tools to measure vulnerability: vulnerability assessment(s) and vulnerability mapping.

Table 11. N-Grams starting with Responsib*

Table 11. N-Grams starting with Responsib*

31This table tends to show that the issue of responsibility is central in the NGO corpus compared to the two others. The idea is confirmed if we adopt a differentiated approach to the use of the lexical unit responsible, distinguishing its meaning as “taking care of” as opposed to “causing”. Most of the occurrences of responsible in the UN table signify that someone is responsible as in “taking care of”, whereas most of the occurrences in the NGO corpus are used to refer to causes of climate change. In the NGO corpus, we find a term to refer to a tool to measure responsibility: responsibility and capability index. This analysis tends to confirm that there is more of a focus on measuring vulnerability in the UN corpus and more of a focus on measuring responsibility in the NGO corpus. These issues do not appear as central in the RES corpus.

32Another interesting distinction we noted between our corpora thanks to the WMatrix categories is the fact that the NGO and RES corpora are much more focused on energy and its production. The second type of cluster that we decided to focus on aims to further our understanding of these differences. We consider the first ten occurrences of two-word units having the structure [energy + NOUN] and the first ten occurrences of two-word units having the structure [ADJECTIVE + energy].

Table 12. [ENERGY+NOUN] in the three sub-corpora

Table 12. [ENERGY+NOUN] in the three sub-corpora

33Although the figures for this type of cluster are higher in the NGO and RES corpora, we find similar units in each. They show a willingness to calculate what is being produced and consumed, and to go towards an integrated and efficient system.

Table 13. [ADJECTIVE+ENERGY] in the three sub-corpora

Table 13. [ADJECTIVE+ENERGY] in the three sub-corpora

34Table 13 confirms the higher number of occurrences of two-word units related to energy in the NGO and RES corpora. It shows renewable energy is the term that is most used to refer to non-conventional sources of energy. The term clean energy is used by NGO and RES corpora. The term good energy is used only in the RES corpus, which may be linked to the fact that one of the companies we selected documents from for this corpus is named Good Energy. Specific types of energies are mentioned in the list for the three corpora but there is no distinct pattern to explain why some energy sources are being mentioned rather than others in each corpus.

35On the whole, the first results concerning distinctive characteristics of our three sub-corpora tended to confirm their usefulness to offer a contrasting view of climate and energy justice, according to our three discourse communities. However, the time-consuming character of the corpus constitution and the discrepancy in sizes of our three sub-corpora led us to question whether a bootstrapping tool could be useful to extend our corpus. Of course, the generic specificity would not be entirely maintained with this method. However, some questions concerning terms, their use and their definitions could be considered more efficiently in an extended corpus. Our interest for using this tool also stemmed from a willingness to test its efficiency and determine its possible limits.

2. The BootCat Corpus

36BootCat stands for “Bootstrapping Corpora And Terms” from the web (Baroni & Bernardini 2004). It is a freely available toolkit which was developed as a simple web-mining device to help translation students, translators and terminologists build specialised corpora. BootCat was thus originally aimed at “users who need relatively large and varied corpora (typically of about 1-2 million words), and who are likely to search the corpus repeatedly for both form- and content-oriented information within a single extended task.” (Bernardini 2006). The resulting corpora have been called “DIY corpora” (Zanettin 2002) or “disposable corpora” (Varantola 2003). In this paper, data collection was not conducted primarily with a view to helping translators, and we sought to gather representative DIY corpora that could then be used for various purposes. Our BootCat corpora are thus somewhat different from our DIY corpora. They were designed as possible complements to these relatively small corpora. Bearing in mind the general advantages and limitations of using webpages as a corpus (i.e. “The web is a dirty corpus, but expected usage is much more frequent than what might considered as noise” Kilgarriff and Grefenstette 2003: 342), our overarching aim was to assess their usefulness for the analysis of specialised discourse. In this section we describe how we went about gathering the data before assessing the relevance of BootCat corpora for a broader set of aims, which could be defined as corpus-based discourse analysis.

2.1. Methods to Build Three BootCat Corpora

  • 5 Seeds are starting points for automatic retrieval. Web-crawlers typically use URLs as seeds, and Bo (...)
  • 6 The term refers to an ordered sequence of a number of items: in BootCat, the user chooses “tuple le (...)

37BootCat starts from seed terms5 that are automatically combined into tuples6 to produce a series of web queries. Although the tuples cannot be controlled, there are at least two ways of discarding irrelevant results. The first one consists in carefully choosing distinctive keywords to be used as seeds. The second one has to do with URLs: BootCat allows users to establish restrictions as well as to manually sort automatically retrieved URLs before downloading the corpus.

2.1.1. Keywords

  • 7 While BootCat outputs just one text file, the SketchEngine creates a table associating each retriev (...)

38We chose to use BootCat within the SketchEngine (Kilgariff et al. 2014) because of recent limitations imposed on the number of free queries with the Bing search engine, and because it allowed us to see more clearly how many words had been retrieved from each URL7. This brought about new limitations: the SketchEngine has fixed parameters and can only take 20 seeds to create a corpus that will also be limited in size (1 million words). The parameters, however, were ideal since we wanted the size of our bootstrapped corpora to remain comparable to that of our three DIY corpora.

39We defined two initial seeds that set the main themes (climate change and energy) and selected the remaining 18 from our comparative table of keywords. First we kept only those keywords that were distinctive of each of our DIY sub-corpus. Then we sorted them according to frequency and kept only the first 18 for each sub-corpus. We used frequency rather than keyness in order to make sure our selected keywords were present in a substantial number of documents (based on Baker 2004: 350-354). The resulting list is presented in Table 14 below.

2.1.2. URLs

40In selecting URLs, we did not seek to achieve discourse homogeneity of the kind that was present in our DIY corpora. Manually checking that each and every retrieved URL corresponded to speakers belonging to the relevant discourse community would have taken us far beyond regular BootCat procedures, which were precisely what we wanted to test. In order to get rid of data that would not be relevant to any of our subcorpora, we excluded the whole of Wikipedia.org, and took out manually the least relevant sources such as the press and blogs. Table 14 shows the final number of URLs used to build our three subcorpora, and observed differences are worth noticing: longer documents were found for the first two subcorpora, and a quick look at URLs suggests that those documents are indeed longer reports of the kind that were gathered in our DIY corpora.

Table 14. The BootCat Corpora

Table 14. The BootCat Corpora

2.2. Comparative Analysis of Results.

2.2.1. Semantic Categories.

41As can be seen in Table 15 below, when it comes to semantic categories, the BootCat corpus reveals a far greater extent of similarity across the three sub corpora. New semantic categories common to all the three sub-corpora emerge in the BootCat data, such as “Cause and Effect Connection”, “Mental object”, “Wanted”, “Giving”. Semantic categories such as “Numbers”, “Substances and Materials”, “Unmatched” (in italics), which were previously common only to NGO and UN corpora, are identified as common to all the three BootCat sub-corpora thus suggesting greater uniformity across our corpus.

42Conversely, the BootCat results are much scarcer when it comes to semantic categories common to NGO and UN sub-corpora, with only three results as compared to nine in the DIY corpus. Only one semantic category, highlighted in bold, is identified in both corpora. This tendency is further confirmed while analysing the semantic categories common to NGO and RES sub-corpora, nearly twice as many results emerge in the DIY corpus. Finally, while the DIY corpus revealed no results common only to UN and RES corpora, “Places” is identified as common to both.

Table 15. Comparative Analysis of DIY and BootCat Results for the Semantic Categories

Table 15. Comparative Analysis of DIY and BootCat Results for the Semantic Categories

43Notwithstanding the varying number of identified semantic categories common to different sub-corpora, it can be noted that our DIY and BootCat corpora are coherent in terms of results inasmuch as the semantic categories identified in one were more often than not found in the other corpus too.

2.2.2. Lexical Units in categories shared by the three sub-corpora

44To go into a more detailed analysis of results, we will consider the lexical units included in semantic categories common to the three sub-corpora. In a comparative purpose, we only present those which were also present in the three DIY corpora that is to say “Weather”, “Change” and “Science and Technology in general”.

45While examining the BootCat results of the most used lexical items in the “Weather” semantic category, two features seem to be noteworthy. On the one hand, there is a striking coherence with 60 per cent of the lexical units (18 out of 30) being shared by all the three sub-corpora, while on the other, very few lexical units are common in combinations of two. Thus only two are shared by the B-NGO and B-UN corpora, hurricane and hurricanes, which being the singular and plural forms of the same noun further reduce the specificity which could be attributed to this combination of sub-corpora. The same is true for the UN and RES, as well as the NGO and RES combinations, as they merely share the lexical units of inundation and cloud, clouds, weather conditions respectively. As a result, it can be concluded that while there is a strong degree of similarity across all three sub-corpora, each sub-corpus remains distinct enough. As for the features characteristic of the way each discourse community treats the issue of the “Weather”, with lexical units such as haze, fog, humid, draught, choppy, the B-NGO sub-corpus seems to place the focus on the quality of air, and possibly on the negative impacts of atmospheric pollution. The major concern of the UN discourse, on the other hand, seems to be the impact of weather on local populations and agriculture with lexical units like floodplain(s), flooded, hailstorm, heatwaves, avalanche, inundations, chinook and rainy specific to its corpus. Finally, the lexical units exclusively present in the RES list of the 30 most used units point to concerns with the development of renewable sources of energy (wind and solar) with monsoon(s), sunny, wind market, wind output, windy, tornado, rains.

46Comparing the above BootCat results with those of the DIY corpus (discussed in section 1), we could state that the thematic coherence of discourses dealing with the issue of climate change (as represented by the semantic tag “Weather”) is largely confirmed with 56% (17 lexical units out of 30) and 60% of lexical units being shared by all three sub-corpora both in the DIY and the BootCat respectively. The links previously observed between the NGO corpus and the two others are further reinforced, whereas the discrepancy between the UN and RES corpora is scaled down by the presence of the lexical unit inundation in both BootCat sub-corpora. More interestingly, while the term monsoon appears exclusively in the DIY UN corpus, it does so in the BootCat RES corpus. Therefore, the discrepancy between the two sub-corpora is to be questioned through further analyses. Another feature specific to the NGO and RES corpora, namely the focus on the energy market, and in particular on the wind and solar energy, is largely confirmed by the BootCat results.

Table 16. Comparative analysis of discourse community specificities (Weather)

Table 16. Comparative analysis of discourse community specificities (Weather)

47A comparative analysis of the specificities identified by the DIY and BootCat are not only incongruent, they further mitigate the specificities of each discourse by establishing new links within combinations of sub-corpora, and most importantly by drawing a link between the UN and RES sub-corpora with the lexical unit monsoon. A new unit ‘humid’ is identified as common to all the three.

48A slightly lower degree of uniformity (13 out 30, i.e. 43%) is established in the ‘Change’ semantic category, which confirms the DIY results (12 out of 30). The strong link within the combination of NGO and RES sub-corpora is reinforced with 8 lexical units being common to both, in line with the DIY results (7/30 common lexical units). However, in this semantic category the link between the NGO and UN sub-corpora is weakened with only 3 shared lexical units in the BootCat and 4 in the DIY corpus. Finally, the emerging link in the BootCat results between the UN and RES sub-corpora is confirmed with 4 lexical units being shared by both, while there were none common to this combination of sub-corpora in the DIY results.

Table 17. Comparative analysis of discourse community specificities (Change)

Table 17. Comparative analysis of discourse community specificities (Change)

49Unlike the above discussed two categories, the comparative analyses in the “Science and Technology” semantic category point to some incongruence between DIY and BootCat results, with merely one third (10 out of 30) of co-occurrences across the three sub-corpora in the former and almost half (14/30) of lexical items common to the three sub-corpora in the latter. As for the possible links between sub-corpora, the DIY results established no combinations within this semantic category, while BootCat results confirmed the links already established in other categories: namely, the strong links of NGO sub-corpus with the other two (4/30 lexical units common in both combinations, NGO-UN and NGO-RES), and the tenuous link between the UN and RES corpora (1/30).

Table 18. Comparative analysis of discourse community specificities (Science & Technology in general)

Table 18. Comparative analysis of discourse community specificities (Science & Technology in general)

50As Table 18 shows, unlike the above two categories, there is greater coherence (demonstrated in bold type) between DIY and BootCat results in the semantic category of “Science and Technology in General”. Concerning the specificities of each discourse community, it can be noted that there are more lexical units with negative connotation in the NGO sub-corpus which suggests a particular concern with noxious effects of chemicals and nuclear energy in general: neodymium, tellurium, radioactive. In contrast, the most used lexical units in the UN discourse seem to point to advances of the science and technology, and are positively connoted. Finally, the lexical units in the RES sub-corpus point to technical and engineering solutions. The most used lexical units in this “Science and Technology” semantic category are positive, as is the case with the UN corpus. Overall, while the latter two shape their discourse mostly pointing to possible solutions, NGOs point to problems that could possibly stem from scientific and technological advances.

2.2.3. Lexical Units on Justice

51To consider the place of issues of justice in our corpora, we began by referring to semantic tags with WMatrix. As in the DIY corpora, we did not find semantic tags related to justice and ethics among the most represented in our corpora. In the B-NGO corpus, “Ethical” appears only 179th in terms of Log-Likelihood, in the B-UN corpus “General Ethics” appears 206th while in the B-RES corpus no related semantic tag appears. As was the case with the DIY corpus, the Antconc word count feature was used in a second stage to find out the number of occurrences of our lexical units of interest in the Bootstrapped corpus. The results are presented in Table 19. The striking absence of our three main terms of interest in the UN and RES sub-corpora in the DIY results are not only confirmed in the BootCat, but amplified with only a few occurrences in NGO and UN sub-corpora. However, further comparative analyses of results reveal a significant incongruence as there is not the same distribution of terms across sub-corpora.

Table 19. Frequency counts of terms of interest in the Bootstrapped corpus with Antconc

Table 19. Frequency counts of terms of interest in the Bootstrapped corpus with Antconc

52Opposite tendencies are observed when it comes to the lexical unit distribution, which had the highest frequency of occurrence in the DIY UN sub-corpus, and the lowest in the RES sub-corpus. In the BootCat data, the occurrence of the lexical unit is the most frequent in the RES corpus and the least frequent in the UN sub-corpus. The same is true for the term equity. Furthermore, while climate justice has a high frequency of occurrence in the DIY NGO corpus, it does not appear at all in the BootCat results. All these differences seem to suggest that the DIY corpus allows finer distinctions, which, however, requires further research. In order to look for such differences, the third point we are going to investigate is whether significant clusters appear distinctively in our three BootCat corpora.

2.2.4. Significant Compounds and Clusters

53We will start by analysing the word clusters with vulnerable.

Table 20. Word clusters with VULNERAB*

Table 20. Word clusters with VULNERAB*

54As Table 20 shows, three word clusters are identified as common to all three sub-corpora. However, the BootCat results do not allow for links to be established between the NGO and UN sub-corpora, as was the case in the DIY corpus. In fact, no combinations of corpora are distinguished.

55Next, we come to the term responsible and its derivatives. The first striking feature of the BootCat corpus is the virtual absence of clusters with the lexeme responsible in the NGO sub-corpus. This contrasts with the DIY results, which revealed a great number of clusters in this semantic category. Furthermore, it is interesting to note that the only cluster that appears in the NGO sub-corpus is shared by the RES sub-corpus.

Table 21. Word clusters with RESPONSIB*

Table 21. Word clusters with RESPONSIB*

56The results in the RES sub-corpus are more centred on the economy, and seem to suggest a national, rather than international perspective on the issues of climate and energy justice (implementing national strategies). This confirms the DIY results, where the RES corpus was equally turned to the energy market.

57Finally, we will examine the results for the “energy” cluster.

58As in the case of the DIY corpus, we consider the first ten occurrences of two-word units having the structure [energy + NOUN] and the first ten occurrences of two-word units having the structure [ADJECTIVE + energy] in our Bootstrapped corpus.

Table 22. [ENERGY+NOUN] in the three sub-corpora

Table 22. [ENERGY+NOUN] in the three sub-corpora

59Though the energy efficiency cluster is common to all the three sub-corpora, its frequency of occurrence is more than doubled in the BootCat corpus (1004 instead of 407 previously). In this sense, the DIY corpus shows a greater degree of homogeneity in the use of this cluster by the three sub-corpora (418, 484 and 407 for the NGO, UN and RES corpora respectively). Both DIY and BootCat corpora identify four clusters common to all the three sub-corpora, namely energy demand, energy use, energy efficiency and energy system in the DIY corpus and energy use, energy consumption, energy efficiency and energy sources in the BootCat. Though the four clusters are not identical in the two corpora, they are semantically close as they refer to the energy demand, energy consumption and sources of energy that constitute the whole energy system. The results in both corpora identify previously observed combinations of sub-corpora and confirm the proximity of the NGO corpus with the other two, and the lack of links between the UN and RES sub-corpora.

60As for features characterising discourse communities, interestingly, both DIY and BootCat corpora point to energy outlook and energy policy as exclusively characteristic of the NGO and UN sub-corpora correspondingly. Energy technologies that would enable energy savings seem to be the main focus of the RES corpus in both DIY and BootCat.

Table 23. [ADJECTIVE+ENERGY] in the three sub-corpora

Table 23. [ADJECTIVE+ENERGY] in the three sub-corpora

61The three clusters international energy, world energy, and global energy in the B-NGO corpus could reveal a greater focus on energy issues at a global scale. The first two of these clusters are equally present in the DIY corpus.

62The UN sub-corpus seems to be concerned with long-term solutions to climate issues with clusters, such as impact energy, building energy and final energy. The cluster national renewable energy seems to confirm a focus on the national scale in the RES sub-corpus. Not surprisingly, the cluster renewable energy ranks top in the RES sub-corpus with 2 882 occurrences.

63In the BootCat corpus, the only cluster that appears as common to the three sub-corpora is renewable energy. Unlike in the DIY corpus, the cluster global energy no longer figures. Similarly, the number of clusters common to NGO and UN corpora gets down to one in the BootCat, instead of three in the DIY corpus. International energy appears as the fourth most common cluster in the UN corpus in the BootCat and the 3rd most common in the DIY. Primary energy and world energy are no longer identified in the BootCat. Nevertheless, in the comparative analysis of the NGO and RES corpora, the BootCat results identify four clusters, sustainable energy, clean energy, global energy, primary energy instead of one, that of clean energy in DIY corpus. More interestingly, what is identified as common to NGO and UN in DIY corpus (primary energy) is done so for the combination of NGO and RES sub-corpora here.

Conclusion

64Our comparative analysis of first results obtained in the DIY and BootCat corpora seems to suggest that BootCat can be a very useful and efficient tool to extend existing small corpora. The specialised lexical units and terms can thus be considered in more contexts and the identification of collocations and multi-word units can be enhanced. To get these results, one must be aware of the importance of building a strongly representative corpus in the first stage, as the keywords extracted from the first corpora will be all the more important as they will be used as seeds to constitute the second. Although we believe this experience shows that BootCat and equivalent bootstrapping tools should not be neglected as potential extensions to existing corpora, it has also shown certain limits. There are variations between discourse communities which were visible in the DIY corpora and which are not in the BootCat corpora. The loss of specificity concerning author, date, and genre also means that there are many questions which could be considered in our DIY corpus that would be meaningless in the BootCat corpus. This enables us to highlight how important it is to question your objectives when you adopt a corpus constitution method. Having a large corpus that can itself be divided into small sub-corpora, according to the questions we are aiming to answer, is definitely more useful than having a single, undifferentiated, corpus on the issue. Proceeding through steps and contrasting small corpora that can then be combined into a big one to answer certain questions seems like a good way of building a corpus (Chareaudau 2009). The study presented here is a preliminary work to check the representativity of our corpora. So far, we have mainly presented quantitative data on the corpora but aim to analyse them in more qualitative terms to further the understanding of the concepts of interest.

Haut de page

Bibliographie

Anthony L. (2014). AntConc (Version 3.4.3) [Computer Software]. Tokyo, Japan: Waseda University. Available at http://www.laurenceanthony.net/.

Baker P. (2004). Querying keywords: questions of difference, frequency and sense in keywords analysis, Journal of English Linguistics 32(4): 346-359.

Baroni M. & Bernardini S. (2004). BootCaT: Bootstrapping corpora and terms from the web”, Proceedings of LREC.

Bernardini S. (2006). Corpora for translator education and translation practice: achievements and challenges, Proceedings of the L4Trans Workshop at LREC.

Bullard R. D. (1990). Dumping in Dixie: Race, class, and environmental quality. Boulder, CO: Westview.

Charaudeau P. (2009). “Dis-moi quel est ton corpus, je te dirai quelle est ta problématique”, Corpus 8 : 37-66.

Dahan A. & Aykut S. (2015). Gouverner le climat ? Vingt ans de négociations internationales. Paris: Presses de Sciences Po.

Davies M. (2010). “The Corpus of Contemporary American English as the first reliable monitor corpus of English”, Literary and Linguistic Computing 25(4): 447-464.

Heffron R. J., McCauley D. & Sovacool B.K. (2015). Resolving society’s energy trilemma through the Energy Justice Metric, Energy Policy 87: 168-176.

Heffron R. J. & MacCauley D. (2017). “The concept of energy justice across the disciplines”, Energy Policy 105: 658-667.

Kilgarriff A., Baisa Vit, Busta J. et al. (2014). The Sketch Engine: ten years on, Lexicography 1: 7-36.

Kilgarriff A. & Grefenstette G. (2003). Introduction to the special issue on the web as corpus”, Comput. Linguist. 29(3): 333-347.

L’Hôte E. & Lemmens M. (2009). “Reframing treason: metaphors of change and progress in new Labour discourse”, CogniTextes [Online], Volume 3.

Belinda M. (1997). Do-it-yourself corpora… with a little bit of help from your friends!, in B. Lewandowska-Tomaszczyk & P. J. Melia (eds.) PALC ’97 Practical Applications in Language Corpora. Lodz: Lodz University Press, 403-410.

Mary Robinson Foundation – Climate Justice. (2013). “Climate Justice Baseline” report, published July 2013. http://www.mrfcj.org/resources/climate-justice-baseline/.

Nicholson S. & Chong D. (2011). “Jumping on the Human Rights Bandwagon: How Rights-Based Linkages can Refocus Climate Politics”, Global Environmental Politics 11(3): 121-136.

Rayson P. (2008). “From key words to key semantic domains”, International Journal of Corpus Linguistics 13(4): 519-549.

Rhaman M. (2016). Climate Justice Framing in Bangladeshi Newspapers, 2007-2011”, South Asia Research 36(2): 186-205.

Rühlemann C. & Aijmer K. (2014). Introduction. Corpus pragmatics: laying the foundations”, in K. Aijmer & C. Rühlemann (eds.) Corpus Pragmatics: A Handbook. Cambridge: Cambridge University Press, 1-26.

Sovacool B.K., Burke M., Baker L. et. al. (2017). “New frontiers and conceptual frameworks for energy justice”, Energy Policy 105: 677-691.

Swales, J. (1990). “The Concept of Discourse Community”, in Genre Analysis: English in Academic and Research Settings. Cambridge: Cambridge University Press, 21-32.

Varantola K. (2003). “Translators and Disposable Corpora”, in F. Zanettin, S. Bernardini & D. Stewart (eds.) Corpora in Translator Education. Manchester: St. Jerome, 55-70.

Zanettin F. (2002). Corpora in translation practice, Proceedings of the First International Workshop on Language Resources (LR) for Translation Work and Research, 10-14.

Haut de page

Notes

1 Log-Likelihood ratios are used within statistical tests to assess differences. Here the comparison is between normalised frequency distributions (see L’Hôte and Lemmens 2009 for more detailed explanations).

2 See http://ucrel.lancs.ac.uk/usas/semtags.txt for a complete list of tags.

3 To identify key semantic categories in a corpus the WMatrix software compares it to a reference corpus already integrated in the software. We used the one named British English 2006.

4 Pollutant associated to conventional energy production.

5 Seeds are starting points for automatic retrieval. Web-crawlers typically use URLs as seeds, and BootCat is an original tool in that it takes word combinations to generate those URLs.

6 The term refers to an ordered sequence of a number of items: in BootCat, the user chooses “tuple length”, i.e. whether they want to use sequences of two, three, four or more seeds.

7 While BootCat outputs just one text file, the SketchEngine creates a table associating each retrieved file to its word count and enables users to delete specific files before compiling the corpus.

Haut de page

Table des illustrations

Titre Table 1. DIY Corpus created through vertical and horizontal methodologies
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-1.png
Fichier image/png, 120k
Titre Table 2. Keywords in each sub-corpus
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-2.png
Fichier image/png, 223k
Titre Table 3. Semantic categories in each sub-corpus
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-3.png
Fichier image/png, 211k
Titre Table 4. The most used lexical items in the category "Weather"
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-4.png
Fichier image/png, 191k
Titre Table 5. The most used lexical items in the "Green Issues" category
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-5.png
Fichier image/png, 178k
Titre Table 6. The most used lexical items in the "Places" category
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-6.png
Fichier image/png, 192k
Titre Table 7. The most used lexical items in the "Change" category
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-7.png
Fichier image/png, 229k
Titre Table 8. The most used lexical items in the "Science and Technology" category
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-8.png
Fichier image/png, 212k
Titre Table 9. Frequency counts with Antconc
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-9.png
Fichier image/png, 107k
Titre Table 10. N-Grams starting with Vulnerab*
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-10.png
Fichier image/png, 142k
Titre Table 11. N-Grams starting with Responsib*
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-11.png
Fichier image/png, 92k
Titre Table 12. [ENERGY+NOUN] in the three sub-corpora
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-12.png
Fichier image/png, 147k
Titre Table 13. [ADJECTIVE+ENERGY] in the three sub-corpora
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-13.png
Fichier image/png, 144k
Titre Table 14. The BootCat Corpora
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-14.png
Fichier image/png, 123k
Titre Table 15. Comparative Analysis of DIY and BootCat Results for the Semantic Categories
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-15.png
Fichier image/png, 99k
Titre Table 16. Comparative analysis of discourse community specificities (Weather)
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-16.png
Fichier image/png, 91k
Titre Table 17. Comparative analysis of discourse community specificities (Change)
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-17.png
Fichier image/png, 79k
Titre Table 18. Comparative analysis of discourse community specificities (Science & Technology in general)
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-18.png
Fichier image/png, 101k
Titre Table 19. Frequency counts of terms of interest in the Bootstrapped corpus with Antconc
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-19.png
Fichier image/png, 92k
Titre Table 20. Word clusters with VULNERAB*
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-20.png
Fichier image/png, 140k
Titre Table 21. Word clusters with RESPONSIB*
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-21.png
Fichier image/png, 77k
Titre Table 22. [ENERGY+NOUN] in the three sub-corpora
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-22.png
Fichier image/png, 167k
Titre Table 23. [ADJECTIVE+ENERGY] in the three sub-corpora
URL http://journals.openedition.org/corpus/docannexe/image/3376/img-23.png
Fichier image/png, 142k
Haut de page

Pour citer cet article

Référence électronique

Camille Biros, Caroline Rossi et Inesa Sahakyan, « Discourse on climate and energy justice: a comparative study of Do It Yourself and Bootstrapped corpora », Corpus [En ligne], 18 | 2018, mis en ligne le 24 septembre 2018, consulté le 14 novembre 2018. URL : http://journals.openedition.org/corpus/3376

Haut de page

Auteurs

Camille Biros

Université de Grenoble

Caroline Rossi

Université de Grenoble

Inesa Sahakyan

Université de Grenoble

Haut de page

Droits d’auteur

© Tous droits réservés

Haut de page
  • Logo Revues électroniques de l’université de Nice
  • OpenEdition Journals