A “Wind of Change” – Shaping Public Opinion of the “Arab Spring” Using Metaphors

Motivation
How does mass media affect the way we think about controversial topics such as the “Arab Spring”? What persuasive role do metaphors play especially in opinion pieces?

During the events of the years 2010-2011 in the Middle East & North Africa region a new discourse was established in the German media; immediately these events were assessed as a “wave” of democratization and liberation, and have been metaphorically labeled “Arab Spring”. Metaphors were frequently used to categorize and understand these events (Möller, 2014; Nunez, 2014).

Given the premise that mass media organizes (Couldry, 2010) and shapes social reality (Luhmann, 1996), we analyze how the Arab Spring is categorized and assessed using metaphorical constructions in newspaper opinion pieces. We show ways in which particularly the use of metaphors reveals how the media tried to achieve acceptance for the events based on our cultural models (Quinn and Holland, 1987), which are grounded on our western knowledge.

According to the Conceptual Metaphor Theory (Lakoff and Johnson, 1980) metaphors are ubiquitous and exhibit a binary source-target domain structure. The knowledge that we choose to function as a source domain illustrates which conventionalized, overt or tacit knowledge we require to understand new or abstract domains (target domains) in terms of our cultural imprints. Metaphors are instantiated on the text surface and give us clues toward our knowledge basis. Thus, the required knowledge can be described in terms of ubiquitous metaphorical patterns that function as semantic “anchors” in texts, and in terms of conceptual knowledge clusters that function as an intertextual semantic knowledge structure.

As such, we constructed a pipeline that automatically detects metaphors appearing within certain grammatical constructions, before clustering them by presumed source and target domains. The results give us insights into how the Arab Spring is metaphorically structured by semantic clusters in opinion pieces.

Corpus and annotation

Our corpus consists of 300 opinion pieces (Ramge and Schuster, 2001) from five German newspapers, Frankfurter Rundschau, Die Zeit, Der Spiegel, taz, and Die Welt, which have been written between December 2010 and November 2011 and cover the Arab Spring.

In nine of these opinion pieces, two of this abstract's authors annotated following grammatical constructions whether they constitute metaphors: adjective-noun (AN) pairs (e.g. “Tunisian spark”), and genitive constructions (GEN) (e.g. “torch of freedom”). Due to their interrelated components they provide a good

insight into the structural systematicity of metaphorical mappings (source domain target domain). The difficulty of the task is reflected in a low inter-annotator agreement (0.45 Krippendorff’s a).

Common sources of annotation disagreement included heavily conventionalized metaphors such as “social network”, personifications like “self-consciousness of a generation”, and metaphors that need a larger context to function. As gold standard for further

training and evaluation we only use the agreed upon

annotations (“annotated”, Table 1).

Sentences AN constructions AN metaphors GEN constructions GEN metaphors

annotated

538

968

116 (12%)

102

29 (28%)

complete

11402

19573

2758

Table 1: Constructions and metaphors in the corpus.

Technical realization
To examine our questions quantitatively, we contrast two approaches to automatically detect metaphors, namely random forests and multilayer-percep-tron. The extracted metaphors are subsequently clustered (Figure 1). To extract AN and GEN constructions we first perform automatic preprocessing, including part-of-speech tagging and dependency parsing.

The random forests approach of Tsvetkov (2014) firmly roots in conceptual metaphor theory, mainly employing features extracted from manually crafted resources such as an abstractness wordlist and supersenses, to classify adjective-noun and subject-verb-object constructions. For use on other languages than English, a bilingual dictionary is required. We manually expand an existing dictionary to cover our corpus, and extend their system to classify GEN metaphors.

The described feature-rich approach will be compared - with regards to what (kind of) metaphors can be found - to the shallow neural network approach by Do Dinh and Gurevych (2016), which does not presuppose any specific metaphor theory. It thus does not make use of external features, but rather learns exclusively from given annotations and their context. Preliminary experiments show that more training data is needed for this bottom-up approach.

To gain further insight into usage of metaphor in our corpus, we cluster the automatically found metaphors - resp. their components - into coarse-grained semantic fields. While there are works using a theory-supported top-down approach, e.g. using source domain lists (Gordon et al., 2015), we employ a rather unsupervised approach, without preselecting the number of clusters or manually fixing cluster centers (similar to Shutova et al. (2010), who use spectral clustering for metaphor detection). To that end, we employ Affinity Propagation (Frey and Dueck, 2007), which we supply with cosine similarities between pretrained word embeddings (Reimers et al., 2014) of the metaphor components.

Figure 1: Few newspaper opinion pieces are annotated (1) and the obtained metaphors are used to leam different models A and B (2) (Tsvetkov, 2014; Do Dinh & Gurevych, 2016). These models are then applied to more articles to extract metaphors (3), which are subsequently clustered (4).

Experiments and discussion
We use cross-validation for the intrinsic evaluation of the metaphor detection part. For GEN metaphors, the tested system achieves 0.63 precision, 0.25 recall, and 0.35 F1-score, with similar performance for AN metaphors. While these results seem low, the actual output of the system when trained on all annotated instances looks promising, and the precision is improved by filtering based on named entities.

Kopf: Schwertern des Islams, Köpfen des verhassten Regimes, Handlanger des Regimes, Schlägern des Regimes, Arm des alten Regimes, Zähne eines Kamms, Kugeln des Regimes, Brust des Leblosen, Gesicht des Exzesses, Gesicht der EZB, Gesicht des Arabischen Frühlings, Gesicht arabischer Demokratien, Gesichter der Demonstranten, Gesichter der Jasmin-Revolution, Gesicht der Revolution, Gesicht der Muslimbruderschaft, Gesicht der ägyptischen Revolution, Gesicht des Landes, Gesicht der tunesischen Revolution, Gesicht der USA, Gesicht ihres Sohnes, Gesichter der Vermissten, Fortsatz des alten Regimes, Wand der Angst, Loch der Diktatur, Schweine des Regimes, Augen vieler Araber, Hände des israelfeindlichen Regimes, Hand der Sozialisten, Hände des Obersten Kommandorates, Händen der Börse

Figure 2: GEN metaphors clustered by first noun, with center “Kopf, hinting at conceptual metaphor POLITICAL SYSTEMS ARE BODIES

The clustering creates an impression of which knowledge (source domain) is required for abstract concepts (target domains), and how abstract concepts are “perspectivized” in the corpus, while also giving an overview of occurring intertextual metaphors. Although the cluster assignment and the metaphor detection leave room for improvement (e.g. Figure 2: “face of her son”), the clusters already reveal the systema-ticity and constraints of metaphorical mappings. Thus, they point to strategies of newspapers that come along with the choice of the (conceptual) source domain.

In Figure 2, bodily parts such as face, head, hand are used as source domains and mapped to political systems or processes (e.g. regime, democracy, revolution). This mapping draws on a long tradition in political and

philosophical history (Musolff, 2004): head and face

play a central role in our culture - comparing political processes with faces or heads conceptualize them as human beings. In this cluster the construction face of indicates that the events are important, thus construed as worthy to support.

Those prototypical examples for ontological metaphors also support the premise of embodied cognition (Johnson, 1987; Rohrer, 2010).

Sturm: Schlussphase des alten Regimes, Sturm des arabischen Umbruchs, Sturm der Moleküle, Sturm der Entrüstung, Inseln der Diktatur, Insel der Stabilität, Wind der Freiheit, Wind der Demokratie, Wind der Revolution, Wind des Wandels

Figure 3: GEN metaphors clustered by first noun, with center “Sturm”, hinting at conceptual metaphor POLITICAL SYSTEMS/PROCESSESAZALUES ARE NATURAL ELEMENTS

The positive properties and the movement character of natural elements such as wind and storm are mapped to the abstract (political) nouns freedom, revolution, or change. They receive a deontic (Hermanns, 1994) character, whereas dictatorship is conceptualized in terms of island which stands for inertia and stability (Figure 3). These examples already show how the chosen metaphors shape dualistic tendencies by categorizing the events on one hand as a dynamic movement (wind, storm) that has to be supported by western democracies, or on the other hand pleading for stability (island), thus implicitly supporting dictatorship.

The analyzed clusters and metaphorical conceptualizations indicate a network of source domains that function as key concepts which structure the discourse of the Arab Spring, an assumption we will focus on in future work.

Conclusion and future work

Our study indicates that metaphorical constructions are important in media because of their ubiquitous use in opinion pieces. The generic extracted source domains already suggest that a specific network of knowledge is used in media to highlight certain political aspects of the Arab Spring. Furthermore, they illustrate how contents are emotionalized and ideologized during these events by metaphors and their framing effects. Usage of natural elements or body parts reduces complexity and conceptualizes the events as an organic development, in short: the Arab states become western democratic states. Thus contributing to the extension of western ideology, metaphors impart implicit cultural values. Combining our cognitive and discourse analytical questions we can summarize that the used “bottom-up” clustering is very helpful to get an explorative impression of the “intertextual consistencies” (Verschuren, 2012: 179) of chosen metaphors. They are good textual “anchors” and starting points to investigate the widespread metaphorical use, and thus knowledge domains, in corpora.

With regard to the state of the art, corpus-based methodologies within the Digital Humanities community will benefit from our research by gaining the possibility to automatically compare thematic corpora by using the relationship of their metaphors to the common main cluster as a metric, therefore obtaining a new way to analyze the conceptual network being used. Our approach can help to facilitate corpus studies, e.g. by analyzing other discourse segments which deal with the implicit construction of identity and alterity within opinion pieces by using metaphors.

In our presentation we will highlight the results and give a structured impression of the mappings and the implications of the used metaphors in our corpus and present in short our methodological basis.

In future studies we will compare the conceptualization strategies of the Arab Spring and “Refugee Crisis” in German media, since we assume that the same metaphors and the same (metaphorical) interpretation patterns occur. Further, we plan to investigate another theory of metaphor which is based on Black (1954, 1977) and Gehring (2010). The latter model is strongly interweaved with current discussions about “Begriff” (Müller-Meiningen and Schmieder, 2016; Gehring, 2005, 2010) and discusses its ideological im-plication(s). Furthermore, the emphasis is placed on the function of metaphors as an epistemological tool by investigating, amongst others, the evolution of ideas and cultural values, e.g. in the historical text collection “Natur&Staat” (1903-11).

Bibliography

Black, M. (1977). More about Metaphor. Dialectica, 31: 431

457.

Black, M. (1954). Metaphor. Proceedings of the Aristotelian

Society. New Series, 55: 273-294.

Couldry, N. (2010). Media discourse and the naturalization of categories. In Wodak, R. and Koller, V. (eds), Handbook of communication in the Public Sphere. Berlin / New York: Walter de Gruyter, pp. 67-88.

Dirven, R., Polzenhagen, F., and Wolf, H.-G. (2010). Cognitive Linguistics, Ideology, and Critical Discourse Analysis. In Geeraerts, D. and Cuyckens, H. (eds), The Oxford

Handbook of Cognitive Linguistics. Oxford: Oxford University Press, pp. 1223-1240.

Do Dinh, E., and Gurevych, I. (2016). Token-Level Metaphor Detection using Neural Networks. In Proceedings of the Fourth Workshop on Metaphor in NLP. San Diego, CA, USA: Association for Computational Linguistics, pp. 28-33.

Frey, B. J., and Dueck, D. (2007). Clustering by Passing Messages Between Data Points. Science, 315(5814): 972976.

Gehring, P. (2005). Vom Begriff zur Metapher. Elemente einer Methode der historischen Metaphernforschung. In Abel, G. (ed), Kreativität. Kolloquiumsbeiträge des XX.

Kongresses der Allgemeinen Gesellschaft für Philosophie in Deutschland. Hamburg: Meiner, pp. 800-815.

Gehring, P. (2009). Das Bild vom Sprachbild. Die Metapher und das Visuelle. In Danneberg, L., Spoerhase, C., and Werle, D. (eds), Begriffe, Metaphern und Imaginationen in Philosophie und Wissenschaftsgeschichte. Wiesbaden: Harrassowitz, pp. 81-101.

Gehring, P. (2010). Erkenntnis durch Metapher? Methodologische Bemerkungen zur Metaphernforschung. In Junge, M. (ed), Metaphern in Wissenskulturen. Wiesbaden: Verlag für Sozialwissenschaften, pp. 203-220.

Gehring, P., and Gurevych, I. (2014). Suchen als Methode? Zu einigen Problemen digitaler Metapherndetektion. Journal Phänomenologie, 41: 99-110.

Gordon, J., Hobbs, J. R., May, J., Mohler, M., Morbini, F., Rink, B., Tomlinson, M., Wertheim, S. (2015). A Corpus

of Rich Metaphor Annotation. In Proceedings of the Third Workshop on Metaphor in NLP. Denver, CO, USA: Association for Computational Linguistics, pp. 56-66.

Hermanns, F. (1994). Schlüssel-, Schlag- und Fahnenwörter; zu Begrifflichkeit und Theorie der lexikalischen “politischen Semantik”. Erste Fassung eines Überblicksartikels zum Forschungsstand in Sachen Schlüsselwort- und Schlagworttheorie und -forschung für den Ergebnisband de. Heidelberg.

Johnson, M. (1987). The body in the mind. The bodily basis of meaning, imagination, and reason. Chicago: Chicago University Press.

Lakoff, G. (2006). Conceptual Metaphor. The contemporary theory of metaphor [1993]. In Geeraerts, D. (ed), Cognitive Linguistics: Basic Readings. Berlin: Mouton de Gruyter, pp. 185-238.

Lakoff, G., and Johnson, M. (1980). Metaphors we live by. Chicago: Chicago University Press.

Möller, N. (2014). Cognitive Metaphor and the “Arab spring.” In Polzenhagen, F., Kleinke, S., Kövecses, Z., and Vogelbacher, S. (eds), Cognitive Explorations into Metaphor and Metonymy. Frankfurt (Main): Peter Lang, pp. 133-148.

Musolff, A. (2003). The heart of the European body politic: British and German perspectives on Europe's central organ. Journal of multilingual and multicultural development., 25: 437-452.

Müller-Meiningen, E., and Schmieder, F. (2016). Begriffsgeschichte und historische Semantik. Berlin: Suhrkamp Verlag.

Luhmann, N. (2009). Die Realität der Massenmedien. Wiesbaden: Verlag für Sozialwissenschaften.

Nunez, A. (2014). Wenn das “Embodiment” politisch wird: Das Image-Schema PATH und seine Realisierung im Mediendiskurs zum „Arabischen Frühling”. In Polzenhagen, F., Kleinke, S., Kövecses, Z., and Vogelbacher, S. (eds), Cognitive Explorations into Metaphor and Metonymy. Frankfurt (Main): Peter Lang: pp. 149-164.

Quinn, N., and Holland, D. (1987). Culture and Cognition. In Holland, D. and Quinn, N. (eds), Cultural models in language and thought. Cambridge: Cambridge University Press, pp. 3-42.

Ramge, H., and Schuster, B.-M. (2001). Kommunikative Funktionen des Zeitungskommentars. In Leonhard, J.-F., Ludwig, H.-W, Schwarze, D., and Straßner, E. (eds), Medienwissenschaft. Ein Handbuch zur Entwicklung der Medien und Kommunikationsformen. Berlin / New York: Mouton de Gruyter, pp. 1702-1712.

Reimers, N., Eckle-Kohler, J., Schnober, C., Kim, J., and

Gurevych, I. (2014). GermEval-2014: Nested Named Entity Recognition with Neural Networks. In Workshop Proceedings of the 12th Edition of the KONVENS Conference. Hildesheim, Germany: Universitätsverlag Hildesheim, pp. 117-120.

Rohrer, T. (2010). Embodiment and experientialism (2010). In Geeraerts, D. and Cuyckens, H. (eds), The Oxford Handbook of Cognitive Linguistics. Oxford: Oxford University Press, pp. 25-47.

Shutova, E., Sun, L., and Korhonen, A. (2010). Metaphor identification using verb and noun clustering. In Proceedings of the 23rd International Conference on Computational Linguistics. Shanghai, China: Association for Computational Linguistics, pp. 1002-1010.

Tsvetkov, Y., Boytsov, L., Gershman, A., Nyberg, E., and

Dyer, C. (2014). Metaphor Detection with Cross-Lingual Model Transfer. In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics. Stroudsburg, PA, USA: Association for Computational Linguistics, pp. 248-258.

Verschuren, J. (2012). Ideology in Language Use. Pragmatic Guidelines for Empirical Research. Cambridge: Cambridge University Press.

Ziegler, H.E., Conrad, J., Haeckel, E. (eds) (1903-1911). Natur und Staat, Beiträge zur naturwissenschaftlichen Gesellschaftslehre. Jena: Gustav Fischer Verlag.