Conclusion 15

Acknowledgements 16

References 17

Appendix A OpinionFinder Results in Trial Test 19

Appendix B List of Voting Sessions used in Analysis 20

Appendix C Top 20 Smallest Deviations in Data Calibration 26

Appendix D Sentimental / Frequency Results 27

Appendix E Combined Prototype Class Demonstration 28

Figure 1 Halliday’s Model of Interpersonal Metafunction 3

Figure 2 Sample Page of a Provisional Record。 7

Figure 3 Sample Entry of the Corpus (CSV Format) 8

Figure 4 Prototype Class Demonstration and Comparison 14

Table 1 List of Parallel Concepts in Halliday’s, Li’s and Quirk’s theories 4

Table 2 Top 10 verbs in normal, UN speech and vote-in-favour speech contexts 10

Table 3 Example Verb Pairs Matching the Phenomenon 12

第 II 页 本 科 毕 业 论   文

1 Introduction

1。1 Research Background

Language profoundly connects and affects politics, culture (Fu, 2004) and economy (Chen, 2003)。 This ability to influence the world is defined as interpersonal metafunction of language by M。

A。 K。 Halliday (2008)。 Understanding the interpersonal usage of language in the practice of Sentiment Analysis and other related Natural Language Processing field (Stuart & Majewski, 2015) provides the way to calculate language which is essential to advance pervasive computing。

The connection and relationship between language and politics have become a hotspot in linguistic research (Sun, Peng, & Liu, 2015) considering the importance of political discourse in both diplomacy and propaganda。 Utilizing and resolving interpersonal metafunction in these discourses enables NLP program to analyze the political tendencies of the context, for example, voting intentions in cases such as a United Nations Security Council’s vote。

Members of the Security Council give speeches in a voting session of a Draft Resolution, stating their country’s position on the matter discussed, therefore potentially indicating their voting intentions。 Achieving prediction of such intentions from interpersonal sentiments revealed in the speech by machine has a broad range of uses。

1。2 Significance of the Study

Unlike other political discourses that heavily utilize interpersonal metafunction of language to transmit the speaker’s persuasion and propaganda intention (Jowett & O’Donnell, 2006), speeches delivered in a Security Council meeting contains less seditious elements and more cautious expressions, “unfriendly” to NLP analyzing。

A trial test in this research used OpinionFinder (Wilson, Wiebe, & Hoffmann, 2005) with default model to identify speaker’s voting intention based on positiveness1 in 1676 outcome-known Security Council speech。 The result shows only a 58。33%2  accuracy in the best-case scenario。 Research on

2   An extremely low threshold is used in this scenario, thus should not be consider applicable。 Applicable results ranged

verbal interpersonal metafunction in a UN meeting context can help improve NLP models on these speech scenarios。