v ictoria rau yi-hsin wu and meng-chien yang

22
Elizabeth Zeitoun, Stacy F. Teng and Joy J. Wu, eds. New Advances in Formosan Linguistics, 533-554 Asia-Pacific Linguistics, 2015 Copyright held by the authors, released under Creative Commons Attribution Licence (CC BY 4.0) 533 22 A corpus-based approach to the classification of Yami emotion* VICTORIA RAU, YI-HSIN WU AND MENG-CHIEN YANG 1 Introduction The point of departure for an investigation of emotion in a language is its lexicon (Saucier & Goldberg 1996). Although emotion concepts, such as happiness, anger, sadness and fear, are intuitively clear and can be found in various languages, we cannot assume all languages have the same set of emotions. 1 Church et al. (1998) found that the “hypercognised” 2 (Levy 1984) emotions in Filipino 3 are anger, anxiety/fear, happiness, contentment, sadness, and arousal, whereas the relatively “hypocognised” emotion domains include feeling tired, guilty, surprised, contemptuous, and aspiring. They also recommended that the terms in all three subcategories of the affective conditions class (i.e., pure affective, affective-behavioral, and affective-cognitive states) in Clore et al.’s (1987) taxonomy of emotion terms be viewed as referring to emotions. Although Church * This study is partially supported by two NSC grants: “A typological study of Austronesian languages in Taiwan and their revitalisation” (NSC100-2420-H-194-011-MY3), 1 November 2011–31 October 2014, and “Yami ontology: Yami lexical semantics and sociogrammar” (NSC 100-2410-H-194-104), 1 August 2011–31 July 2012.Various parts of our research have been presented at the 2012 CLDC (Rau, Wu, Yang & Hu 2012), the 2012 International Conference on Landscape, Seascape, and the Spatial Imagination (Rau, Wu & Yang 2012), and the International Conference on Asian Language Processing (Yang, Rau & Wu 2012). We would like to thank the three anonymous reviewers for their helpful comments and the rigorous work of the editors of this volume, which has significantly improved the technical writing of this paper. All remaining errors are ours. 1 Emotion can be used as either an uncountable or countable noun, depending on whether the collective concept of emotion is intended (singular) or individual references of emotion are intended (plural). 2 Hypercognised and hypocognised emotions refer to the dichotomy between most important/maximally lexicalised and least important/minimally lexicalised emotion terms. 3 The word ‘Filipino’ is used by Church et al. (1998). Filipino is the official name of the national language of the Philippines, primarily based on Tagalog (http://www.ethnologue .com/country/PH/languages). Although Tagalog is more frequent in common parlance, Filipino or Pilipino underlines its role as the national language. The name Pilipino is also used in the textbook title Pilipino through Self-Instruction (Wolff et al. 1991) to emphasise its role as a widely used second language for inter-group communication.

Upload: others

Post on 04-Apr-2022

3 views

Category:

Documents


0 download

TRANSCRIPT

Elizabeth Zeitoun, Stacy F. Teng and Joy J. Wu, eds. New Advances in Formosan Linguistics, 533-554 Asia-Pacific Linguistics, 2015 Copyright held by the authors, released under Creative Commons Attribution Licence (CC BY 4.0) 533
22 A corpus-based approach to the classification of Yami emotion*
VICTORIA RAU, YI-HSIN WU AND MENG-CHIEN YANG
1 Introduction The point of departure for an investigation of emotion in a language is its lexicon
(Saucier & Goldberg 1996). Although emotion concepts, such as happiness, anger, sadness and fear, are intuitively clear and can be found in various languages, we cannot assume all languages have the same set of emotions.1 Church et al. (1998) found that the “hypercognised”2 (Levy 1984) emotions in Filipino3 are anger, anxiety/fear, happiness, contentment, sadness, and arousal, whereas the relatively “hypocognised” emotion domains include feeling tired, guilty, surprised, contemptuous, and aspiring. They also recommended that the terms in all three subcategories of the affective conditions class (i.e., pure affective, affective-behavioral, and affective-cognitive states) in Clore et al.’s (1987) taxonomy of emotion terms be viewed as referring to emotions. Although Church
* This study is partially supported by two NSC grants: “A typological study of Austronesian
languages in Taiwan and their revitalisation” (NSC100-2420-H-194-011-MY3), 1 November 2011–31 October 2014, and “Yami ontology: Yami lexical semantics and sociogrammar” (NSC 100-2410-H-194-104), 1 August 2011–31 July 2012.Various parts of our research have been presented at the 2012 CLDC (Rau, Wu, Yang & Hu 2012), the 2012 International Conference on Landscape, Seascape, and the Spatial Imagination (Rau, Wu & Yang 2012), and the International Conference on Asian Language Processing (Yang, Rau & Wu 2012). We would like to thank the three anonymous reviewers for their helpful comments and the rigorous work of the editors of this volume, which has significantly improved the technical writing of this paper. All remaining errors are ours.
1 Emotion can be used as either an uncountable or countable noun, depending on whether the collective concept of emotion is intended (singular) or individual references of emotion are intended (plural).
2 Hypercognised and hypocognised emotions refer to the dichotomy between most important/maximally lexicalised and least important/minimally lexicalised emotion terms.
3 The word ‘Filipino’ is used by Church et al. (1998). Filipino is the official name of the national language of the Philippines, primarily based on Tagalog (http://www.ethnologue .com/country/PH/languages). Although Tagalog is more frequent in common parlance, Filipino or Pilipino underlines its role as the national language. The name Pilipino is also used in the textbook title Pilipino through Self-Instruction (Wolff et al. 1991) to emphasise its role as a widely used second language for inter-group communication.
534 Victoria Rau, Yi-Hsin Wu and Meng-Chien Yang
et al. used a quantitative experimental method to validate Clore et al.’s emotion model, they relied on a translation of emotion terms from English rather than searching directly in Filipino. The translation approach is based on the researchers’ assumptions of universals in emotion. Although this has its place to provide “etic” data, one cannot be certain that the equivalents can be found in translation, not to mention that there is always something lost in translation. Thus to yield an “emic”4 perspective of emotion from a language, it is imperative to extract emotion terms directly from the language, but the problem is how to determine what constitutes an emotion term in a language.
Cognitive linguists have proposed to construe emotion based on a grammatical model for Formosan languages, such as Tsou (Huang 2002). Following Talmy’s (2000) suggestion that emotion events are inherently causal, Hsieh (2011) examined emotional causality in Kavalan, Paiwan, and Saisiyat. Their theoretical approach has provided a framework for the present study to explore Yami emotion. As Yami is the only Philippine language in Taiwan, it is also important to compare the results of Yami emotions with the results of Filipino obtained by Church et al. (1998).
How is emotion defined? Wierzbicka (1992) suggests that emotion concepts can be defined in terms of universal primitives, such as ‘good’, ‘bad’, ‘do’, ‘happen’, ‘know’, and ‘want’, and prototypical scripts formulated in terms of ‘thoughts’, ‘wants’, and ‘feelings’. In other words, an emotion event involves someone’s cognition, affection, or feelings about something that happened to someone or the fact that someone did something. The thoughts/wants/feelings can be evaluated as positive or negative. To discover what meets the definition of emotion from an emic perspective, this study aims at a grammatical model encoded by the prefix ika- ‘the reason/cause for a certain feeling’ to conceptualise Yami emotion concepts, complemented by an ontological approach to compare Yami emotions with Filipino. As recent interdisciplinary investigations on Yami fish names (Hu & Rau 2013) and Yami fish ontology (Tai et al. 2008; Rau et al. 2009) have brought us a better understanding of how metaphors are used in describing fish names and place names, we intend to adopt a similar interdisciplinary approach to investigating the classification and ontology of Yami emotion with the goal of building a complete ontology of the targeted language and culture.
Our aim is to explore the following three questions: 1. What are the hypercognised and hypocognised emotions in Yami? 2. Are there more positive or negative emotion terms in Yami? 3. Do Yami emotions share the same distinct domains as Filipino emotions? The organisation of this paper is as follows. After the brief introduction above, an
introduction to Yami morphology, with a focus on ka- and ika-, is presented. After that, the results of classification of Yami emotion are presented, followed by a comparison with Filipino emotion based on the results of a cluster analysis of Yami emotion. This paper ends with an evaluation of the application of this approach to the study of emotion.
4 The terms ‘emic’ and ‘etic’ are used in anthropology to contrast an insider’s view from an
outsider’s view. They are derived from the linguistic distinction between ‘phonemic’ and ‘phonetic’. Different sounds which are ‘phonetically’ different may be perceived by native speakers of a language as either the same or different ‘phonemically’ depending on the phonological system of the language.
A corpus-based approach to the classification of Yami emotion 535
2 Yami morphology: ika- To understand why the prefix ika- was chosen as the point of departure for the
investigation of emotion, a brief discussion of Yami morphology is in order. Yami verbs are classified either as dynamic or stative. Transitive verbs can be marked by one of the four focus markers: agent focus (AF) m-, patient focus (PF) -en, locative focus (LF) -an, and instrument focus5 (IF) i-. Stative verbs, which are mostly intransitive (i.e., agent focus) (see Table 1), are marked with the ma- prefix. Note that among the three types of ma-, only the second type is stative.
Table 1: Yami ma- verbs
ma-cimoy ‘rain’ (1) ma- agent focus ma-ngay ‘go’
(2) ma- agent focus stative verb
ma-saray ma-tava
‘happy’ ‘fat’
ma-cita ‘can see, visible’ (3) ma- patient focus potentive verb ma-hap ‘can get’
Dynamic transitive verbs (see Table 2), on the other hand, are marked by the p- prefix.
Here are some contrasting examples of verb forms with the m- and p- prefixes.
Table 2: Yami dynamic verbs
Intransitive Transitive mi-palit (AF mi-the root is palit ‘change’) ‘exchange’
pi-palit-en (pi-root-PF.en) ‘exchange’
panazang-an (paN-root-LF.an) ‘place where one bought’
maka-vonas (AF maka- the root is vonas ‘remove’) ‘can remove’
paka-vonas-en (paka-root-PF.en) ‘must remove’
maci-vazay (AF maci- the root is vazay ‘work’) ‘engage in work’
paci-vazay-an (paci-root-LF.an) ‘engage in work with someone’
The stative ka- as described by Zeitoun & Huang (2000) only appears when the transitive forms are affixed with the instrument focus prefix i- to indicate the O argument is in IF (instrument or reason of the action), as illustrated in Table 3. Note that the Yami prefix ka- is polysemous.6 It can form nouns (e.g., ka-tangked ‘nearby’, ka-paganam ‘dance’), verbs with a separate inflectional paradigm from the regular focus system (e.g., ka-doa ‘two in total’, ka-tangara ‘looked up just now’, ka-lavi ‘why cry (blaming)’, ka-teneng ‘then understand’, exclamatory sentences (e.g., ka-zakat ‘Go to hell!’, ka-lowlaw ‘so 5 Instrument Focus covers instrument, beneficiary, and reference. 6 A preliminary discussion of the functions of the ka- prefix can be found in Rau & Dong
(2006:132–134). For a more current description of ka-, consult Rau & Dong (2010, forthcoming).
536 Victoria Rau, Yi-Hsin Wu and Meng-Chien Yang
bloody good!’), or stative verb stems to derive verbs of instrument focus (e.g., i-ka-rahet ‘consider something bad because …’). The stative ka- is invisible in intransitive verbs, e.g., marahet ‘bad’, but when a transitive verb prefixed with i- is formed from a stative root, the ka- prefix marks its stative root, e.g., ikarahet ‘consider something bad because ...’.
Table 3: Yami stative verbs
Intransitive Transitive mazakat ‘killed, dead’ i-ka-zakat ‘killed because…’ marahet ‘bad’ i-ka-rahet ‘consider something bad because…’ mam’ing ‘smile’ i-ka-m’ing7 ‘amused because…’.
The emotion-related ika-8 served as a promising point of departure for searching Yami feeling and emotion based on a bottom-up approach because it helped us find the majority of Yami emotion roots. The stative roots (defined as roots that can co-occur with the ka- prefix) identified as emotion led us to find other derived verbs. For example, ikangsah ‘feel bored because of such and such (IF)’ can lead to other derived forms, such as mangsah ‘feel bored (AF)’ or angsahen ‘feel impatient about someone (PF)’ with the same root angsah ‘bored’. Take ika’oya ‘feel angry because of such and such (IF)’ as another example. We can find several other derived verbs with the same root ’oya ‘angry’: m’oya ‘angry with someone (AF)’, mi’oya’oya ‘very upset (AF)’, ’oyan ‘reason to be angry (LF)’, and i’oya ‘get upset with someone (IF)’. As illustrated above, this bottom-up approach, rooted in the form of the language, provides a reliable basis to search for the iconic relationship of isomorphism, i.e., same form, same function.
3 Methods This study adopted a corpus-based approach to find Yami emotion terms from the ika-
construction and compare the classification of Yami emotion terms with that of Filipino based on a cluster analysis. The methods are described in the following four steps.
7 As the orthography of the laryngeal features of /h/ and glottal stop in Yami remain to be
worked out, currently there are various ways of spelling i-ka-m’ing and mam’ing, such as ikamiying, ikamihing, mamiying, and mamihing. For an updated version of Yami phonology and orthography, see Rau & Dong (forthcoming).
8 The other homophonous, monomorphemic ika- encoding ordinal number, such as ika-dwa ‘the second’ is not related to emotion and was excluded from our discussion.
A corpus-based approach to the classification of Yami emotion 537
Step 1: Search for tokens with ika- from the Yami language documentation website To find potential Yami emotion terms, we began our study by extracting all 1763
tokens of ika- from our Yami corpus, consisting of sixty-three texts and the New Testament data from the Yami-language documentation website.9 The extraction of all the ika- tokens helped us identify the “construction” meaning (Goldberg 1995:4) of the grammatical model ika- as ‘the reason/cause for a certain feeling’. Although ika- is composed of the instrument focus i- and the prefix ka- in stative verbs, the frequent pairing of the form ika- with the meaning of ‘the reason/cause for a certain feeling’ has created a new “construction” which cannot be predicted from the composition of i- and ka- in Yami. Under this cause frame (Dirven 1997), we noticed that ika- can be prefixed with a wide range of word classes, from pronouns (e.g., ikaiya ‘he is the cause/reason’) to negation markers (e.g., ikabeken, ikabo ‘reason for being not’) and stative verbs (e.g., ikamo ‘reason to feel embarrassed’). Overall, ika- is most frequently prefixed to roots of stative verbs (e.g., masaray ‘happy’, ma’oya ‘angry’) and bare-root attributive modifiers (e.g., aro ‘many’, apia ‘good’) to form the most prototypical cause construction (e.g., ikasaray ‘reason to be happy’, ika’oya ‘reason to be indignant’, ikaro ‘reason to be abundant’, and ikapia ‘reason to be good’). However, the wide range of words co-occurring with the ika- construction still made it difficult to classify emotion according to Clore et al.’s (1987) taxonomy. Thus we decided to set aside the ika- tokens temporarily and search the emotion lexicon by analyzing twenty narrative texts (Rau & Dong 2006) with clear story lines, as the narrative context made the task of identifying and coding emotion words much easier.
Step 2: Search for Chinese translations of emotion expressions The search for emotion terms based on reading the twenty Yami texts with Chinese
translations (Rau & Dong 2006) helped us identify 258 potential emotion expressions, not all of which included ika-. We coded the 258 terms into nine categories. The first six categories fit Clore et al.’s (1987) framework, but the other three categories include interjections, curses, and metaphors/metonyms. The coding was jointly decided by the first two authors. Clore et al. (1987) made a distinction between internal and external conditions. As we did not find any token that would fit nicely in the category of “external” conditions, defined as (1) “subjective” evaluations of character or stable characteristics (e.g., attractive, trustworthy) and (2) “objective” conditions, such as things done to a person (e.g., abandoned, insulted), “external” conditions were excluded from the study.
The remaining internal conditions were further divided between mental and non-mental states, as shown in Figure 1.
9 The three Yami websites are Yami language documentation (http://yamiproject.cs.pu.edu
.tw/yami), Yami e-learning (http://yamiproject.cs.pu.edu.tw/elearn), and Yami online dictionary (http://yamibow.cs.pu.edu.tw).
538 Victoria Rau, Yi-Hsin Wu and Meng-Chien Yang
Internal conditions
Pure affective ikarilaw ‘sympathize’ ikaniahey ‘fear’ ikákey ‘like’
Pure cognitive ikakdeng ‘important’ ikanehed ‘true, certain’ ikacilo ‘noisy’
Physical ikabsoy ‘satiated’ ikasaki ‘drunk’ ikakaha ‘sleepy’
Affective-behavior ikam’ing ‘reason for laughing’ ikasnek ‘ashamed’ ikaciwciw ‘scare away’
Cognitive-behavior ikalma ‘lazy’ ikapili ‘picky’ ikaoyat ‘diligent’
Affective-cognitive ikagom ‘overbearing’ ikeylamnay ‘relaxed’ ikahanang ‘calm’
Figure 1: Classification of Yami emotion based on Church et al. (1998)
The internal non-mental states refer to physical and bodily states (e.g., sleepy, seasick). According to Clore et al. (1987), internal mental states consist of affective conditions and cognitive conditions. Under the category of affective conditions, we further separated pure affective states (e.g., afraid, angry, happy) from affective-behavioral states and affective-cognitive states, depending on whether the affective emotion is followed by an action (e.g., scare away, fight) or a cognitive consequence of the emotion (e.g., impatient, sorrowful). The category of cognitive conditions was similarly further divided into pure cognitive states and cognitive-behavioral states, with the former referring to the internal mental states in which cognition is dominant (e.g., smelly, stuffy) and the latter being followed by an action (e.g., picky, discreet). The other categories of emotion identified from the twenty texts included interjections/curses (e.g., ouch, damn) and metaphors/metonyms (e.g., the body is as healthy as light feathers or someone being as despised as goats). We can see in Table 4 that the emotion expressions from the twenty texts are divided into nine categories. Note that the first six categories contain the ika- prefix, while the last three do not.
A corpus-based approach to the classification of Yami emotion 539
Table 4: Coding categories for Yami emotion
Category Yami English10 1. Physical and bodily states ikakaha ‘sleepy because…’ 2. Pure affective states ikaniahey ‘afraid because…’ 3. Affective-behavioral states ikavozoaw ‘scare away because…’ 4. Affective-cognitive states ikaotok ‘impatient because…’ 5. Pure cognitive states ikangot ‘smelly because…’ 6. Cognitive-behavioral states ikapili ‘picky because…’ 7. Interjections ananay ‘ouch’ 8. Curses mo kavazat ‘Damn!’ 9. Metaphors/metonyms11 nimananat so velek ‘terrified (in the stomach)’
Step 3: Identification of the final set of Yami emotion terms As steps 1 and 2 led us to ascertain that the ika- prefix is really the key area to search
for emotion terms in Yami, we began our final search to find all the Yami emotion terms in 166 texts (including the twenty texts mentioned above and 146 other narratives from the three Yami websites). The final search yielded 12612 Yami emotion terms with the ika- prefix to serve as the database for categorisation and analysis in the present study. After the linguistic classification of the emotion terms was completed, a diagram was drawn using the Protégé program to represent the Yami emotion ontology.
Step 4: A cluster-analysis of Yami emotions for comparison with Filipino To explore the possibility of comparing our results with the previous study on Filipino,
a hierarchical cluster analysis was conducted to produce distinct domains of Yami emotion.
Following Church et al.’s (1998:78) procedure, a between-clusters linkage algorithm was calculated to produce comparable dendrograms for comparison with Filipino emotions. The cross-relationship between the emotion terms was calculated to build the hierarchical structure, using the knowledge extracted from our proposed ontological computation procedure. In contrast to Church et al.’s questionnaire methods for data collection, our study used a bottom-up corpus approach to create and grow the ontologies of the emotion concepts manually. These factors were used to calculate judgment values for evaluating whether an ika- emotion term could be put into a specific English emotion 10 As the English translations of the ika- ‘the reason/cause for a certain feeling’ examples are all
‘feel such and such because…’, we will not repeat the same frame but only translate the emotion terms in the rest of the paper.
11 Several body parts have been identified as related to emotion in metaphors. Due to the scope of this paper, we only list some examples here with the keywords bolded and will leave a systematic study on Yami “embodiment” in cognitive linguistics (Lakoff & Johnson 1999) for future investigation.
(1) ji anisomalap o pahad na. . ‘His soul has indeed flown away.’
(2) malaw no velek a kalawan. ‘worry to the stomach, i.e., very worried’
(3) do keyngeyngen na no oo. ‘sick to the head, i.e., have a headache’ 12 The 258 potential emotion expressions in step 2 contained both ika- words and three other
categories (exclamations, curses, and metaphors). After the 132 expressions without ika- were excluded, this yielded the final 126 terms. As our focus is on classification of different types of emotion, it is not our concern to report the tokens of each individual type.
540 Victoria Rau, Yi-Hsin Wu and Meng-Chien Yang
cluster. The clusters in each domain are represented by the key words and the frequency numbers of Yami ika- words with similar semantics (near synonyms) shown in parentheses after the ika- word. The relative distance from each ika- emotion term with the English translation was calculated using these factors. Finally, dendrograms were drawn to visualise the clustering results.13
4 Results Based on the analysis of the 126 Yami emotion terms with the ika- prefix, the
following section first presents the quantitative results of the most and least important emotion terms in Yami, followed by the distribution of positive and negative emotion terms, and ontological representation of semantic categories of Yami emotion. The second part of the results demonstrates the similarities and differences between Yami and Filipino.
4.1 Most important and least important emotions in Yami
What is considered the most or least important emotion in Yami is based on an interpretation of the frequency distribution of the 126 types. Overall, the pure cognitive category (e.g., good, bad, intelligent, difficult) constitutes the majority of Yami emotion terms, while affective-cognitive (e.g., lonely) and cognitive-behavioral categories (e.g., lazy) are the least frequent. The discovery of the most and the least important emotion domains in Yami generally matches Church et al.’s (1988) findings in Filipino data, except that there are no external conditions in Yami. As shown in Figure 1, Yami emotion based on the 126 terms with the ika- prefix constitutes three internal conditions, affective, cognitive, and physical, with the three affective conditions and the two cognitive conditions “hypercognised”. This finding supports Church et al.’s recommendation that the terms in all three subcategories of the affection conditions class in Clore et al.’s (1987) taxonomy of emotion terms be viewed as referring to emotions.
As shown in Table 5 and Figure 2, almost half of the emotion terms are cognitive (44% pure cognitive, e.g., ikacilo ‘noisy’) or cognitive related (5% cognitive-behavior, e.g., ikalma ‘lazy’), one third are affective (13% pure affective, e.g., ikarilaw ‘sympathise’; 13% affective-behavioral, e.g., ikami'ing ‘amused’; and 6% affective-cognitive, e.g., ikagom ‘overbearing’), and less than one fifth are physical (19% pure physical, e.g., ikabsoy ‘satiated’). Interestingly, if we had not investigated the ika- construction, we would not have discovered the saliency of the cognitive categories in Yami emotion.
13 The procedure reduplicated the illustrative dendrograms in Church et al.’s study for
comparison. The detailed steps of our ontological simulation for processing the ika- emotion terms supporting Church et al.’s study is described in Yang et al. (2012).
A corpus-based approach to the classification of Yami emotion 541
Table 5: Distribution of 126 Yami emotion terms (ika- verbs)
Internal conditions
Pure affective
Affective- behavioral
Affective- cognitive
Pure cognitive
Cognitive- behavioral
Pure physical
N=126 17 16 7 56 6 24 100% 13% 13% 6% 44% 5% 19%
pure physical 19%
pure affective 13%
pure cognitive 44%
Figure 2: Classification of Yami Emotion
4.2 Positive and negative Yami emotion
As ‘good’ and ‘bad’ are part of the universal primitives in Wierzbicka’s (1992) definition of emotion, we assume that emotion can be coded as a continuum of positive and negative feelings. Based on the Chinese translations of these emotion terms, we assigned all the Yami emotions to five levels: positive, relatively positive, neutral, relatively negative, and negative (see appendix). Relatively positive and relatively negative evaluations are determined in relation to prototypes of the two ends. For example, ikasaray ‘happy’ is taken as prototypically positive while ika’oya ‘angry’ prototypically negative. Relatively negative and relatively positive evaluations describe the positive and negative emotions with lesser degrees in comparison with the two extremes. For example, ikanig ‘embarrassed, ashamed’ is classified as relatively negative in relation to the negative end of ‘angry’. Similarly, ikabsoy ‘satiated’ is classified as relatively positive in relation to the positive end of ‘happy’.
Some emotion terms may not be “valenced” or evaluated as positive or negative. For example, ‘surprise’ and ‘amazement’ do not imply anything good or bad (Wierzbicka
Emotion in Yami
542 Victoria Rau, Yi-Hsin Wu and Meng-Chien Yang
1992:550); therefore, it is necessary to have a neutral category when it comes to evaluation of emotion, such as ikakaha ‘sleepy’.
The results indicate that Yami has more negative emotions, as shown in Figure 3. The distribution of Yami emotion is negative 47% (23% negative, e.g., ika’oya ‘angry’; 24% relatively negative, e.g., ikanig ‘embarrassed, ashamed’), neutral 30% (e.g., ikakaha ‘sleepy’), and positive 23% (12% positive, e.g., ikasaray ‘happy’; 11% relatively positive, e.g., ikabsoy ‘satiated’).
This assignment of postive and negative emotion based on Chinese–English translation remains tentative, and requires further corroboration with native Yami speakers. In particular, the neutral category and the fine-grained classifications of the relatively positive and relatively negative evaluations of Yami emotions may require a future field study to elicit the ‘emic’ judgments from the Yami speech community to validate our classification. However, the generalisation that Yami has more prototypically negative (23%) than positive type of emotion probably still holds, given the low percentage of positive emotion (12%).
Positive and Negative Tendency of Yami Emotions
Relatively Negative
23%
Figure 3: Distribution of positive and negative emotion terms in Yami
4.3 Semantic categories of Yami emotion
The same set of emotion data was further categorised based on near synomyms. Table 6 lists all the semantic categories of Yami emotion, divided into Clore et al.’s three categories: (i) affective conditions, (ii) cognitive conditions, and (iii) physical and bodily states.
A corpus-based approach to the classification of Yami emotion 543
Table 6: Semantic categories of Yami emotion
Clore et al.’s categories Semantic categories of Yami emotion Affective conditions Anger, fear, happiness, longingness, love, nervousness,
sorrow, sympathy, worry, evaluation of size, curse, forgetfulness, jealousy, noise/calmness, shame, stinginess, danger, overbearingness, relaxation
Cognitive conditions Age, evaluation of size, boredness, certainty, cleverness, cold/heat, curiosity, danger/safety, darkness, difficulty, relaxation, external states of things, distance, fortune, goodness/badness, greatness, importance, independence, weight, noise/calmness, quantity, correctness, smoothness, states of human body, taboo, taste/smell, watchmacallit, diligence/laziness
Physical and bodily states Disagreement, nitpicking, scheme
The same table can be further represented by the diagram drawn using the Protégé program to represent the Yami emotion ontology, as shown in Figure 4. The diagram also helps us visualise the overrepresentation of cognitive conditions in encoding Yami emotion. In addition, several semantic categories show cross-sectional representations, the most important of which is ‘fear’, occurring not only in pure affective and affective-cognitive conditions, but also in physical and bodily states.
544 Victoria Rau, Yi-Hsin Wu and Meng-Chien Yang
Figure 4: Ontology describing categories of near synonyms
A corpus-based approach to the classification of Yami emotion 545
4.4 Comparison of Yami emotions with Filipino emotions
Our hierarchical cluster analysis of Yami emotion terms yielded eleven distinctive domains (see Figure 5), i.e., happy, aroused, contented, emotionless, contemptuous, angry, sad, tired, quiet/shy, anxious, and aspiring, almost identical to the Filipino emotion terms, except that the guilty domain is lacking in Yami. However, this does not mean Yami does not have the guilty domain. The Yami word for guilty miraraten (mi-raraten) is derived from the stem raraten ‘guilt, sin’ (< rahet ‘bad’). As the ika- prefix would derive ikarahet ‘consider something bad because…’, instead of guilty, this explains why the guilty domain is missing from the cluster analysis.
How well these domains are clustered can be evaluated by the weight. Figure 5 depicts the Yami emotion clusters derived from the ontological calculation. The dendrograms of this figure show the selected clusters in each domain. The scale shown is a simulation calculation derived by the weight function of semantic distance between each cluster in the emotional ontology. This simulation is used to emulate the judgment process in Church et al.’s study. If the weight of the lexical word is close to 0.9, this indicates the word fits well in its semantic domain. On the other hand, if the weight is close to 0.1, the word does not fit the domain well.
The domains in Figure 5 form a hierarchy of fitness: happy (0.8), angry (0.8), anxious (0.7), aroused (0.6), sad (0.6), tired (0.6), emotionless (0.5), contented (0.5), contemptuous (0.4), quiet/shy (0.3), and aspiring (0.3). Thus, the cluster analysis provides further confirmation of our identification of the eight most important (or hypercognised) emotion categories and the three least important (or hypocognised) emotion categories in Yami, if we use 0.5 as an arbitrary threshold.
Although cluster analysis is a useful quantitative tool to yield preliminary results in our analysis, we cannot avoid cross-sectional reprentations. Some emotion terms, albeit a negligible minority, were categorised into two different domains. For example, ikazoay ‘feel proud’ was put in both “emotionless” and “aspiring” domains. This points out that assignment of lexical items to different semantic categories by the third author also awaits future validation by the speech community members.
In summary, our cluster analysis of Yami emotion identified all eleven emotion domains found by Church et al. (1998), except the guilty domain. The reason, as explained previously, is that ika-rahet ‘consider something bad because …’ only encodes the emotion of anxiety/worry, whereas guilty is encoded by mi-raraten. The most important emotions in Yami are anxiety/fear, arousal, contentment, anger, happiness, sadness, and “emotionless” (i.e., feeling bored and other cognitive conditions) whereas the least important emotion domains include feeling tired, quiet/shy, contemptuous, and aspiring. Our findings also confirm that the terms in all three subcategories of the affection conditions class in Clore et al.’s (1987) taxonomy of emotion terms are viewed as referring to emotions.
546 Victoria Rau, Yi-Hsin Wu and Meng-Chien Yang
A corpus-based approach to the classification of Yami emotion 547
Figure 5: Hierarchical cluster analysis of Yami ika- emotion terms
548 Victoria Rau, Yi-Hsin Wu and Meng-Chien Yang
4.5 Evaluation of the current methods
Before we conclude, an evaluation of the corpus linguistics–cum–ontology methods in the study of Yami emotion is in order. The strength of this interdisciplinary approach to investigate Yami emotion lies in the identification of a grammatical model, i.e., Yami prefix ika- ‘the reason/cause for a certain feeling’, as a keyword for systematic search in the online corpus. A detailed analysis of Yami narratives in step 2 resulted in further data on emotion that overlapped with the ika- words to yield the final 126 emotion terms in our analysis.
The usefulness of the computerised representations of the Yami ontology as shown in Figures 4 and 5 in computational linguistics depends solely on the accuracy of the corpus linguistic analysis of the Yami data. This prompted us to further validate our classification in future fieldwork. However, a word of caution is necessary here. Although the assignment of the emotion terms into different categories was determined by the first two authors solely on the basis of clear operational definitions, this was probably the best approach to our study given the circumstances. An attempt was made for the first author to bring the list of emotion terms (in the appendix) to Orchid Island for the community members to either confirm or reject the validity of our classification. However, this task incited a debate on what emotion is, and no consensus was reached. Nonetheless, the preliminary results we found in the study have paved the way for two immediate follow-up studies: (1) an evaluation of positive and negative Yami emotion by community members using a five-point Likert scale and (2) a classification of emotion terms into the eleven domains in the cluster analysis by community members.
5 Conclusion This study has classified Yami emotion into six internal conditions. Like Filipino, the
most important emotions in Yami are anxiety/fear, arousal, contentment, anger, happiness, sadness, and emotionless (e.g., feeling bored), whereas the least important emotion domains include feeling tired, quiet/shy, contemptuous, and aspiring. In general, Yami emotions basically share the same distinct domains as Filipino emotions and similarly the language contains more negative emotion terms than positive ones.
This study, albeit preliminary, has demonstrated how to use an interdisciplinary approach to study Yami ontology, using emotion as a semantic domain. Future investigation can apply the same methods to cover wider semantic domains in preparation for building a complete ontology of a targeted language and culture.
A corpus-based approach to the classification of Yami emotion 549
Appendix
Classification of the 126 Yami emotion terms
Positive Yami English Category (Church et al. 1988) ikagága ‘happy’ pure ikáglaw ‘love’ pure ikaizay ‘great’ cognitive ikakey ‘like (something)’ pure ikákza ‘like (someone, something)’ pure ikamiying, ikami'ing
‘laugh’ affective-behavior
Relatively Positive Yami English Category (Church et al. 1988) ikabsoy ‘satiated’ physical ikacigzang ‘strong, hard’ physical ikahanang ‘calm’ affective-cognitive ikakdeng ‘important’ cognitive ikalamnay ‘relaxed’ cognitive ikamoay ‘plump’ cognitive ikaoyat ‘diligent’ cognitive-behavior ikapzat ‘safe’ cognitive ikaraevaes ‘fitting’ cognitive ikasazovaz ‘relaxing’ affective-behavior ikasingat ‘important/expensive’ cognitive ikasonong ‘smooth’ cognitive ikawadwad ‘clear, clean’ cognitive ikeylamnay14 ‘relaxed’ affective-cognitive
14 It may be a variant of ikalamnay ‘relaxed’.
550 Victoria Rau, Yi-Hsin Wu and Meng-Chien Yang
Neutral Yami English Category (Church et al. 1988) ika’amang ‘fainted’ physical ikabezbez ‘hurried’ physical ikaganinam ‘sweet’ cognitive ikahamang ‘forget’ affective-behavior ikahangno ‘smells fragrant’ cognitive ikahep ‘dark’ cognitive ikahithitkahen ‘love to sleep’ physical ikakaha ‘sleepy’ physical ikakoan, ikaikikoan ‘watchmacallit’ cognitive ikakoat ‘burning heat’ cognitive ikakopad ‘bitter’ cognitive ikaladan ‘older’ cognitive ikalak ‘curious’ cognitive ikalavayo ‘young’ cognitive ikanehed ‘true/certain’ cognitive ikangilin ‘lucky’ cognitive ikangongyod ‘real, certain’ cognitive ikangot ‘smelly, stinky’ cognitive ikangsah ‘bored’ cognitive ikanoyong ‘real’ cognitive ikaotok ‘bored’ cognitive ikapait ‘salty’ cognitive ikapaw ‘miss (someone, something)’ pure ikarehmet ‘heavy’ cognitive ikarekmeh ‘cold (weather)’ cognitive ikasagpaw ‘heavy’ cognitive ikasinasina ‘divergent’ cognitive-behavior ikasngen ‘too close’ cognitive ikasoliket ‘sticky’ cognitive ikateleh ‘deaf’ cognitive ikavaheng ‘black’ cognitive ikavaw ‘cool’ cognitive ikavokay ‘dry (powder)’ cognitive ikaynaw/ ikeynaw ‘stinky, fishy’ cognitive ikazemek ‘broken’ cognitive ikazeziak ‘all speak loudly’ affective-behavior ikehma ‘soft’ cognitive ikeyngen ‘muscle ache’ physical
A corpus-based approach to the classification of Yami emotion 551
Relatively Negative Yami English Category (Church et al. 1988) ikaa ‘gluttonous’ physical ikaciciaw ‘too talkative’ cognitive ikacilo ‘noisy’ cognitive ikagolang ‘thin’ physical ikagom ‘overbearing’ affective-cognitive ikahango ‘seasick’ physical ikahen ‘very cold’ cognitive ikakaram ‘as small as a mouse’ cognitive ikakcin ‘hungry’ physical ikalanan ‘gluttonous’ physical ikalikey ‘too small’ affective-behavior ikalinlin ‘faint, too weak to stand’ physical ikalita ‘scheme’ cognitive-behavior ikalotoy ‘bulging stomach’ physical ikamez ‘chilly’ cognitive ikamo ‘embarrassed’ affective-behavior ikanig ‘ashamed, embarrassed’ affective-behavior ikapereh ‘few’ cognitive ikapili ‘picky’ cognitive-behavior ikaraway ‘become ugly’ cognitive-behavior ikasaki ‘drunk’ physical ikasnek ‘ashamed’ affective-behavior ikasngisngit ‘sharp pain’ physical ikaspet ‘dangerous or complicated’ cognitive ikatahaw ‘weak’ physical ikateyci ‘feel disgusted’ physical ikayod ‘tired’ physical ikazazomay ‘sick of eating something’ physical ikaziknan ‘exhausted’ physical ikazonat ‘inflame’ physical
Negative Yami English Category (Church et al. 1988) ikaciblis ‘cursed’ affective-behavior ikaciwciw ‘scared away’ affective-behavior ikakaniaw ‘taboo’ cognitive ikakezes ‘nervous’ pure ikalag ‘taboo’ cognitive ikalas ‘wrong’ cognitive ikálaw ‘worried’ pure ikalma ‘lazy’ cognitive-behavior ikaloit ‘dirty’ cognitive ikamáma ‘worried’ pure
552 Victoria Rau, Yi-Hsin Wu and Meng-Chien Yang
Negative Yami English Category (Church et al. 1988) ikameneng ‘heartbroken’ pure ikamogaw ‘afraid’ pure ikananat ‘scared, as if the stomach is lifted’ physical ikangazicin ‘disgusted’ physical ikanginanawa ‘dangerous’ affective-cognitive ikaniahey ‘afraid’ pure ika'ogto ‘frightened’ pure ika'oya ‘angry’ pure ikarahet ‘bad/feel upset’ cognitive ikararaten ‘stingy’ affective-behavior ikarokaw ‘lonely, sorrowful’ affective-cognitive ikasalit ‘difficult’ cognitive ikasozi ‘angry (violent)’ pure ikatamoad ‘embarrassing’ affective-behavior ikavozoaw ‘scared away’ affective-behavior ikaynanahet ‘jealous’ affective-behavior ikaywam/ ikeywam ‘afraid’ pure ikeynanahet ‘selfish/jealous’ affective-cognitive ikeyzaw ‘frightened’ affective-cognitive
A corpus-based approach to the classification of Yami emotion 553
References
Church, A. Timothy, Marcia S. Katigbak, Jose Alberto S. Reyes and Stacia M. Jensen. 1998. Language and organisation of Filipino emotion concepts: Comparing emotion concepts and dimensions across cultures. Cognition and Emotion 12.1:63–92.
Clore, Gerald L., Andrew Ortony and Mark A. Foss. 1987. The psychological foundations of the affective lexicon. Journal of Personality and Social Psychology 53.4:751–766.
Dirven, René. 1997. Emotions as cause and the cause of emotions. In The language of emotions: Conceptualization, expression, and theoretical foundation, ed. by Susanne Niemeier and René Dirven, 55–83. Amsterdam and Philadelphia: John Benjamins.
Dong, Maa-Neu, Victoria D. Rau and Ann Hui-Huan Chang. (eds.) 2012. dawu yu cidian [Yami (Tao) dictionary]. Taipei: National Taiwan University Press.
Goldberg, Adele E. 1995. Constructions: A construction grammar approach to argument structure. Cognitive Theory of Language and Culture Series. Chicago: University of Chicago Press.
Hsieh, Fuhui. 2011. The conceptualization of emotion events in three Formosan languages: Kavalan, Paiwan, and Saisiyat. Oceanic Linguistics 50.1:65–92.
Hu, Jackson and Victoria D. Rau. 2013. dawu yuming de shengtai tixian han renzhi yuyi [Ecological embodiment and cognitive semantics of Tao fish names] In Analyzing language and discourse as intercultural and intracultural mediation, ed. by Ming-Yu Tseng, 187–227. Kaohsiung: National Sun Yat-sen University. [In Chinese]
Huang, Shuanfan. 2002. Tsou is different: A cognitive perspective on language, emotion, and body. Cognitive Linguistics 13.2:167–186.
Lakoff, George and Mark Johnson. 1999. Philosophy in the flesh: The embodied mind and its challenge to Western thought. New York: Basic Books.
Levy, Robert I. 1984. The emotions in comparative perspective. In Approaches to emotion, ed. by Klaus R. Scherer & Paul Ekman, 397–412. Hillsdale, NJ: Lawrence Erlbaum Associates.
Rau, D. Victoria and Maa-Neu Dong. 2010. Serves you right: Ka- as an attitudinal marker in Yami. Paper presented at the 4th Conference on Language, Discourse, and Cognition. 1-2 May 2010. Taipei: National Taiwan University.
Rau, D. Victoria and Maa-Neu Dong. Forthcoming. dawu yu yufa cankao shu [Yami reference grammar] [In Chinese]
Rau, D. Victoria and Maa-Neu Dong. 2006. Yami texts with reference grammar and dictionary. Language and Linguistics Monograph Series A-10. Taipei: Academia Sinica.
Rau, D. Victoria, Yi-Hsin Wu and Meng-Chien Yang. 2012. What can ika- tell us about Yami emotion? Paper presented at the 2012 International Conference on
554 Victoria Rau, Yi-Hsin Wu and Meng-Chien Yang
Landscape, Seascape, and the Spatial Imagination. 2–3 November, Kaohsiung: National Sun Yat-sen University.
Rau, D. Victoria, Yi-Hsin Wu, Meng-Chien Yang and Jackson Hu. 2012. What can ika- tell us about Yami emotion? Poster presented at the 2012 CLDC, 4–6 May. Taipei: National Taiwan University.
Rau, D. Victoria, Meng-Chien Yang, Hui-Huan Ann Chang and Maa-Neu Dong. 2009. Online dictionary and ontology building for Austronesian languages in Taiwan. Journal of Language Documentation and Conservation 3.2:192–212. http://nflrc.hawaii.edu/ldc/.
Saucier, Gerard and Lewis R. Goldberg. 1996. The language of personality: Lexical perspective on the five-factor model. In The five-factor model of personality: Theoretical perspectives, ed. by Jerry S. Wiggins, 21–50. New York: Guilford Press.
Tai, Yin-Sheng, D. Victoria Rau and Meng-Chien Yang. 2008. A semantic study on Yami ontology in traditional songs. IJCNLP 2008. In Proceedings of the 6th Workshop on Asian Language Resources, 81–84, 11–12 January. Hyderabad, India. http://www.aclweb.org/anthology/I/I08/I08-7011.pdf.
Talmy, Leonard. 2000. Toward a cognitive semantics. Cambridge, MA: MIT Press. Yang, Meng-Chien, D. Victoria Rau and Yi-Hsin Wu. 2012. Analyzing and classifying
the Yami emotion phrases using ontological structure and computation. Paper presented at the International Conference on Asian Language Processing (IALP), 13–15 November. Hanoi, Vietnam.
Wierzbicka, Anna. 1992. Defining emotion concepts. Cognitive Science 16.4:539–581. Wolff, John U., with Maria Theresa C. Centeno and Der-Hwa V. Rau. 1991. Pilipino
through self-instruction. Ithaca, NY: Cornell University Press. Zeitoun, Elizabeth and Lillian M. Huang. 2000. Concerning ka-, an overlooked marker of
verbal derivation in Formosan languages. Oceanic Linguistics 39.2:391–414.