Journal of Linguistics and Language Teaching (JLLT)

Journal of Linguistics and Language Teaching

Volume 10 (2019) Issue 1, pp. 87-118

Language Maintenance and Language Shift

in a Taiwanese Aboriginal Village:

A Domain Approach

Ya-Ling Chang (Yilan, Taiwan (R.O.C.))

Abstract

This study explores language shift by scrutinising language proficiency and habitual language use in one of the biggest aboriginal Pangch villages in Taiwan. It aims to better understand the factors attributed to the erosion of the Pangcah language and how language policy shaped its linguistic structures. In a limited perspective, data were drawn from a survey of language proficiency and language use in various settings. The domain analysis model by Fishman (1964, 1965, 1972), incorporating Multiple Analysis of Variance (MANOVA) and Logistic Regression (LR) of SPSS was adopted. The results of the sociolinguistic survey show that age is a variable which is related to language change over time. Education has its impact on language proficiency in Mandarin (Mandarin Chinese). In language use, there is a general shift towards the dominant language(s), mainly Mandarin. The language choice among children, in the family domain in particular, appears to be shifting towards Mandarin Chinese monolingualism. This shows some evidence that the heritage language is not transmitted to future generations. Language use in in-group communication in other domains as friendship, religion and shop shows male Pangcah as a better language keepers than females However, in most of the domains, whether high or low, where Han Chinese are present, there is a general shift to Mandarin, which shows little resistance towards the domaince of Mandarin.. Thus, reflection on some of these findings shows that the Pancah language shift patterns are closely related to the long-term colonial hegemonic language policies and hierarchy which are integral to the sociopolitical economy of the district.

Key words: Language maintenance, language shift, Pangcah, language use, language proficiency, domain approach

1 Introduction

Before the occupation of the Dutch and Chinese from the 17th century onwards, Taiwan was an unclaimed island inhabited by various groups of indigenous people - the earliest settlers on the island. Since then, Taiwanese indigenous people have been ruled by six different regimes: the Dutch, the Spanish, the Ming Dynasty, the Qing Dynasty, the Japanese and the Republic of China, the Taiwanese Government (The Chinese Nationalist Party and the Democratic Progress Party). While these historical events led to abundant ethnic and linguistic resources, they simultaneously allowed ethnic and linguistic conflicts to arise.

Ethnicity is a highly politicised issue in Taiwan, often involving the tension between inclusion and exclusion under the definition of ‘Taiwanese’. After the lifting of martial law in 1987, a more common division of ethnic boundaries was formed under the term ‘Four Major Ethnic Groups’, comprising Southern Min, Hakka, Mainlanders, and Austronesian indigenous peoples. As language groups and linguistic affiliations have not been included in the census, the distribution outlined by Huang (1995: 21) can be summarised in the following: Mainlanders (Han Chinese) 13%, Southern Min (Han Chinese) 73.3%, Hakka (Han Chinese 12%) and Austronesian aborigines 1.7%. Ethnic groups in Taiwan are basically divided along linguistic lines.

The language situation in Taiwan as a whole is complex. Two overarching language families are the Sino-Tibetan and the Austronesian languages. Sino-Tibetan includes Mandarin Chinese(hereafter “Mandarin“), Southern Min (a general term, with the language also called ‘Taiwanese’ or ‘Southern Fukienese’), Hakka and various Chinese languages / varieties spoken by the Mainlanders. The Austronesian languages are described as a collection of mutually unintelligible tongues spoken by the 16 officially recognised aboriginal peoples. Due to the success of the National Language Movement after the 1950s, which promoted Mandarin as a means of communication, it is fair to say that Mandarin has replaced Southern Min and Hakka as the most commonly used language. It is generally believed that most people are bilingual and are able to speak Mandarin and at least one other language. Many aboriginal people are trilingual, speaking Southern Min, one of the aboriginal languages or Hakka, and either Mandarin or Japanese, which elderly Pangcah people acquired during the Japanese colonial period (1895-1945).

The aboriginal languages have become in-group languages, confined to private domains because they are demographically small. Those who speak them are also economically and politically marginalised, compared to the majority of Han Chinese in Taiwan. This paper therefore presents a case study of Pangcah language maintenance and shift by examining language proficiency and use so as to better understand the way in which other languages compete with and impact on Pangcah.

Fishman (1964) notes that there are

three key aspects which an ‘informed evaluation’ of language maintenance and language shift (LMLS) should consider: habitual language use, behaviour towards language, and socio-cultural change process. (cited by Li 1997: 365).

In a limited scope, one of these aspects will be examined – habitual language use in Tafalong, a rural, compact and linguistically beleaguered aboriginal Pangcah village, which comprises 4197 people of plain aboriginese (including a very small number of non-Pangch aboriginal) (Household Office, Kuang-fu Township of Hua-lien County, 2018). The issues related to language shift and maintenance are sizeable, and there is no single simple model to examine this phenomenon. Therefore, rather than attempting to exhaust possible models of analysis or factors attributed to the erosion of the aboriginal language, this paper takes a limited perspective to seek an answer to the research question regarding the extent to which language proficiency and language use in Tafalong have declined. Language use and proficiency are examined though survey data, using Fishman's (1965, 1972) domain approach .

2 Models of Language Maintenance and Language Shift

“The study of LM [language maintenance] and LS [language shift] is a multidisciplinary and interdisciplinary enterprise” (Pauwels, 2004: 721). Researchers from different academic disciplines work on language maintenance and language shift. Earlier divisions looking into the methodology of language maintenance and shift were based on academic disciplines. Garcia (2003) and Fasold (1984) make a distinction between two broad fields: anthropology / anthropological linguistics and sociology / the sociology of language. According to these two researchers, participant observation is often applied by (linguistic) anthropologists when investigating language choice in individual communities; sociologists, in contrast, adopt surveys or questionnaires to generalise a wider, generalised view of language shift and language maintenance in various communities. Thus, these two academic disciplines create different interpretations based on the approaches which they have adopted: the former uses smaller-scale qualitative data; the latter, large-scale quantitative data.

In order to obtain a comprehensible picture of social changes contributing to language maintenance and shift, which is different from the divisions of academic associations as described above, researchers have tried to find ways to bridge both micro and macro analyses of languages in different ways. Based on levels of social structure, Milroy (2001: 39-40) categorises the following research paradigms of analysis: “political and economic determinants of language behaviour, community level patterns of language choice / language mixing, and language behaviours at interactional level”. Analysis at macro-level, as observed by Milroy (2001), aims to distinguish the social, political and economic factors contributing to language choice behaviours by individuals in various domains or institutions.

Another line of research is community-level ethnographic analysis, which looks for the connections between individual language choice and both mundane discourse practice and social organisations. It is likely that local level analysis can be linked to the macro-level structure. This suggests that such analysis can bridge the gap between the micro and macro level of language planning.

The third approach deals not only with micro analysis within the bilingual conversational interaction but also the linkage of social values attributed to it. Each of the three categories of research models recognised by Milroy (2001) by no means has a clear distinction between their micro or macro level analysis; that is to say, whilst studies might aim at a micro-level analysis, the research aim seems to be to build bridges between both levels of analysis in order to achieve a better understanding of language maintenance and language shift.

After having given a general overview on the research paradigms for language maintenance and language shift, we will now turn to theoretical studies on the specific approach of domain approach adopted in this study, as it provides grounds for later discussion and analysis.

2.1 Domain Approach

On the basis of Fishman’s theories (1964, 1965, 1972), domain analysis has been widely used as an approach to explore language shift and language maintenance. This is due to the fact that such an approach enables researchers to link the language investigated with its context, which is integrated with participants, settings and topics. This approach provides a means for the systemic analysis of social and cultural change related to language maintenance and shift. Nevertheless, this domain approach is not without problems. Dittmar (1976) indicates that the compartmentalised nature of the domain does not necessarily reflect the language in real social contexts. Apart from this, Baker (2006) also suggests that the amount of quality time (i.e. the time spent in real conversation) should also be taken into account since it is also an important element of language use. Data for the domain-analysis approach are often limited to self-reported language use survey data. Totally self-reported survey data are seen to have limited validation despite Fasold’s (1984) claims that such data can be valuable under careful investigation. Thus, the result of survey data is can be more suggestive than confirmatory of the language use of the target informants.

3 Research Design

3.1 Goals for Data Collection and Description of Data

The main purpose of this quantitative study is to better understand the language use of the Pangcah people, in the hope that it will provide a blueprint for further research on language maintenance and language planning. Also, by scrutinising language use, it is hoped that comprehensive insights into the correlation between language acquisition, language proficiency and language use can be obtained.

The present study is based on a small-scale survey comprising one hundred and four questionnaires (Appendix 1) which centre on a small collective, i.e. an aboriginal Pangcah village. All the structured questionnaires were conducted individually by the researcher.

3.2 Sampling Technique and Research Instruments

IIn the present study, convenience sampling was adopted due to the fact that most young people aged twenty to forty were not available due to the fact that they were seeking jobs in the cities. The questionnaire included closed-ended questions. Additional open ended questions were only used when the respondent’s information about the topic might differ from someone else’s. The questionnaire comprised three sections, namely, Language Proficiency, Language Choice for use with different interlocutors in daily life and Personal Information.

The first section contained a self-assessment of language proficiency level to gain ground for further understanding of the relationship between language ability and language use. In the second section, the questions sought to investigate the respondents’ language use with different interlocutors, these being limited mostly to the people whom the respondents were likely to encounter on an everyday basis. While not exhaustive, this section covers a wide range of possible situations. The last section was related to background information covering the respondents’ gender, age and education (Appendix 1).

3.3 The Field

Tafalong is a multilingual and multiethnic farming community, consisting of Pangcah, Southern Min and Hakka people, who constitute one third of the population each. In addition, there are a few people from ethnic minorities, including aboriginal Bunun, aboriginal Atayal and foreigners mostly from South-east Asia, becoming inhabitants through intermarriage.

Young speakers of Pangcah aged under 30 are mostly semi-speakers of Pangcah, who are capable of comprehending a certain amount of Pangcah but cannot speak it. Their situation is more or less similar to that of the five to ten children in the Gapun community of Papua New Guinea, who possessed a ‘good passive understanding’ of the local language, as described by Kulick (1992: 217). Although one might be surprised to find one or two children who speak fluent Pangcah as their native language, it emerges that they were generally raised in a family in which the grandparents do not speak Mandarin or possess very low Mandarin language ability. Thus, Mandarin is the only language of most young Pangcah people. People in Tafalong are aware of language shift, for they face everyday-communication difficulties or language breakdowns with their children or grandchildren. However, the awareness of language death is not very distinct, due to the fact that Tafalong is a big Pangcah tribal village and it appears to most people that the language is still alive. Many Pangcah have learnt to speak Southern Min through different channels, such as the media, neighbours or work. They do, however, have less knowledge about Hakka in spite of the fact that Hakka in Tafalong also accounts for one third of the population.

3.4 The Participants

The participants are mostly the school pupils’ relatives; some are people with whom the researcher was in touch on a regular basis. Working at the local elementary school, the researcher often asked pupils to request permission from their relatives to go to their houses to conduct the present survey. Permission was hardly ever refused.

3.5 Methodology

In this study, language shift and language maintenance are assessed through language use in self-reported data in order to draw a general picture of language choice behaviour in everyday life, i.e., the extent of language shift. Language proficiency and language use are investigated by means of a sociolinguistic survey. Language use in particular (Fishman 1964, 1965, 1972) is examined through a domain approach combined with mainly closed-ended questions requesting self-reported language use administered to 104 targeted villagers (mostly engaged in farming) of different age groups, genders and education levels.

3.6 Piloting and Revision of the Questionnaire

The preliminary work of the research design was to identify the problem of the endangered Taiwanese aboriginal language, Pangch. The hypothesis under scrutiny was that Pangch language use has decreased significantly among the Pangcah people. Before designing the questionnaire, the sample-and-survey method had to be decided upon. Based on the subsidiary questions, the questionnaire was then carried out. Following the design of the questionnaire, a pilot test was conducted, and at least ten villagers were asked to fill in the questionnaire. Then the results of the pilot study were analysed and the questionnaire was modified accordingly to increase its reliability and validity. After the informants had filled in the questionnaires, these were collected and examined to check for any relevant answers which might be missing. In such cases, the respondents were asked to provide the respective answers.

3.7 Coding, Data Processing and Analysis

Since the questions in the survey were mainly closed-ended, except for personal information, the response alternatives and the question grids were designed so as to fit in the computer coding system so as to facilitate the coding of the data. The data were analysed using SPSS.

4 Data Analysis

The following data analysis comprises two parts: the analysis of language proficiency and the analysis of language use.

4.1 The Analysis of Language Proficiency

As mentioned above, the language used to be analysed is based on the valid data from 104 respondents. The following tables demonstrate the numbers of respondents interviewed in groups according to age, gender and education:

Variables	Categories	Number
Gender	Male	48
	Female	56
Age	21-30	13
	31-40	22
	41-50	22
	51-60	17
	61-70	17
	Over 70	13
Education	Up to Primary School (aged 6-11)	48
	Up to Junior High School (aged 12-14)	20
	Up to Senior High School (aged 15-17)	26
	Up to College Level (including five-year junior college, for those aged over 15)	1
Table 1: Community Member Interview Profile 1

Language use and language proficiency are closely related and are both indicators of language maintenance and shift. When a language shift is taking place, language use and language proficiency both decline. Before we proceed to look at the language use of the subjects with different interlocutors, a preliminary examination of language proficiency is first provided to make it easier to interpret the language use. Multiple Analysis of Variance (henceforth MANOVA) is applied to the investigation of language proficiency and Logistic Regression (henceforth LR) is used for language use.

The reasons for adapting the two statistical techniques MANOVA and LR in investigating language proficiency and language use are as follows: MANOVA is a statistical tool for testing the significance of the relationship between two or more dependent variables. In this case, there are four dependent variables, namely the language proficiency of Pangcah, Japanese, Mandarin and Southern Min. The power of MANOVA enables us to investigate the effects of several independent variables, as well as to examine the effect of single treatment. Since, in this case, there are three independent variables (age, gender and education), it can be useful to study not only the relationship between language proficiency and each individual variable (the main effect) but also its relationship with the interaction of variables (age and education, education and gender, age and gender, and age, gender and education). Examining such relationships (the increase and decrease of language proficiency in their relation to age, gender and education) enables us to locate the effect of language policy on the language proficiency of the investigated languages (Pangcah, Japanese, Mandarin and Southern Min). MANOVA is applied at a significance level of 0.05.

Logistic Regression (LR) deals with situations where there are fewer than three groups of dependent variables. In this area of language use, the dependent variable refers to the respondent’s self-reported language choices with different interlocutors. The outcome variables are two combinations of language choice, namely Pangcah as the most frequently used language, and the dominant language(s) as the most frequently used language(s) which are both categorical. Thus, LR is employed.

4.1.1 The Analysis of Language Proficiency

In the present study, language proficiency focuses on the respondent’s self-assessment of their language ability based on a five-point scale from ‘not able to produce and to comprehend sensible conversation, sentences and vocabulary’ to ‘very fluent’. The languages reported include the major languages used in the field, namely Pangcah, Japanese, Mandarin and Southern Min.

The languages which the informants acquired either in their neighbourhood or during their work or study, such as Hakka, Indonesian, Toruko and English, were also reported. However, they are not included because the cases reported account for a small number of speakers only, and they are less frequently used within in-group communication.

There are three demographic variables selected as the independent variables, namely cage, gender and education. Cage is the categorised version of age which specifies six age groups. Under MANOVA, tests of different models for the relationship between the main effects (cage, gender and education), two-way interactions and the three-way interaction of the language proficiency of Pangcah, Japanese, Mandarin and Southern Min, have been performed to obtain a result.

Before we go further into analysing the results, it may be helpful to understand the process of analysis in conjunction with Table 2 (Multivariate Tests), Table 3 (Tests of Between-Subjects Effects), and Table 4 (Parameter Estimates). The functions for presenting the results of each within MANOVA are as follows: the multivariate tests display the overall relationship between the independent and the dependent variables of the respective proficiency of the four languages as a whole; the tests of Between-Subjects Effects provide the significance level of the effect of the independent variables on the language proficiency in each individual language. Parameter Estimates provide detailed information about the way in which the respective proficiency of each language is affected by age, gender and education in terms of their categorisations.

Within MANOVA, the best model is finalised under various stages of entering and dropping-out factors and their interactions. Table 2 below presents the Multivariate Tests of the model. As shown in the table, this model includes only the main effects since the interaction terms for independent variables appear to be insignificant.

Effect		Value	F	Hypothesis df	Error df	Sig.
Intercept	Pillai's Trace	.984	1397.059(a)	4.000	91.000	.000
	Wilks' Lambda	.016	1397.059(a)	4.000	91.000	.000
	Hotelling's Trace	61.409	1397.059(a)	4.000	91.000	.000
	Roy's Largest Root	61.409	1397.059(a)	4.000	91.000	.000
gen	Pillai's Trace	.094	2.374(a)	4.000	91.000	.058
	Wilks' Lambda	.906	2.374(a)	4.000	91.000	.058
	Hotelling's Trace	.104	2.374(a)	4.000	91.000	.058
	Roy's Largest Root	.104	2.374(a)	4.000	91.000	.058
cage	Pillai's Trace	1.082	6.971	20.000	376.000	.000
	Wilks' Lambda	.149	11.733	20.000	302.763	.000
	Hotelling's Trace	4.241	18.980	20.000	358.000	.000
	Roy's Largest Root	3.892	73.164(b)	5.000	94.000	.000
edu	Pillai's Trace	.196	1.622	12.000	279.000	.085
	Wilks' Lambda	.810	1.667	12.000	241.055	.075
	Hotelling's Trace	.228	1.703	12.000	269.000	.066
	Roy's Largest Root	.193	4.498(b)	4.000	93.000	.002
a Exact statistic b The statistic is an upper bound on F that yields a lower bound on the significance level. c Design: Intercept+gen+cage+edu
Table 2: Multivariate Tests (c)

From Table 2, it can be inferred that the three independent variables are all significant for language proficiency. However, they are significant in different degrees. Cage has a strong overall significance for overall language proficiency (P<0 .0005="" font="">gender has a borderline significance (P=0.058), and education has a minor significance (P<0 .1="" font="">

In order to understand the effect of the three independent variables - gender (gen), education (edu) and age (cage) - for each individual language, we move on to Table 3-Tests of Between-Subjects Effects:

Source	Dependent Variable	Type III Sum of Squares	df	Mean Square	F	Sig.
Corrected Model	LangP	52.654(a)	9	5.850	16.612	.000
	LangJ	191.847(b)	9	21.316	20.138	.000
	LangM	44.404(c)	9	4.934	17.092	.000
	LangS	66.462(d)	9	7.385	7.612	.000
Intercept	LangP	1051.667	1	1051.667	2986.102	.000
	LangJ	253.925	1	253.925	239.890	.000
	LangM	927.268	1	927.268	3212.240	.000
	LangS	350.170	1	350.170	360.953	.000
gen	LangP	1.077	1	1.077	3.058	.084
	LangJ	.010	1	.010	.010	.922
	LangM	.305	1	.305	1.057	.307
	LangS	6.103	1	6.103	6.291	.014
cage	LangP	16.600	5	3.320	9.427	.000
	LangJ	139.676	5	27.935	26.391	.000
	LangM	20.103	5	4.021	13.928	.000
	LangS	36.467	5	7.293	7.518	.000
edu	LangP	1.476	3	.492	1.397	.249
	LangJ	3.255	3	1.085	1.025	.385
	LangM	3.363	3	1.121	3.884	.012
	LangS	1.852	3	.617	.636	.593
Error	LangP	33.106	94	.352
	LangJ	99.499	94	1.059
	LangM	27.135	94	.289
	LangS	91.192	94	.970
Total	LangP	2007.000	104
	LangJ	624.000	104
	LangM	1610.000	104
	LangS	778.000	104
Corrected Total	LangP	85.760	103
	LangJ	291.346	103
	LangM	71.538	103
	LangS	157.654	103
a R Squared = .614 (Adjusted R Squared = .577); b R Squared = .658 (Adjusted R Squared = .626) c R Squared = .621 (Adjusted R Squared = .584); d R Squared = .422 (Adjusted R Squared = .366)

Table 3: Tests of Between-Subjects Effects

The relationship between age and language proficiency is significant for all the languages in question (P< 0.05). Education is significantly related to language proficiency in Mandarin only (P<0 .05="" font="" is="" language="" of="" proficiency="" that="" the="">Mandarin (Mandarin) is closely related to both age and education. Gender has a strong effect on the language proficiency of Southern Min but only a marginal effect on Pangcah.

In this section, we will closely examine how the independent variables are related to language proficiency.

In Table 4 below, the B values (B parameter estimates) indicate the scores of proficiency. These scores are relative numbers based on the last reference groups in each variable categorisation as they are presented with zero values in the table. The positive and negative of the values denote whether proficiency has increased or decreased. The absolute value of B denotes the extent of the difference of proficiency with regards to the value of the reference group:

Dependent Variable	Parameter	B	Std. Error	t	Sig.	95% Confidence Interval
						Lower Bound	Upper Bound
Pangcah	Intercept	4.477	.329	13.626	.000	3.824	5.129
	[gen=1]	.223	.127	1.749	.084	-.030	.475
	[gen=2]	0(a)	.	.	.	.	.
	[cage=1]	-1.939	.298	-6.510	.000	-2.530	-1.348
	[cage=2]	-.830	.235	-3.526	.001	-1.297	-.363
	[cage=3]	-.649	.214	-3.027	.003	-1.074	-.223
	[cage=4]	-.317	.219	-1.449	.151	-.752	.118
	[cage=5]	-.047	.222	-.213	.832	-.489	.394
	[cage=6]	0(a)	.	.	.	.	.
	[edu=1]	.511	.294	1.734	.086	-.074	1.095
	[edu=2]	.194	.282	.687	.494	-.367	.755
	[edu=3]	.242	.246	.980	.330	-.248	.731
	[edu=4]	0(a)	.	.	.	.	.
Japanese	Intercept	4.830	.570	8.480	.000	3.699	5.961
	[gen=1]	-.022	.221	-.099	.922	-.460	.416
	[gen=2]	0(a)	.	.	.	.	.
	[cage=1]	-4.325	.516	-8.376	.000	-5.350	-3.300
	[cage=2]	-4.250	.408	-10.415	.000	-5.060	-3.440
	[cage=3]	-3.418	.372	-9.197	.000	-4.156	-2.680
	[cage=4]	-2.230	.380	-5.871	.000	-2.983	-1.476
	[cage=5]	-1.200	.386	-3.111	.002	-1.965	-.434
	[cage=6]	0(a)	.	.	.	.	.
	[edu=1]	-.594	.511	-1.163	.248	-1.607	.420
	[edu=2]	-.089	.490	-.182	.856	-1.061	.883
	[edu=3]	-.069	.427	-.162	.872	-.917	.779
	[edu=4]	0(a)	.	.	.	.	.
Mandarin	Intercept	3.596	.297	12.089	.000	3.005	4.186
	[gen=1]	-.118	.115	-1.028	.307	-.347	.110
	[gen=2]	0(a)	.	.	.	.	.
	[cage=1]	1.254	.270	4.652	.000	.719	1.790
	[cage=2]	1.282	.213	6.017	.000	.859	1.705
	[cage=3]	1.290	.194	6.649	.000	.905	1.676
	[cage=4]	1.198	.198	6.041	.000	.804	1.592
	[cage=5]	.303	.201	1.507	.135	-.096	.703
	[cage=6]	0(a)	.	.	.	.	.
	[edu=1]	-.878	.267	-3.292	.001	-1.407	-.348
	[edu=2]	-.530	.256	-2.073	.041	-1.038	-.022
	[edu=3]	-.537	.223	-2.407	.018	-.980	-.094
	[edu=4]	0(a)	.	.	.	.	.
Southern-Min	Intercept	.889	.545	1.631	.106	-.193	1.972
	[gen=1]	.530	.211	2.508	.014	.110	.949
	[gen=2]	0(a)	.	.	.	.	.
	[cage=1]	1.586	.494	3.208	.002	.604	2.567
	[cage=2]	1.731	.391	4.432	.000	.956	2.507
	[cage=3]	1.895	.356	5.325	.000	1.188	2.601
	[cage=4]	1.587	.364	4.365	.000	.865	2.309
	[cage=5]	.576	.369	1.561	.122	-.157	1.309
	[cage=6]	0(a)	.	.	.	.	.
	[edu=1]	-.148	.489	-.303	.763	-1.118	.823
	[edu=2]	.276	.469	.589	.557	-.654	1.207
	[edu=3]	.019	.409	.047	.963	-.793	.831
	[edu=4]	0(a)	.	.	.	.	.
a This parameter is set to zero because it is redundant.

Table 4: Parameter Estimates2

For the relationship between gender and language proficiency, there is only a small difference between men and women in their language proficiency in Pangcah, Japanese, and Mandarin. The differences lie in the fact that men are more proficient in Pangcah than women but less proficient than women in Japanese and Mandarin. Compared with the language proficiency of Southern Min, the men’s command of Pangcah is significantly better than women’s. This finding can be attributed to the fact that the men go out to work in the urban areas. Many of them work on construction sites or do low-level manual work where Southern Min is required for communication with the Han Chinese.

The relationships between age and language proficiency in Pangcah, Japanese, Mandarin, and Southern Min will be considered separately. Cage 6 (age over 60) will be taken as the reference group to serve as a base scale. Its parameter value therefore is zero. In Pangcah, proficiency is the highest among the oldest people and declines with age decreases, with scores going from zero into larger negative values. As may have been expected, there is not much difference between Cage 5 and Cage 6. Accordingly, proficiency starts declining from Cage 4 on, as it refers to the younger groups, and there is a sharp decrease in language proficiency between Cage 2 and Cage 1.

Like Pangcah, Japanese also undergoes language shift. However, the degree of the shift is much greater than that of Pangcah as the absolute value (between 0 and 5) is much higher than that of Pangcah (between 0 and 2). In relation to the degrees of language shift, there is not much difference between Cage 6 and Cage 5, Cage 5 and Cage 4, Cage 4 and Cage 3, and Cage 3 to Cage 2, respectively. As it comes down to the last two age groups (aged below 40), there is very little difference in proficiency.

Unlike the proficiency levels of Pangcah and Japanese, the proficiency level of Mandarin is maintained. Pangcah people aged over 60 have a lower proficiency level in Mandarin. However, those who are under 60 have a similar language proficiency, as it is shown that the B values of Cage 4, Cage 3, Cage 2 and Cage 1 fall between a positive value of 1.1 and 1.3. As for Southern Min, the situation is similar, in spite of the much lower proficiency level than that in Mandarin.

Figure 1 provides an overall picture of the relationships between age groups and language proficiency. The use of age as a social variable permits examination of the role of language policy with respect to the language proficiency, as age is an important variable in sociolinguistic surveys (Tsao 1997, Cenoz 2003). At the same time, it can provide a timeline for cross-check with policy implementation.

Figure 1: Relationship between age groups and language proficiency3

As one might expect, and as Figure 1 indicates, the older the person, the better their mastery of Pangcah; the situation is the same for Japanese. Oral mastery of Pangcah is steady although facing a gradual decline before meeting the crossing point with Mandarin. A sharp decline appears in the Pangcah mastery of those subjects who are between thirty and forty .An explanation for this finding is that in between 1970 and 1986, the national language movement was at its most intense period of implementation. Mandarin was heavily reinforced in educational settings, whilst the use of Japanese was strictly forbidden in surroundings such as at school and in the media. Most elderly Pangcah people who are over seventy are able to speak Japanese under the influence of the Japanese colonisation. In 1945, the Japanese were defeated and the Chinese nationalists took over Taiwan. The crossing point of the Japanese and Mandarin languages corresponds to these political events. After 1945, spoken Japanese continued declining. However, among those aged between 31 and 40, there appears to be an increasing ability to speak Japanese. However, from the point of language proficiency, this improvement is not significant. This means that most average Pangcah people, although they claim to have some Japanese language ability, do not have sufficient skills to carry out a conversation in Japanese. Still, in the village, one may hear some fragments of Japanese, such as the use of some Japanese vocabulary items, used by people of this age group.

For Mandarin, the situation is exactly the opposite: the younger the person, the better his or her Mandarin.

For Southern Min, the situation is slightly less clear. Based on this survey, it appears that their language ability is increasing slightly but not to any significant degree. The above data suggest that in Tafalong, competence among the villagers is incontrovertibly increasing. Competence of Southern Min may increase if the Democratic Progressive Party, whose supporters are mainly Southern Min, remains the ruling party. In contrast, Japanese is fading away, leaving many of those aged between 50 and 60 as semi-speakers. For Pangcah, we see a sharp decrease in ability due to the ‘linguicism’ (a term used by Phillipson 1992) of the national language movement, though it is not possible to see a distinctive fall in this figure due to the fact that those who are under 20 are left out because only 3 out of 280 young Pangcah were active Pangcah speakers in Tafalong. However, although this is not the only factor, it is clear that, based on the findings of the present study, the change of political entities and language policies genuinely influences the ecology of the Taiwanese aboriginal languages.

As for the relationship between education and language proficiency of Pangcah, the only real difference is Edu 1. However, Edu 2 and Edu 3 are quite similar and neither of them is very different from Edu 4 (Parameter value = 0). This means that those who have a primary level of education tend to have higher language proficiency.

For Japanese, the situation is reversed. Those subjects who only have a primary school level of education are less proficient in Japanese. There is very little difference in the remaining categories. For Mandarin, those who have high-school education, as compared to those who have earned a college degree, are slightly less proficient. Those who have primary education only tend to have the lowest proficiency in Mandarin. From the cases of Japanese and Mandarin, it can be concluded that those who have a low level of education have a low proficiency level in the language of instruction. For Southern Min, the proficiency level is similar among those who have at least senior-high-school education. Those who reached junior-high-school level have the lowest proficiency and those who reached primary-school level have the lowest proficiency level.

4.1.2 Language Use

In the following section, the first part of the research question, which deals with language choice, i.e. Are gender, age and education related to the language choice of the respondents in communication with different interlocutors in different settings? will be analysed. In this study, a modified version of the domain approach of Fishman (1964, 1965, 1972) is adopted. Seven domains are included: family domain, friendship domain, religious domain, shopping domain, school domain, work domain, and official-institution domain. Each domain consists of different combinations of interlocutors (Table 5).

A preliminary and overall examination of language choice in relation to its domain and the interlocutors reported by the informants under the descriptive statistics will be presented first (Table 5), then followed by a more detailed analysis of our results concerning the relationships between language choice with different interlocutors and independent variables as age, gender and education.

Domain	Interlocutor	Language Use Patterns		No interlocutor involved
		Pangcah as the most frequently used language*	Dominant language(s) as the most frequently used language(s)**
Family	Grandparents	89.5%	3.8%	6.7%
	Parents	88.5%	11.5%	0.0%
	Husband/Wife	65.4%	29.8%	4.8%
	Children	3.8%	80.8%	15.4%
Friendship	Pangcah Friends	80.8%	19.2%	0.0%
	Pangcah and Chinese Friends	47.1%	49.1%	3.8%
	Chinese Friends	3.8%	80.8%	15.4%
Religion	God(s)	62.5%	29.8%	7.7%
	Priests	35.6%	38.4%	26.0%
	Church Friends	62.5%	16.4%	22.1%
Shopping	Salespersons in Pangcah Stores	78.8%	21.2%	0.0%
	Salespersons in Chinese Stores	11.5%	88.5%	0.0%
School	School Staff	7.7%	59.6%	32.7%
Work	Colleagues	22.1%	38.5%	39.4%
Official Institutions	Staff	20.2%	75.0%	4.8%
* Pangcah as a category listed in the language choice refers to the (total or predominant) use of Pangcah. ** The term dominant language(s) refers to Japanese, Mandarin or Southern Min. This category hints to the dominant use of at least one of the languages listed above.

Table 5: Percentage of Self-reported Language Choice between Pangcah and Respective Dominant Language(s) with or without Interlocutors Involved

From the above figures, it can be concluded that the heritage language is still used in the domain associated with Pangcah ethnicity. What is worth noting is that within the family domain, most informants claim that the heritage language is still most frequently used in their communication with parents and grandparents. However, only 3.8% of informants within the valid sample of 104 claimed that their most frequently used language spoken with their children is Pangcah. This shows figure that the dominant language, in most cases Mandarin, almost replace Pangcah in the communication between the informants and their children in the household.

After this preliminary result, our data were further analysed through LR in pursuing the research question: Are age, gender and education related to the choice of language use between Pangcah and the dominant language(s)? The independent variables are age, gender and education; the dependent variable is the self-reported information about language choice between Pangcah and the dominant languages. Similar to MANOVA, in LR, various models with one, two or three variables and their interaction terms are are decided in order to improve our model. Apart from this, the categories in both dependent and independent variables are combined with other categories due to the sparsity of data: age is split into three age-groups (21-40 years, 41-60 years, 61 onwards); the levels of education are merged into two (below and above secondary education); the language choice is categorised into two (Pangcah as the most frequently used language and (a) dominant language(s) as the most frequently used language(s)). Table 6 presents the factors which are significantly related to the language choice used with the specified interlocutors, using the Likelihood Ratio Test within LR (P<0 .05="" font="">:

Domain	Interlocutor	Significant factor (s)
Family	Grandparents	Age
	Parents	Age
	Husband/Wife	Cage*
	Children	Age
Friendship	Pangcah Friends	Age, Gender
	Pangcah and Chinese Friends	Cage
	Chinese Friends	Age
Religion	God(s)	Cage, Gender
	Priests	Cage, Education
	Church Friends	Age
Shopping	Salespersons in Pangcah Stores	Age, Gender
	Salespersons in Chinese Stores	Cage
Education	School Staff	Age
Employment	Colleagues	Age
Official Institutions	Staff	Age
* ‘Cage’ here refers to the categories of age.

Table 6: The Significant Factors for the Self-reported Language Choice between Pangcah and the Dominant Language(s)

As shown in the above table, in some situations, Age appears to be the significant factor; in other situations it is Cage. The reason for the flexibility in considering both Cage and Age as independent variables is that, while operating the variables, it was found that, in some situations, the role of age variable was the better predictor, while in other situations, a categorised version of age (cage) was a better predictor.

From Table 6, it can particularly be deduced that Age / Cage is significant across the board in relation to language use within the seven domains specified. Education is only significant in situations when the informants talk to priests (Domain of Religion). Gender has an effect on the informants’ conversation with Pangcah friends (Domain of Friendship), with Gods(s) (Domain of Religion) and with salespersons in Pangcah shops (Domain of Shopping).

4.2 Domains

Turning to Parameter Estimates in LR enables us to examine some details about how age, gender and education affect the language choice of the participants with the fifteen sets of interlocutors whom they interact with in the different domains.

4.2.1 Family Domain

Table 7: Parameter Estimates of Language Use in the Family Domain

For every year of increase in age, a person is 1.115 times as likely to speak the heritage language rather than the dominant languages to their grandparents. Alternatively we might say that for twenty years of increase in age, a person is 8.82 times as likely to speak the heritage language rather than the dominant language(s) to their grandparents (1.115²⁰=8.82).

Similarly, for every year’s increase in age, a person is 1.128 times as likely to speak the heritage language rather than the dominant language(s) to their parents.

Members of the younger Pangcah generation aged between 21-40 (cage = 1) are only 0.045 times as likely to speak Pangcah to their spouse or partner as a heritage language. Those informants who are aged between 41-60 are 0.593 times as likely to speak the heritage language to their spouse or partner.

For every year of increase in age, a person is 1.147 times as likely to speak the heritage language rather than the dominant language(s) to their children. Alternatively we might say that for every twenty years’ of increase in age, a person is 15.533 times as likely to speak the heritage language rather than the dominant language(s) to their grandparents (1.147²⁰=15.533).

4.2.2 Friendship Domain

Table 8: Parameter Estimates of Language Use in the Friendship Domain

For each year of increase in age, a person is 1.270 times as likely to speak the heritage language rather than the dominant language(s) to their Pangcah peers. Pangcah males are 4.857 times as likely as females to speak the heritage language with their ethnic friends.

The use of Pangcah in communication with friends of mixed races (both Pangcah and Han Chinese) is closely related to age Representatives of the younger generation are 0.125 times as likely to speak the heritage language than those belonging to the oldest group, i.e. aged over 60. Those informants who are between 41-60 are 0.579 times as likely to speak the heritage language rather than the dominant language(s) as the oldest group.

For every year of increase in age, a person is 1.124 times as likely to speak the heritage language rather than the dominant language(s) with their Chinese friends.

4.2.3 Religious Domain

Table 9: Parameter Estimates of Language Use in the Religious Domain

There are two variables related to language use in prayers, namely Gender and Cage. Compared to Pangcah women, Pangcah men are 3.804 times as likely to speak the heritage language rather than the dominant language(s) in religious contexts. The younger generation is 0.015 times as likely to speak the heritage language as the oldest group, i.e informants over 60. Those who are between 41 and 60 are 0.351 as likely to speak the heritage language rather than the dominant language as compared to the oldest group.

Compared to those who received at least secondary education, those whose education is lower than this level are 4.868 times as likely to speak with priests using the heritage language rather than the dominant language(s). If we take elderly people as the reference group, the youngest generation is 0.098 times as likely to speak the heritage language with their priest. Similarly, if we compare the elderly and the middle-aged group, the middle-aged Pangcah are 0.103 times as likely to speak the heritage language rather than the dominant language(s) with their priest. The chance of speaking the heritage language for either the younger generation or the middle-aged generation is almost the same.

For every one year of increase in age, a person is 1.122 times as likely to speak the heritage language rather than the dominant language(s) with their Chinese friends.

4.2.4 Shopping Domain

Table 10: Parameter Estimates of Language Use in the Shopping Domain

For every year of increase in age, a person is 1.102 times as likely to speak the heritage language rather than the dominant language(s) with the salespersons in Pangcah stores. Compared to Pangcah women, Pangcah men are 5.406 times as likely to speak the heritage language as the dominant language(s) with the salespersons in Pangcah stores.

Compared to elderly people aged over 60, those aged under 60 are 5% to 6% times as likely to speak the heritage language rather than the dominant language(s) with the salespersons in Chinese stores.

4.2.5 Education Domain

Table 11: Parameter Estimates of Language Use in the Education Domain

For every year of increase in age, a person is 1.202 times as likely to speak the heritage language rather than the dominant language(s) to the school staff.

4.2.6 Work Domain

Table 12: Parameter Estimates of Language Use in the Work Domain

For every year of increase in age, a person is 1.047 times as likely to speak the heritage language rather than the dominant languages to colleagues at work.

4.2.7 Official Institution Domain

Table 13: Parameter Estimates of Language Use in the Official Institution Domain

For every year of increase in age, a person is 1.166 times as likely to speak the heritage language rather than the dominant language(s) with government clerks. Compared to women, men are 0.307 times as likely to speak the heritage language other than the dominant language(s) with government clerks in official institutions.

As mentioned above, age (cage) is unequivocally correlated with the pattern of the language choice in communication with different interlocutors in various domains across the board. Family, the most important site for heritage language transmission, is examined below. In Figure 2 below the relationships between age and the probability of most frequently using Pangcah with grandparents, parents and children in the family domain are presented:

Figure 2: The Probability of Most Frequently Using Pangcah with Grandparents, Parents and Children

Figure 2 shows that there is a general downward trend in the probability of using Pangcah with the three sets of interlocutors as age decreases. That is, the older Pangcah people are, the higher the possibility of their using Pangcah, and the opposite is also true. The probability of using Pangcah with children in particular differs noticeably between the two in that the decline in the possibility of using Pangcah runs from the eldest to the youngest and the degree of decrease is much larger, because the slope is much steeper than for the other two. In contrast, the probability of speaking Pangcah with parents and grandparents begins to decline more distinctively at the age of 50.

Although gender was not presented as being significantly related to language choice with most of the interlocutors, the results obtained indicate that Pangcah men tend to use more Pangcah in intragroup communication (i.e. among Pangcah friends, to God(s) and with salespersons in Pangcah stores). In official domains, such as in government institutions, the results show that Pangcah women tend to use more Pangcah than men do. However, this is the case in situations where there is only one case and the relationship between language choice patterns and gender is borderline significant (P = 0.096). This finding does not represent wholly convincing evidence for claiming that Pangcah women are better language maintainers in official settings. Previous studies (Gal 1979, Milroy 1980, and Hill 1987) suggest that women play a leading role in language shift due to the social pressure which they feel, such as insecurity, stigmatisation, or devaluation. In the present case study, the gender difference on language shift needs further research, since nowadays, the status of aboriginal people in Taiwan in general is at the bottom stratum of society. Moreover, sense of insecurity and lack of self-esteem are not just shared only by women. Pangcah is a matrilineal society, in which gender relationships differ from those in the above-mentioned studies. Based on the present study, it is safe to state that Pangcah men are better language maintainers especially in the inner domains, such as talking to ethnic friends, in Pangcah stores and talking to God(s). The level of education is only significant in conversations with priests.

5 Conclusion

In the present study, the relationships between three relevant variables (age, gender and education), language proficiency and language use in various domains within the community in question have been analysed. The findings of the sociolinguistic survey of language proficiency and language use show that age is a variable which is related to the language shift and maintenance of various languages in Tafalong, one of the biggest Taiwanese indigenous villages. Education has its impact on language proficiency in Mandarin.

In language use, there is a general shift towards the dominant language(s), mainly Mandarin. The language choice of adults with children, in the family domain in particular, appears to be shifting towards Mandarin monolingualism. This shows that the heritage language is not transmitted to future generations.

Language use in in-group communication in other low domains like friendship, religion and shops shows that male Pangcah are better language maintainer than female ones. However, in most of the domains, whether high or low, where Han Chinese are present, there is a general shift to Mandarin, which shows little resistance towards the dominance of Mandarin.

Within less than a century, the linguistic hierarchy in Tafalong presented an unstable diglossia under the Japanese colonial period (1895-1945), and the Chinese Nationalist regime (1945-1987), respectively. Despite the fact that Tafalong shows a dense and multiplex network structure based on the fact that intermarriage among villagers is common and cooperation remains one of the important merits of Pangcah tribal daily practice, modernisation has changed both the culture and lifestyle, the younger generations tending to increase their language contact with Han Chinese (Southern-min and Hakka) through doing university studies or taking urban jobs. This has brought some motivation for language change.

The ‘active’ speakers of Pangcah in Tafalong are numerous in comparison with those of other nearly extinct aboriginal languages in Taiwan. However, the language shift of Pangcah has been too accelerated for it to be overlooked. That is, the language shift described has occurred within three generations, and children no longer acquire Pangcah, substituting Mandarin as their native language. This is a sign of earlier language death under the long-term linguistic hegemony, as severe language policies were imposed during the Japanese and nationalist colonial eras. Mandarin has clearly become the sole language in use nowadays amongst most aboriginal children and adolescents.

In light of the above findings, the following recommendations may serve as a basis for developing Taiwan’s indigenous language maintenance:

• As the present study is rather limited in scope, a large-scale language survey is needed. The results of such a survey would provide an initial basis for making decisions about future language policy.
• Although the analysis of language use and proficiency presented here has provided a sketch of language maintenance and language shift of a Pangcah village, the survey data by themselves do not provide information about cultural values. For this reason, other devices (for example, the analysis of language ideologies embedded within language practices) for measuring the causes of language shift are needed in order to produce more explanatory results.
• As this paper shows that the Pangch people studied cannot withstand and resist the dominance of Mandarin, documentation should be useful for resolving issues concerning literacy, such as aboriginal materials, teacher training and linguistic studies, which arise from the lack of a comprehensive corpus. This will also be of benefit for preserving linguistic data as a human resource for future generations and research.

From the examination of Pangcah, one of Taiwan’s endangered indigenous languages, it is asserted that Pangcah will be difficult to maintain. According to Williams’ (1991) findings, the following sociolinguistic realities are of importance in any attempt at language maintenance: a language can be maintained when it is used by its speakers on a daily basis; however, efforts undertaken by policy-makers, language educators and activists are likely to be in vain if the indigenous languages are not enabled to become vehicles of social, political and economic advancement.

Appendices

Appendix I: Questionnaire for Language Proficiency and Language Use

Questionnaire

This questionnaire has been designed to investigate the language use among the aboriginal Pangcah people. Your information is extremely important for this project. Please answer fully and frankly so as to make this study valuable. All questionnaires are anonymous and will be treated confidentially. I am grateful to you for taking the time to complete this questionnaire!

I: Language Proficiency

2. The following section is about your overall language proficiency. Please assess your language proficiency level in the spaces provided. The scales are as follows:
3. 5 - “very fluent”
4. 4 - “fluent and no problem with communication”
5. 3 - “able to communicate but sometimes with difficulties”
6. 2 - “understand most of the conversation but only within sentence-structure level when communicating”
7. 1 - “comprehension is very limited and only able to produce a few words or phrases”;
8. 0 - “totally unintelligible”

Languages	Language Proficiency

Appendix II: Language Use in Daily Lives

What is / are the language(s) which you most frequently use in talking to the following interlocutors?

1. 1= I use Pangacah most frequently;
2. 2= I use Japanese, Mandarin (Mandarin) and / or Southern Min most frequently

No	Interlocutors	Language(s) Used	Notes
1	Grandparents and their generation
2	Parents and their generation
3	Spouse / Partner
4	Pangcah friends
5	Pangcah and Han Chinese friends
6	Han Chinese friends
7	Children and their generation
8	Colleagues
9	Clerks working in the government sectors
10	God(s)
11	Priests
12	Church friends
13	School staff
14	Sales persons in Pangcah-running shops
15	Sales persons in Chinese-running shops

Appendix III: Personal Information

1. Gender: __________

2. Which year and what month were you born in? ____________

3. Education __________________________________________

References

Baker, Colin (2006). Psycho-sociological analysis in language policy research. In T. Ricento (ed.) An Introduction to Language Policy: Theory and Method (pp 210-228). Malden, MA: Blackwell.

Dittmar, Nobert (1976). Sociolinguistics: A critical survey of theory and application. London: Edward Arnold.

Fasold, Ralph (1984). The Sociolinguistics of Society. Oxford: Blackwell.

Fishman, Joshua, A. (1972). Domains and the relationship between micro- and macrosociolinguistics. In John. Joseph. Gumperz and Dell. Hymes (eds.) Directions in Sociolinguistics (pp 435-453). New York: Holt Rinehart and Wilston.

Fishman, Joshua, A. (1965). Who speaks what language to whom and when? La Linguistique 2, 67-88.

Fishman, Joshua, A. (1964). Language maintenance and language shift as a field of inquiry. Linguistics 9, 32-70.

Gal, Susan (1979). Language Shift: Social determinants of linguistic change in bilingual Austria. New York: Academic Press.

Garcia, MaryEllen (2003). Recent research on language maintenance. Annual Review of Applied Linguistics, 23, 22-43.

Hill, Jane H. (1987). Women’s speech in modern Mexicano. In Susan, Philips, Susan, Steele, and Christine,Tanz (eds.) Language, Gender and Sex in Comparative Perspective (pp 121-60). Cambridge: Cambridge University Press.

Huang, Shuan-fan (1995). Language, Society and Ethnicity: A study of the sociology of language in Taiwan. Taipei: Crane. [in Chinese]

Kulick, Don (1992). Language Shift and Cultural Reproduction: Socialization, self and syncretism in a Papua New Guinean village. Cambridge: Cambridge University Press.

Household Office, Kuang-fu Township of Hua-lien County, (2018). Statistics for the population of Kuang-Fu Township. Retrieved 3, March 2018, from http://em.hl.gov.tw/population_list.php?menu=&typeid=3077

Li, Wei. (1997). Language shift in the Teochew Community. Journal of Multilingual and Multicultural Development. 18, 5, 364-384.

Milroy, Lesley. (2001). Bridging the micro-macro gap: social change, social networks and bilingual repertoires. In Jetske, Klatter-Folmer and Piet. van Avermaet (eds.), Theories on Maintenance and Loss of Minority Languages: Towards a more integrated explanatory framework. Munster: Waxmann Verlag GmbH.

Milroy, Lesley. (1980 & 1987, 2^nd ed). Language and Social Networks. Oxford: Blackwell.

Pauwels, Anne. (2004). Language maintenance. In Allan, Davies and Catherine, Elder (eds.), The Handbook of Applied Linguistics (pp 719-737). Oxford: Blackwell.

Phillipson, Robert. (1992). Linguistic Imperialism. Oxford: Oxford University Press.

Author:

Ya-ling Chang

Assistant Professor

Department of Foreign Languages and Literature

National Yilan University

#1, Sec. 1, Shennong Rd

Yilan City

Yilan County

Taiwan 26047

1 It should be noted that the above group does not include subjects who are twenty years old or below because these people are mostly Mandarin speakers only.

2 Coding for age categorisation: over 70 years old = cage 6, 61-70 = cage 5, 51-60 = cage 4, 41-50 = cage 3, 31-40 = cage 2 and blow 30 =cage 1. Coding for Education: College = Edu 4, Senior High School = Edu3, Junior High School = Edu 2, and Primary School = Edu 1. Coding for gender: male = Gen 1, female = Gen 2.

3 5 stands for ‘very fluent’, 4 for ‘fluent and no problem with communication’, 3 for ‘able to communicate but sometimes have difficulties’, 2 for ‘understand most of the conversation but only within sentence-structure level’, 1 for ‘comprehension is very limited and only able to produce a few words or phrases’; 0 for ‘totally unintelligible’.