Attitudes towards the use of Tharu are consistently strong and positive. All indications are that Tharu will remain in use as a mother tongue for the foreseeable future. Attitudes are tentatively positive
towards literature in Tharu; however, attitudes towards literature in Kathoriya Tharu are somewhat negative, though this assessment is very preliminary.
3.5 Tharu culture
Like many tribal groups in India, the Tharu are changing rapidly. In many ways they are maintaining an uncomfortable balance between their traditional culture and the encroaching Hindu culture. Tharu
culture today is a blend of both worlds. Among the younger generation, education is providing opportunity for advancement in the national culture; the other key force of change is immigration into
traditional Tharu areas by outside groups. The maintenance of language, a key measure of change in a culture, suggests that the Tharu will continue to maintain a strong sense of cultural identity in the
foreseeable future.
4 Study of dialect areas
The purpose of a dialect area study is to define in quantifiable terms the differences existing between speech varieties in a given geographic area. In order to understand the extent of dialect differences
throughout the western Indo-Nepal Tarai, a dialect area study was carried out which consisted of 1 wordlist comparison, and 2 dialect intelligibility testing.
4.1 Wordlist comparison
4.1.1 Procedures
Comparing wordlists between two points is one method of measuring the similarity of those two speech varieties. This systematic study of vocabularies is known as a lexical similarity study. Speech varieties
that have more words in common higher lexical similarity, generally understand each other better than those communities that have fewer words in common.
Analysis of the wordlists was by means of grouping similar words together for each English gloss and calculating the percentage of similar words between any two wordlists. Similarity is based on
phonological similarity and not strictly on cognate relationship, using similarity counting procedures outlined in Blair 1990:31–32. After the words were grouped according to these counting procedures, an
analysis was run using the compass algorithm in Wimbish’s Wordsurv program 1989. Additional phonological regularities were identified in this way, and the groupings readjusted to account for these.
Wordlist similarity, or lexical similarity, below 60 percent typically corresponds with inadequate intelligibility between the compared varieties reflective of distinct languages. Lexical similarity above
90 percent typically corresponds with high intelligibility between the compared varieties reflective of very closely related dialects. Dialect intelligibility testing is not usually required for either situation.
Lexical similarity between 60 and 90 percent warrants fuller investigation by means of dialect intelligibility testing to determine the nature of the relationship between the two speech varieties Blair
1990:23.
A 210-item wordlist was elicited from mother-tongue speakers of each speech variety under investigation, and transcribed using first the Devanagari script and then the International Phonetic
Alphabet. In most cases, wordlists were double-checked with other mother-tongue speakers to check for errors and to weed out items which appear to be different but which are really only synonyms. Past
experience shows that double-checking wordlists tends to increase lexical similarity percentages. In some cases, elicitation was also done with several mother-tongue speakers present, thus providing a
measure of built-in double-checking from the beginning. Specific details for each wordlist are given in Appendix A.
Because Hindi plays such a strong role as an inter-group language there is often the problem of eliciting a Hindi word when a local word is commonly used. For this reason, a strong effort was made at
all times to elicit local Tharu words if they were still in use.
4.1.2 Results
A total of sixteen wordlists were compared: Table 2 identifies each list by the three letter code, name, and location it represents. The maps show the location of these wordlist points.
Table 2. Wordlists, source locations, and identity codes
Code Speech variety
Location: Village, Tehsil, District, Country
BNM Bhuksa Tharu
Madnapur, Gandepur, Nainital, India BNT
Bhuksa Tharu Thari, Ramnagar, Nainital, India
RNK Rana Tharu
Sugia, Khatima, Nainital, India RNS
Rana Tharu Sisaikera, Sitarganj, Nainital, India
RNs Rana Tharu
Sisana, Sitarganj, Nainital, India RkM
Rana Tharu Majhgam, Kanchanpur, Kanchanpur, Nepal
RKB Rana Tharu
Bangama, Nighasan, Kheri, India TkN
Thakur Tharu Naibasti, Mahendranagar, Kanchanpur, Nepal
KkP Kathoriya Tharu
Pavera, Pavera, Kailali, Nepal SkP
Sunha Tharu Piparia, Mahendranagar, Kanchanpur, Nepal
DKS Dangora Tharu
Sivratnapur, Asuliya, Kelali, Nepal DDK
Dang Tharu Kotani, Dang, Dangdeokuri, Nepal
DGC Dangora Tharu
Chandanpur, Tulsipur, Gonda, India DkR
Dangora Tharu Rajipur, Kanchanpur, Kanchanpur, Nepal
CCC Chitwan Tharu
Chitwan, Chitwan, Chitwan, Nepal HIN
Hindi Standard, Uttar Pradesh, Hindi, India
The matrix in table 3 presents the results of all comparisons, expressed as a percentage of lexical similarity.
Table 3. Lexical similarity for all wordlist points BNM—Buksa
93 BNT—Buksa BUKSA
77 76 RNK—Rana 79 76 97 RNS—Rana
79 77 97 97 RNs—Rana RANA
77 73 91 89 93 RkM—Rana 77 75 90 91 89 88 RKB—Rana
76 75 84 83 87 85 83 TkN—Thakur 69 66 74 74 76 74 79 74 KkP—Kathoriya
LINK DIALECTS 67 65 71 71 74 70 70 71 73 SkP—Sunha
65 63 67 67 68 66 72 68 79 74 DKS—Dangora 59 58 64 63 64 63 66 63 76 73 92 DDK—Dang
DANGORA 64 61 68 68 69 69 71 66 79 72 89 88 DGC—Dangora
60 58 63 63 66 66 65 67 74 72 85 86 82 DkR—Dangora 58 56 56 57 57 57 58 60 63 63 65 63 61 67 CCC—Chitwan
OTHERS 83 80 70 70 71 68 70 72 68 66 64 59 58 65 60 HIN—Hindi
As much as possible in this table, varieties that are more similar are placed next to one another, and spaces are inserted to draw attention to dialect groupings. Several observations can be made about this
display of lexical similarity. Percentages vary from a high of 97 to a low of 56. It is also clear that the speech varieties cluster into several groups. Excluding Chitwan CCC and Hindi HIN, there are four
main subgroups: Buksa, Rana, Dangora, and a “link” group. Figure 1 graphically shows groupings and the similarity percentage at which any two groups can be linked. These percentages are from table 3. For
example, the Rana and Buksa groups are linked and become one group which shares lexical similarity of 73 percent or greater.
Figure 1. Grouping of test points based on lexical similarity data.
Subgroups of Tharu based on lexical similarity Two Buksa points BNM, BNT form the Buksa subgroup that unites at 93 percent lexical similarity. The
Buksa and Rana subgroups share at least 73 percent lexical similarity with each other. The Dangora group shares a lexical similarity of 82 percent or greater among four points DKS,
DDK, DGC, and DkR. This group spans considerable distance geographically, yet maintains a high degree of similarity.
Sunha SkP and Kathoriya KkP do not fit easily in any of the other subgroups. Sunha shares 70 percent or greater lexical similarity with the Rana group and 72 percent or greater with the Dangora
group. Kathoriya shares 74 percent or greater similarity with both the Rana and Dangora groups. Both Sunha and Kathoriya seem to form bridges between the Rana and Dangora groups: Rana and Dangora
only share 63 percent or greater lexical similarity with each other.
Other neighboring languages The geographically nearest Tharu variety examined in this survey is Chitwan Tharu CCC, using a
partial wordlist from Leal 1978. It is clearly quite different from the other varieties, sharing from a low of 56 percent lexical similarity with one of the Rana varieties to a high of 67 percent with one of the
Dangora varieties.
The lexical similarity results show that the Buksa subgroup is most closely related to Hindi 80 percent or greater; the Rana subgroup shares 68 percent or greater similarity with Hindi; the Dangora
subgroup is least similar of all the subgroups with Hindi 58–65 percent. Comparing lexical similarity with Hindi is particularly troublesome: it is very difficult to distinguish
between Hindi words that have become part of the local vocabulary, and those that have been given because the researchers were using Hindi. The Buksa wordlists may be especially affected by this
problem: Buksa is lexically closer to Hindi than to any of the other Tharu varieties, results which are a bit suspicious. In regard to this, see §4.2.2 for a discussion of the relationship between lexical similarity
and understanding of a recorded text.
4.2 Dialect intelligibility
4.2.1 Procedures
The procedures for testing dialect intelligibility are those described by Casad 1974 and Blair 1990. These procedures will be described briefly here.
For each speech variety under investigation a short two to four minute narrative story is recorded from a mother-tongue speaker. Each story is transcribed phonetically often in Devanagri first and
translated into English, thus facilitating the development of a list of simple questions about the story. Questions for each story are translated and recorded into each of the other speech varieties under
investigation. An RTT is then developed which consists of a story played one time through, followed by that same story interspersed with questions about that story. Questions are always in the mother-tongue
of the subject, and are only played one time each.
The places from which tests are developed are called reference points. The places in which tests are administered are called test points. A test developed and administered in the same place is known as a
hometown test. Each RTT must first be screened by a panel of ten mother-tongue speakers who validate the test by scoring nearly perfectly on the test in its final form of ten questions. Questions missed by
more than one mother-tongue speaker are usually eliminated. Likewise, each subject must score at least 80 percent on an RTT in their own mother-tongue before they qualify to take an RTT in another speech
variety. In this way the validity of each test and the suitability of each subject is ensured.
Generally speaking, if a sample of ten people from a test point averages 80 percent or higher on a recorded text test, then that community is said to adequately understand the dialect of the reference
point—the test point and the reference point are varieties of the same language.
These procedures were followed as closely as possible, but were altered in a few instances. Casad 1974 calls for screening potential questions with a panel of local speakers. In this survey, that
generally meant only one speaker at first; if an initial screening of the RTT revealed mistakes in the translation of the questions, we attempted to correct the mistaken questions before proceeding. The
initial form of the hometown test generally had between fifteen and thirty questions, which helped ensure ten good questions in the final test. In several instances subjects were allowed to continue even
though they scored less than 80 percent on a hometown test that was constructed from their own village. If they scored 90 percent or higher on the next test, we accepted them as suitable subjects, and assumed
that they had just needed a bit more time to adjust to the researchers and the testing method. In another instance, several subjects were tested on a series of tests without first taking a hometown test. These
were accepted only because they all scored 100 percent on the next test, proving their suitability as subjects.
The result of intelligibility testing is expressed as a percentage based on the mean average score of a sample of usually ten people. In order to ensure that what is being tested is inherent intelligibility and
not acquired intelligibility, standard deviation is calculated. A high standard deviation above 12 or 13 indicates relatively wide variation in subjects’ test performance. A common cause for such wide variation
is that some subjects have acquired intelligibility through contact with people from other speech varieties. Figure 2 shows the relationship between standard deviation and average score on a dialect
intelligibility test Blair 1990:25.
Standard Deviation High
Low
A ver
ag e S
co re
High Situation 1
Many people understand the story on the test tape well, but
some have difficulty. Situation 2
Most people understand the story on the test tape.
Low Situation 3
Many people cannot understand the story, but a few
are able to answer correctly. Situation 4
Few people are able to understand the story on the test
tape.
Figure 2. Relationship between standard deviation and average score on an inteulligibility test.
4.2.2 Results
Seven recorded text tests were developed and tested in this survey. The texts and their questions are included in Appendix A. Results from testing among these different Tharu varieties are shown in table 4.
The tests reference points are listed horizontally across the top; the places where each test was administered are listed vertically down the left side. The top number is the average for the sample; the
middle number is the standard deviation; the bottom number is the sample size. For example, ten subjects from DKS scored 91 percent on the test developed in RNs, with a standard deviation of 9.9.
Table 4. Summary results of recorded text test RTT BNM RNs
RKB KkP
DKS DGC
DDK BNM
98 95
— —
— —
— 4.4
10.1 —
— —
— —
13 10
— —
— —
— RNs
— 97
99 —
— 52
51 —
8.1 3.2
— —
18.1 14.5
— 19
10 —
— 10
10
BNM RNs RKB
KkP DKS
DGC DDK
RKB 100
96 98
90 88
78 69
7.0 4.6
12.5 9.8
8.6 16.3
10 10
23 10
10 10
10 KkP
— 96
95 94
93 91
84 —
6.0 7.1
10.7 8.3
8.8 17.1
— 10
10 19
10 10
10 DKS
— 91
— 97
95 83
87 —
9.9 —
4.8 8.3
11.6 15.7
— 10
— 10
20 10
10 DGC
— 73
75 95
79 96
72 —
18.5 18.6
7.0 15.1
7.3 14.8
— 11
11 10
11 23
10 DDK
— 68
— 95
— 71
93
— 15.5
— 5.3
— 12.9
8.5 —
10 —
10 —
10 18
The scores on the downward diagonal from left to right in bold are the hometown test scores. These scores ranged from a low of 93 on a sample of 18 in Dang District DDK, to a high of 98 on a
sample of 23 in Bangama RKB. Subjects missed questions on a hometown test because, in our opinion, a hometown test is so easy that their attention wandered. Other tests required, and received, greater
attention. It is interesting to note that in several instances subjects performed better on a test tape from another village than they did on a test from their own village.
Analyzing the scores in each vertical column reveals how well different test points understand the speech variety of that reference point. In general the Rana Tharu reference points RNs and RKB were
not understood well at the Dangora Tharu test points DDK and DGC. DDK only scored 68 percent on the RTT from RNs, with wide variation in understanding among the subjects as seen by the standard
deviation of 15.5. Likewise, the Dangora reference points were not understood well at the Rana test points: RNs only scored 51 on the test from DDK. RNs and DDK represent the geographic extremes in this
survey.
Intermediate geographically, and also according to lexical similarity, is Kathoriya Tharu KkP. All test points tested on KkP, RKB, DKS, DGC, and DDK averaged at least 90 percent. Test points RNs and
BNM were not tested on KkP; however, we can extrapolate from the results we do have. Average RTT scores among RNs, RKB, and BNM are uniformly high—all above 95 percent. Based on these high scores,
we could have chosen any one of these points as representative of the other two. This suggests that RNs and BNM should not score significantly different from RKB 90 percent on the KkP test. Therefore, there
is one point, KkP, that seems to be adequately understood by all test points in this survey.
Comparison of lexical similarity scores with RTT scores raises some questions. Lexical similarity scores are surprisingly low between the Rana and Buksa groups in light of the fact that RTT scores
between RNs and BNM are at least 95 percent. One reason for this, and there are several possible, is that the Buksa wordlists show a bias towards Hindi in their similarity: Buksa appears to be more lexically
similar with Hindi, which makes it appear less similar with the other Tharu varieties.
5 Bilingualism
Bilingualism is ability in a second language that is the result of learning, either formally as in school or informally as in the bazaar. A recorded text test RTT and a sentence repetition test SRT were used
in this survey to evaluate bilingual ability in Hindi. Bilingualism was only tested at the test points in India. It was not possible to test Nepali ability at any of the test points in Nepal.
5.1 Recorded text testing
5.1.1 Procedures
A recorded text test can be used to provide a preliminary assessment of a community’s learned ability in a second language. The procedures for evaluating bilingual ability using this method have been adapted
from dialect intelligibility testing as discussed in §4.2.1. Since this method only evaluates comprehension ability, it is not adequate for evaluating higher levels of bilingual proficiency Blair 1990:74. Used in
conjunction with sentence repetition testing we have a double check on our results, obtaining a more accurate understanding of bilingualism than if only one method were used.
When using recorded text tests in bilingualism testing, care must be taken to test a sample that is representative of the demographic characteristics in the community. A sample of subjects should
therefore include men and women, young and old, educated and uneducated, and traveled and untraveled, in the same proportion as the population as a whole.
5.1.2 Results
A Hindi recorded text test was developed in Bareilly, Uttar Pradesh. This was then tested at five test points as a preliminary, or pilot test, of bilingualism in Hindi. The two test points not tested are in Nepal
and are more influenced by Nepali. Subjects in DKS in Nepal on the border did relatively poorly on the Hindi RTT. On the basis of these results no further testing of Hindi was done in Nepal. All of the test
points in India did quite well, showing good understanding of a simple narrative text. These results are summarized in table 5.
Table 5. Hindi RTT Results HinRTT
Results from this pilot test show that further bilingualism testing is needed. Simple narrative material is understood well by the small samples tested; further testing was needed using the sentence
repetition test to more thoroughly investigate Hindi proficiency.
5.2 Sentence repetition testing
5.2.1 Procedures
A sentence repetition test SRT consists of a set of 15 carefully selected sentences recorded on a cassette tape. Each sentence is played once for each subject. Subjects are evaluated, according to a four point
scale 0–3, on their ability to accurately repeat each sentence. Essentially any deviation from the recorded sentences is counted as an error. A subject’s ability to accurately repeat sentences of increasing
difficulty is directly correlated with the overall ability to speak and understand the language. The higher the score, the greater the bilingual ability. Though an SRT is quite time consuming and difficult to
develop, once developed it is very quick and easy to administer, making it possible to evaluate a large sample in a community in a very short time. This procedure provides a more complete and accurate
evaluation of a community’s bilingual ability than recorded text testing. Radloff 1991 provides complete procedures for constructing and administering a sentence repetition test.
In a community different levels of bilingual ability frequently pattern with such demographic factors as sex, age, education, and amount of travel. These factors, therefore, must be adequately
TEST POINTS RNs
RKB BNM
DKS DGC
DDK KkP
HIN HINDI
91 92
91 74.5
93.5 —
— 100
RTT 9.4
7.9 11.0
12.1 6.7
— —
10 10
10 10
10 —
— 10
accounted for in the sample tested using the SRT. A small sample of at least five to ten people must be tested for each different combination of demographic factors that the researchers expect to have a
significant effect on bilingual ability. The specific demographic factors are determined by observation, by informal interviewing, and from census data. Our samples were chosen keeping in mind the factors of
sex, education, and age, with the greatest importance placed on education.
SRT results are expressed as a point total out of 45 possible points. They are interpreted according to a corresponding bilingualism proficiency level, or reported proficiency evaluation RPE level. These
RPE levels range from 0+ very minimal proficiency to 4+ approaching the proficiency of a native speaker. Probably at least a level 3 proficiency is required to adequately understand most philosophical
or religious material Kindell 1991:28.
5
A phonetic transcription of the SRT is included in Appendix B, along with a more detailed and practical description of the RPE levels than is presented here. Table 6
relates Hindi SRT score with the equivalent RPE level Varenkamp 1991:9 and Radloff 1991:242. Table 6. Score ranges on Hindi SRT corresponding to RPE levels
SRT Score = out of 45 RPE level
Proficiency descriptions
44–45 = 4+
[Near-native-speaker proficiency] 38–43 =
4 [Excellent proficiency]
32–37 = 3+
[Very good, general proficiency] 26–31 =
3 [Good, general proficiency]
20–25 = 2+
[Good, basic proficiency] 14–19 =
2 [Adequate, basic proficiency]
8–13 = 1+
[Limited, basic proficiency] 4–7 =
1 [Minimal, limited proficiency]
0–3 = 0+
[Very minimal proficiency]
5.2.2 Results
A demographic profile of a community provides the basis for interpreting the results of bilingualism testing. The percentage of the community with certain social characteristics e.g., younger, uneducated,
female should be compared with that social group’s average proficiency. Table 7 gives a demographic profile for Tharu populations in this survey. This profile is based on detailed census figures for Kailali
District in Nepal Government of Nepal 1984a and a Tharu village in Gonda District Singh 1988:16, in addition to general literacy statistics for districts in Uttar Pradesh Bose 1991. The range of percentages
in each social category reflects the approximate spread among the various Tharu areas.
A total of 190 subjects in five Tharu villages were tested on the Hindi sentence repetition test. In each village a stratified sample was selected that was as representative as possible of the overall village
population, based on demographic profiles developed for each village.
5
This assumes that RPE levels can be equated with the FSI foreign service institute proficiency levels referred to in Kindell 1991. The exact relationship between FSI and RPE levels has not yet been demonstrated.
Table 7 Demographic profile of Tharu villages UNEDUCATED PRIMARY ED
HIGHER ED. SEX
AGE 0 YEARS
1–5 YEARS 6+ YEARS
TOTAL MALE
YOUNGER 15–34
17–22 4–7
4–5 29
MALE OLDER
19–22 1–3
0–1 23
35+ MALE
TOTALS 36–44
5–10 4–6
52 FEMALE YOUNGER
15–30 26–28
1–3 0–1
30 FEMALE
OLDER 16–18
0–2 18
35+ FEMALE
TOTALS 42–46
1–5 0–1
48 TOTALS
78–90 6–15
4–7 100
Overall results for each village are compared in table 8, along with results for just the uneducated part of the sample: x= mean average, s= standard deviation, N= sample size, L= equivalent RPE level
of the average. Table 8. Hindi SRT results by village
VILLAGE OVERALL
UNEDUCATED x
s N
L x
s N
L Sisana
RNs 27.7
12.5 42
3 19.9
11.3 21
2+ Mathpuri
BNM 28.5
9.0 47
3 24.7
9.1 26
2+ Bangama
RKB 21.8
11.8 42
2+ 16.1
9.5 27
2 Bhusahar
Hunchawa 19.3
9.4 39
2 14.7
7.3 23
2 Chandenpur
DGC 19.4
13.1 20
2 10.1
6.2 10
1+ Chandenpur was used as a pilot test point. A larger and more representative sample
was tested in the nearby village of Bhusahar Hunchawa. A few comments can be made about this general display of data. There is a slight decline in average
score from west to east: Sisana and Mathpuri are in Nainital District and have the highest scores; Bhusahar Hunchawa and Chandenpur are in Gonda District and have the lowest scores. There are several
reasons for this. Standard Hindi is spoken in Nainital District, but in Gonda District a non-standard
variety of Hindi is spoken. Also, the Tharu villages in Gonda District are more remote with less opportunities for learning Hindi than in Nainital District.
The difference in average score between most pairs of villages is not statistically significant. However, SRT results from Mathpuri are significantly different from Bangama, Bhusahar Hunchawa, and
Chandenpur, but not from Sisana.
7
Figure 3 shows the distribution of bilingual proficiency levels for both Mathpuri and Bhusahar Hunchawa. This figure shows that ability in Hindi is quite varied within a
village, and also between villages.
Figure 3. Distribution of Hindi proficiency in two Tharu villages with percentage of sample at each level. Keeping in mind that there is wide variation in Hindi ability among the villages, detailed results by
social category are presented together in table 9. Table 9. Summary of SRT results
UNEDUCATED 0 YEARS
EDUCATED 1 YEAR OR MORE
AGE GROUP: YOUNGER OLDER YOUNGER OLDER TOTAL
15–34 35+
15–34 35+
MALE x= 25.4
x= 18.7 x= 32.4 x= 27.3 x= 27.2
s= 9.8 s= 11.2 s= 7.8
s= 10.6 s= 10.9 N= 17
N= 30 N= 52
N= 13 N= 113
L= 2+ L= 2
L= 3+ L= 3
L= 3
7
Significant at p.05 on a chi-squared test.
UNEDUCATED 0 YEARS
EDUCATED 1 YEAR OR MORE
FEMALE x= 17.1
x= 13.7 x= 35.6 x= 17.3 x= 19.3
s= 8.4 s= 8.2
s= 6.6 s= 5.5
s= 11.1 N= 34
N= 26 N= 14
N= 3 N= 77
L= 2 L= 2
L= 3+ L= 2
L= 2 TOTAL
x= 18.1 x= 31.7
x= 24.0 s= 10.0
s= 8.7 s=11.6
N= 107 N= 83
N= 190 L= 2
L= 3+ L= 2+
These results show a typical pattern of second-language acquisition: the educated have a significantly better Hindi ability than the uneducated; for those who are uneducated, men have
significantly better Hindi ability than women; and younger people have better Hindi ability than older people. In every social category average Hindi ability is RPE level 2 or higher, characteristic of at least
“adequate, basic proficiency.” Those who have completed at least one year of formal education average RPE level 3+, characteristic of “very good, general proficiency.” Those who are educated can probably
use Hindi satisfactorily in most situations, though they may have some difficulty using Hindi for communication of philosophical or religious subjects.
However, interpretation of these results must consider the extent of education in the Tharu communities. A vast majority of the population, perhaps as much as 90 percent, still falls in the category
of uneducated; this segment of the population will not be able to adequately use Hindi for communication of complex subject matter.
6 Language use and attitudes, and language vitality
6.1 Procedures
A study of language use patterns attempts to describe which speech varieties a community uses in different social situations. These situations, called domains, are contexts in which the use of one
language variety is considered more appropriate than another Fasold 1984:183. A study of language attitudes attempts to describe people’s attitude towards the different speech
varieties that are known to them, and about the choices people should make with regard to language use. The primary method for studying both language use and language attitudes among the Tharu was
the use of orally administered questionnaires. Observation was also used. During initial wordlist collection and recorded text testing, a preliminary form of the language use
and attitude questionnaire was used, mostly informally, to see which questions were most appropriate and useful. Some questions that were useful to guide the researchers in the early stages of the survey
were not included in the final form of the questionnaire. This final form was administered at two test points in conjunction with bilingualism testing.
The questions were asked in Hindi, adding a potential bias from use of the prestige language by the researchers. Some inconsistency in how questions were asked, especially those probing language
attitudes, has led to results that are less reliable than they ought to be. The following questions comprised the preliminary form of the questionnaire. Those marked with
an asterisk were asked on the final form of the questionnaire; there are missing numbers for questions discussed in §1.2.2.
1. What do you call your language?
2. What other languages do you speak?
3. Do you ever speak Hindi with other Tharus?
4. What language do you speak to merchants in the bazaar?
5. What language is spoken in your home?
6 What language do the children use when playing?
10. Are there Tharus who speak differently from you? 11. …Where?
12. Do you understand the Tharu spoken in Kheri District? 13. …In Gonda District? [In Nainital District?]
14. Where is the sweetest Tharu spoken? 15. What language do you use in private worship?
16. In what language should a mother speak to her young child? 17. Should Tharu children learn to speak Hindi?
18.a Would it be a good thing for books to be written in Rana Tharu? 18.b …In Kathoriya Tharu?
19. Would you want your child or you to marry someone who spoke only Hindi? 20. …Only Tharu?
24. Do you think children here will still be speaking Tharu in 50 years?
6.2 Results