1. Introduction
In Japan, as in many other nations, the official characterization of the standard language variety contends that it is not the language variety of any one specific region of the country (Shibatani 1990: 185-187). While this very diplomatic conceptualization of the standard does serve a purpose, in reality, many (though by no means all) Japanese speakers consider it common sense that hyôjungo (standard Japanese) is a language variety spoken in and around Tokyo. In this paper, I employ a "Perceptual Dialectology" methodology, inspired by the work of Dennis Preston (1988) in the United States, in an attempt to shed some light on the geographical perceptions Japanese speakers have of the standard variety of their language. I propose some methodological innovations, namely techniques for the mapping of means and standardized values, the utilization of "proportion tests" (Davis 1990) to determine statistical significances between two maps, and the mapping of these significance test results themselves.

2. Speech Variety Labels
In a series of surveys conducted in eight regions of Japan, I collected hand-drawn maps from 756 informants native to their respective regions. I converted the geographical boundaries of the 5517 hand-drawn perceptual speech variety regions into numerical data using a digitizer, and used privately-developed software to produce numerous "composite" perceptual dialect maps. (For a detailed outline of this methodology, see Long 1990a or Long forthcoming b.)

Here I will examine the types of language variety labels which the informants applied to the speech regions they drew, but first I should clarify that there were two situations in which informants' responses were regarded as unusable and not included in the data. One was cases in which informants wrote a speech variety label on the map but did not draw boundaries to define the region of its usage. The other was the opposite case in which informants drew the boundaries for a region but did not label it. Only terms which clearly referred to language were regarded as speech variety labels. Thus, an area labeled "Hiroshima-ben" 'Hiroshima dialect' was included in the data, but one simply labeled "Hiroshima" was ignored.

There were a total of 257 different speech variety labels used. These may be divided into two main types: locative labels which employ a word indicative of a place, and non-locative labels which do not. Locative labels consist of two parts, a locative descriptor plus a variety descriptor. Both Japanese and English have a large array of variety descriptors. Informants in Preston's (1989:44) study of perceptual dialects of the United States gave the following variety descriptors (shown with frequency of occurrence): accent (45), English (25), drawl (14), speech (5), twang (3), jargon (2), style (1), lingo (1), dialect (1), slangs (sic, 1), and talk (1).

In this study, the most common variety descriptor was ben 'dialect', used 4855 times and accounting for 88% of the total labels. Interestingly, the term hôgen 'dialect', preferred by linguists, accounted for only 4.1% (228 occurrences) of the total. The other descriptors (with their frequencies) were as follows: go 'language' (378 occurrences; 6.9%), kotoba 'talk' (43 occurrences; 0.8%), namari 'accent' (9 occurrences; 0.2%), guchi 'language' (Okinawan term, 2 occurrences), kei '-related' (1 occurrence), gengo 'language' (1 occurrence).1 With locative descriptors, as well, we can identify certain subtypes, namely specific locatives which refer to a specific place, such as "Tokyo-ben", and generic locatives which refer to a general type of place, such as "shima-kotoba" 'island talk'.

3. Is "Standard" a Language Variety with a Region?
First, let us examine how many of our informants gave "standard Japanese" as the variety spoken in a specific region.2 In the survey, we did not ask informants to identify, for example, the area where they thought people used "standard Japanese" or "Kyoto Dialect", etc. Rather we asked them to "draw lines to designate areas where the speech differed".3 Thus, both the question of how many speech regions Japan can be divided into, and of what those speech divisions are was left to the discretion of the informant. The smallest number of perceptual dialect regions drawn by an informant was 2, and the greatest number drawn by a single informant was 23. The mean number of regions drawn by one informant was 7.3.

In light of this, we will begin our discussion of the perceptional regions for "standard Japanese" with an examination of how many informants labeled areas as standard on their maps. 281 informants, roughly one-third of the total, drew "standard Japanese" as a region on their maps, but as we see in table 1, there were great differences among the percentages for each informant region.

These results do indeed tell us something about the dialect perceptions (and attitudes) of the Osaka informants. The fact that they view the speech of the Kantô region as a dialect, rather than the standard, to a significantly higher degree is evidence of their reluctance to concede to a Tokyo-based standard. In other words, compared to informants of other regions (even the regions of Western Japan in our study, which are generally less fond of Tokyo speech than are the regions of Eastern Japan), Osaka informants think of the speech of the Tokyo area not as "standard" but as the dialect they call "Kantô-ben".

4. Percentage Map
In our study, we obtained data from informant groups native to eight different regions of Japan, and as we saw above, between 23 and 58% of the informants chose to indicate "standard" as a geographically-defined language variety in their answers. Due to space restrictions, we cannot examine all of the "standard" maps here, but we will look at the maps drawn by the Nagoya and the Osaka informants, in an effort to find both the common and unique aspects of the response patterns of the two groups.

First, look at the map of standard-speaking areas compiled from the responses of the Nagoya informants (figure 1, which has the locations mentioned in this paper labeled for reference). The strongest tendency we find here is to indicate a core area centered around Tokyo and radiating outward from there. A look at the Osaka informants' results in figure 2 shows us that this is a shared tendency. In fact, this tendency is seen in the maps of all eight informant groups. Something else which strikes us about figure 1 is the way in which the home region of the Nagoya informants is not included within the standard area. It would appear, in fact, that the home region has been carefully excluded, since Nagoya itself is included only at the lowest level (1-19%), while it is almost surrounded by areas of the 20-39 percentage range.

Figure 1, Nagoya informants "standard" (percentages)
In contrast, figure 2 shows us that the level of Osaka people who perceive the Nagoya region as being standard-speaking is in the 20-39% range, and thus exceeds the level indicated by the Nagoya informants themselves. With the Osaka informants, in addition to the fair number who perceive "standard" areas to the east of them, there is even a small number who perceive standard-speaking areas west of themselves in the Hiroshima area. Here too, however, we find that none of the Osaka informants identify the home region as "standard", and thus we may see this as another example of the "home region exclusion" tendency.

Figure 2, Osaka informants "standard" (percentages)

This tendency to exclude the home region from the standard-speaking area also turned up in the maps of several other informant groups,4 and to this extent we can view this phenomenon as a common characteristic among many of the maps. While the phenomenon itself can be seen as a single tendency, it naturally produces a degree of variation into the data, however, since the area which comprises the home region varies from one informant group to the next. We shall discuss these differences between the informants groups' maps later in the paper.

Additionally, in figure 2 we see a strong perception of the northernmost island of Hokkaidô (seen as an inset in the top left of the maps) as an area of standard usage. Although this tendency does not show up in the Nagoya map, it did appear in six of the eight informant groups. Even so, the perception of Hokkaidô as "standard" was strongest in the Osaka map with scores in the 20-39% range, while the other informant regions were in the 1-19 percent range.5

On comparison of the two maps, we see that the Tokyo-centered core area is weaker in the Osaka informants' map than it is for the Nagoyans (and, incidentally, for the other six groups whose maps are not included in this paper as well). This weakened core is, as a mater of fact, directly related to the strength of the Hokkaidô results here. Whereas the informants in Nagoya (and the other six survey locations) all basically agreed that "standard" was centered in the Tokyo area, the votes for the Osaka informants were split. While some informants drew lines to indicate that "standard" was used in both Hokkaidô and Tokyo, many others indicated strictly the Hokkaidô area, excluding Tokyo altogether. This disagreement within the Osaka informant group resulted in a "standard" with not a single core area, but one split into two parts, neither of which obtained the extremely high percentages seen in the other maps.

Reviewing the results above, we can identify three clear tendencies for the perceptual region of "standard", outlined as follows: (1) a Tokyo core (a tendency, seen in the maps of all eight regions, to indicate a core area centered around Tokyo and radiating outward from there); (2) a Hokkaidô supplement (a less common tendency to include the northern island of Hokkaidô); (3) home region exclusion (a tendency to avoid the inclusion of the home region within the "standard" area).

Examining these maps visually, our natural inclination is to look for dissimilarities in them. Such dissimilarities give us insight into how people's perceptions of standard Japanese differ from one region to another. Visual comparisons of the two maps are, however, impressionistic and thus lacking in accuracy. The fact that the percentage values in the maps have been grouped into five levels further reduces the accuracy of our visual perceptions. For this reason, I searched for a method to determine those areas whose levels of perception differed at a statistically significant level. I will describe this procedure next.

5. Significance Test Results
One method for determining the unique characteristics of two informant groups is to compare their results using what is commonly referred to as the "proportion" test. The proportion test is used when the original data is in the form of a "yes/no" type question and the two values being compared are percentages. Let us take a moment here to understand just what values are being used in the calculations, because a grasp of this will facilitate our understanding of how the other statistical procedures outlined below were applied as well.

In order to calculate the informants' responses in our study, I divided the map of Japan into a dot matrix with over 27,000 dots. I began with the line(s) drawn by informant 1 to designate "standard" and determined which of the dots were included inside the "standard" area. I repeated this procedure with all the informants who chose to indicate "standard". In this way, I determined, for each dot in the matrix, what percentage of the informants had included that dot within the area they called "standard". Thus, for each of the 27,000 some odd dots on the map, I obtained eight percentage figures, one for each of the informant groups.

To determine if there was a statistically significant difference in the percentage of Osaka informants and Nagoya informants who included a certain dot within the "standard" area, I performed the proportion test on the two percentages. I then repeated this test on the rest of the dots in the map matrix and illustrated the results of these 27,000 significance tests on the map. In figure 3, I applied a diagonal striped pattern to those areas which were included in the "standard" maps of either the Osaka or the Nagoya informants but which showed no significant difference. The dark spotted pattern designates those areas where the Osaka informants' values were higher to a statistically significant degree. The lightly spotted pattern shows the same for those areas in which the Nagoya informants' values were higher.

Figure 3, Osaka and Nagoya informants "standard" (proportion test results)

In other words, figure 3 shows (1) areas which Osaka informants were more prone to label "standard", (2) areas which Nagoya informants were more inclined to include in the area called "standard", and (3) areas which showed no significant difference between the two. In the first category, we find the island of Hokkaidô. In the second, we find parts of the Kantô region around Tokyo, as well as a sizable portion of the Kansai region (the region which contains Osaka). These results can be interpreted in the light of previous studies which have shown that speakers from Osaka (and the other parts of the Kansai region) have a low regard for the speech of Tokyo and the Kantô region around it (see for example Long 1990b; Long forthcoming a; Inoue 1989: 161, 187). This negative impression has manifested itself in the data here in the form of the reluctance shown by many Osaka informants to acknowledge the Kantô area as being standard-speaking. As we saw in figure 2, the percentage of Osaka informants who included Hokkaidô in the standard area was rather high, and we see in figure 3 that, when compared to the Nagoya informants, this difference is so great it is statistically significant. Furthermore, we see that significantly more of the Nagoya informants are willing to include portions of the Kansai region in their standard-speaking area than are their Osaka counterparts. In other words, we see that the "home region exclusion" tendency mentioned earlier has been statistically verified. Although the number of Nagoya informants who included this strip of the Kansai region in the standard was low, the number of Osaka informants who did so was zero.

There is a limit to the usefulness of these proportion test maps, however. Since we have 8 informant groups, a thorough cross-comparison would require us to create and examine 32 different maps. Thus, we are forced to search for more efficient ways of analyzing informant-group differences.

6. The Eight-Region Means Map
One method for obtaining an overall view of the entire body of data is to calculate the mean values for the eight informant groups. Mapping these mean values gives us the map in figure 4. Here we see two tendencies mentioned previously, that is the appearance of a strong Tokyo-centered core area, and a weaker tendency to include Hokkaidô in the standard-speaking area. What's more, producing this average map gives us a new basis for comparison. Rather than comparing the Nagoya map with the Osaka map, and then the Nagoya with the Kagoshima, the Nagoya with the Tokyo, and so on, we can now compare each of the eight maps with this average map to examine, if not how each maps differs from each other map, at least how each map differs from the average.

Figure 4, Eight informant groups "standard" (means)

Look again at the Osaka informants' "standard" percentage map in figure 2. We see here that a high percentage of informants (between 60 and 79%), have included the area around Tokyo in their "standard" area. At first, this would appear to indicate that a large portion of the Osaka informants see the Tokyo area as "standard", and to a degree it does. But when we compare these results with those of the eight-region average map in which the Tokyo area ranks at the 80-100% level, we see that the Osaka informants' values are not very high. On the contrary, they are lower than average. The next step now is to determine whether our visual impressions of differences between the Osaka map and the average are indeed statistically significant. We can accomplish this by standardizing the values for each informant group's map.

7. Standardized Values Map
I standardized the percentages for each dot on each of the eight maps in order to see the levels on the maps not in isolation but in relation to the percentage levels of the other seven maps. This standardization of the scores allows us, in effect, to compare each of the eight region maps with the average. In the standard score maps, the eight-region average (for that dot on the matrix) is set at 50 and the percentages are recalculated so that those higher than average fall proportionately above 50, and those lower fall proportionately below 50. Moreover, scores greater than 69.6, or less than 30.4 differ from the average to such an extent that the difference is statistically significant.

Let us take another look at the Osaka informants' "standard" areas, but this time, rather than simply percentages, we will look at the standardized values (figure 5). Whereas with the percentage map, it appeared that a large number of our Osaka informants acknowledged the Tokyo area as a standard-speaking one, the standardized values map reveals this impression to be quite misleading. Even though in figure 2 the area in and around Tokyo ranked the highest with numbers in the 60-79% percentage range, we find that these figures are actually much lower than what we would expect given the eight-region average. This is indicated on the map by the extremely light pattern. Moreover, in the center of this light area, we find that, in fact, the small area comprising Tokyo prefecture itself is so much lower than the average (falling below the 30.4 level) that the difference is statistically significant.

Figure 5, Osaka informants "standard" (standardized values)

The high percentage of Osaka informants who included Hokkaidô is amplified in the standardized scores as well. For, although five of the seven other regions also included Hokkaidô to some extent in their "standard" areas, we see that in the case of the Osaka informants, the tendency was so strong that the difference is statistically significant. Furthermore, the inclusion of Hiroshima (and environs) to the west of Osaka was only indicated by a small percentage of the Osaka informants and thus appeared as a very weak result on the percentage map, but the fact that this area was not included whatsoever in any of the seven other maps causes it to receive above-average (though not statistically significant) scores on the standardized map here.

8. Standard Deviations Map
Above, we have seen disagreements among different regions of Japan as to just what areas speak "standard". Now, we will look at a way to measure this degree of disagreement quantitatively and illustrate it on a map. Just as we standardized the scores for each dot on the map matrix above, we can calculate the standard deviation for each dot on the matrix as well. The higher the standard deviation is, the more disparity there was among the eight regions' "standard Japanese" values for that dot. In other words, a high standard deviation score shows that there was a great deal of disparity among the percentages of the eight informant groups.

On the standard deviation map in figure 6, we find that the Tokyo area has a very low standard deviation. This is because, in spite of the fact that the informants from the Osaka region had significantly low scores for this area, the remaining seven groups were basically in agreement about this area's standardness. (All of them placed it in the 80-100% range.) The area which encountered the most disagreement among the eight groups was the Chûbu region situated in the center of the main island. This is indeed an area whose standardness is often disputed, both by linguists as well as the average person on the street.

Figure 6, Eight informant groups "standard" (standard deviations)

9. Topics for Future Research
In this paper, I introduced some innovations to the perceptual dialectology methodology to examine the perceptions of standard Japanese held by informants native to two different geographical regions. I developed a technique for comparing two composite maps and for showing those areas which differ significantly in a map format. Using this method, I found that informants from different regions of Japan had noticeably different (often to a statistically significant degree) ideas about what areas speak standard Japanese. In particular, I found that informants from Osaka are reluctant to acknowledge the Tokyo region as a standard-speaking one, a finding which reinforces the evidence of Osaka's linguistic rivalry with Tokyo seen in previous studies. Furthermore, I found that standardized score maps were an effective tool in revealing unique qualities of informant groups which had been obscured in simple percentage maps. Finally, I demonstrated how the mapping of standard deviations could be used to determine what areas were the subject of disagreement among informant groups.

In one of the survey questions not analyzed in this paper, informants were asked to list specific traits which they thought characterized the speech regions they had indicated (Long 1993, Long forthcoming b). Professional linguists have long asserted the existence of both "standard Japanese" and of a "Tokyo dialect". Preliminary analyses of the specific linguistic characteristics which our informants associate with these two speech varieties suggest, however, that non-linguists may not differentiate between the two in strictly linguistic terms. In the future, I wish to pursue this type of problem in an attempt to shed new light on not just the "where" issues of standard language perception, but the "what" issues as well.

Notes