20.11.2014 Views

Translation Universals.pdf - ymerleksi - home

Translation Universals.pdf - ymerleksi - home

Translation Universals.pdf - ymerleksi - home

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

76 Anna Mauranen<br />

After setting up the frequency bands, and excluding the proper names,<br />

the procedure was as follows. At the first stage, the original Finnish texts’<br />

rank ordered vocabulary was adopted as the standard for comparison, i.e.<br />

the reference corpus. The other corpora were compared to this by noting<br />

the deviation of each item from the standard, that is, the difference in the<br />

item’s rank order position from the position of the same item in the standard.<br />

The deviations occurred either upwards (the item had a higher rank in the<br />

translation corpus than the standard), or downwards (the item’s rank was<br />

lower), which meant that they might have cancelled each other out unless only<br />

their eigenvalues (points of difference in the rank) were noted. The eigenvalues<br />

were then summed for an aggregate estimate of the difference between the<br />

reference corpus and each subcorpus at a time. The same procedure was then<br />

applied to comparisons between translational corpora at the second stage, with<br />

the subcorpus of mixed languages as the reference corpus.<br />

This experimental method was developed as a tentative measure for comparing<br />

the relative distances between corpora. It goes without saying that such<br />

a measure remains partial because it is based on lexical rank order differences<br />

only, but in the absence of comprehensive overall measures it can be used as<br />

a pointer, in conjunction with other measures where possible. On account of<br />

its exploratory character it is not well suited for existing tests of statistical significance<br />

– even nonparametric tests which in principle might be considered,<br />

make such assumptions about the populations which do not apply to data consisting<br />

of a mass of running text. What we can hope from the present method<br />

is, then, a rough outline of the degree to which corpora might differ from each<br />

other, and expect the outline to be filled out with complementary means.<br />

6. Findings<br />

Applying the comparative method above to the subcorpora resulted in some<br />

interesting, even surprising observations. What we find is that it was the<br />

medium-frequency band (50–79), not the lowest, which actually shows the<br />

greatest overall differences (Tables 1–4 below). However, there was some<br />

variation. Thus, in comparison with original Finnish, the English subcorpus<br />

is an exception: it shows a steady increase of deviations as we go down the<br />

frequency list.<br />

As a test of universality vs. SL specific interference, I suggested above that<br />

if universality overrides bilingual interference, there should be little difference<br />

between texts from mixed SLs and texts from particular SLs, but all of these

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!