word rank | frequency | n-gram |
---|---|---|
1 | 21922 | k- |
2 | 14631 | t- |
3 | 13988 | p- |
4 | 12855 | s- |
5 | 11832 | v- |
word rank | frequency | n-gram |
---|---|---|
1 | 5789 | ko- |
2 | 5237 | ka- |
3 | 4659 | va- |
4 | 3231 | te- |
5 | 3163 | ma- |
word rank | frequency | n-gram |
---|---|---|
1 | 1316 | val- |
2 | 997 | väl- |
3 | 946 | koo- |
4 | 942 | või- |
5 | 915 | rah- |
word rank | frequency | n-gram |
---|---|---|
1 | 576 | välj- |
2 | 547 | vast- |
3 | 530 | vali- |
4 | 466 | suur- |
5 | 450 | kesk- |
word rank | frequency | n-gram |
---|---|---|
1 | 477 | välja- |
2 | 375 | riigi- |
3 | 329 | kooli- |
4 | 317 | vastu- |
5 | 304 | välis- |
The tables show the most frequent letter-N-grams at the beginning of words for N=1…5. Their frequency is count without multiplicity, otherwise the stopwords would dominate the tables.
As shown in the above example (German), word prefixes are clearly visible. In the above example, ver- and ein- are prefixes, and Sch- is not. At the end of a prefix we typically have a wide variety of possible continuations. Hence a prefix of length k will be prominent in the table for N=k, but typically not in the table for N=k+1. The prominent entries Schw- and Schl- for N=4 tell us that Sch- is no prefix.
Zipf’s diagram is plotted with both axis in logarithmic scale, hence we expect nearly straight lines. The graphs look more typical for larger N. Especially for N=3 we find only a small number of trigrams resulting in a sharp decay.
For a language unknown to the reader, the data can easily be used to see whether prefixes do exist and to find the most prominent examples.
For counting, only words with a minimum character length of 10 were considered.
Because only a word list is needed, the tables above can be generated from a relatively small corpus.
For N=3:
SELECT @pos:=(@pos+1), xx.* from (SELECT @pos:=0) r, (select count(*) as cnt, concat(left(word,3),"-") FROM words WHERE w_id>100 group by left(word,3) order by cnt desc) xx limit 5;
For more insight in a language, longer lists might be useful.
Is there a need for larger N
Most frequent word endings
Most frequent letter-N-grams
Number of letter-N-Grams at word beginnings
Number of letter-N-Grams at word endings