diff options
Diffstat (limited to 'corpus')
| -rwxr-xr-x | corpus/paste-files.pl | 1 | ||||
| -rwxr-xr-x | corpus/support/quote-norm.pl | 3 | ||||
| -rw-r--r-- | corpus/support/token_list | 2 | ||||
| -rwxr-xr-x | corpus/tokenize-anything.sh | 2 |
4 files changed, 7 insertions, 1 deletions
diff --git a/corpus/paste-files.pl b/corpus/paste-files.pl index 4cb424ad..ef2cd937 100755 --- a/corpus/paste-files.pl +++ b/corpus/paste-files.pl @@ -34,6 +34,7 @@ while(1) { $done = 1; last; } + $r =~ s/\r//g; chomp $r; if ($r =~ /\|\|\|/) { $r = ''; diff --git a/corpus/support/quote-norm.pl b/corpus/support/quote-norm.pl index e4e5055e..b104e73c 100755 --- a/corpus/support/quote-norm.pl +++ b/corpus/support/quote-norm.pl @@ -11,6 +11,8 @@ while(<STDIN>) { s/&\s*squot\s*;/'/gi; s/&\s*quot\s*;/"/gi; s/&\s*amp\s*;/&/gi; + s/&\s*nbsp\s*;/&/gi; + s/&\s*#\s*160\s*;/ /gi; s/ (\d\d): (\d\d)/ $1:$2/g; s/[\x{20a0}]\x{20ac}]/ EUR /g; s/[\x{00A3}]/ GBP /g; @@ -20,6 +22,7 @@ while(<STDIN>) { s/&\s*#45\s*;/--/g; s/�c/--/g; s/ ,,/ "/g; + s/„/"/g; s/``/"/g; s/''/"/g; s/[「」]/"/g; diff --git a/corpus/support/token_list b/corpus/support/token_list index d470cb22..366cd7ff 100644 --- a/corpus/support/token_list +++ b/corpus/support/token_list @@ -37,6 +37,8 @@ tel. 10. ##################### abbreviation: words that contain period. +EE.UU. +ee.uu. U.A.E Ala. Ph.D. diff --git a/corpus/tokenize-anything.sh b/corpus/tokenize-anything.sh index 1a24193d..028992cf 100755 --- a/corpus/tokenize-anything.sh +++ b/corpus/tokenize-anything.sh @@ -9,5 +9,5 @@ $SUPPORT/utf8-normalize.sh | sed -e 's/ al - / al-/g' | $SUPPORT/fix-contract.pl | sed -e 's/^ //' | sed -e 's/ $//' | - perl -e 'while(<>){s/(\d+)(\.+)$/$1 ./;print;}' + perl -e 'while(<>){s/(\d+)(\.+)$/$1 ./; s/(\d+)(\.+) \|\|\|/$1 . |||/; print;}' |