From 899032c9728c7a1c9c97f624ba0cc49b0814277b Mon Sep 17 00:00:00 2001
From: Chris Dyer <cdyer@moto.clab.cs.cmu.edu>
Date: Sun, 3 Jan 2016 16:17:35 -0500
Subject: corpus stats script

---
 corpus/corpus-stats.pl | 50 ++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 50 insertions(+)
 create mode 100755 corpus/corpus-stats.pl

diff --git a/corpus/corpus-stats.pl b/corpus/corpus-stats.pl
new file mode 100755
index 00000000..0bbd49b4
--- /dev/null
+++ b/corpus/corpus-stats.pl
@@ -0,0 +1,50 @@
+#!/usr/bin/perl -w
+use strict;
+
+my $f = <>;
+my $IS_PARALLEL = ($f =~ / \|\|\| /);
+if ($IS_PARALLEL) {
+  die "This script is only valid for monolingual corpora, but file contains |||\n";
+}
+
+my %d;
+my $tc = 0;
+my $lc = 0;
+while($f) {
+  $lc++;
+  chomp $f;
+  my @toks = split /\s+/, $f;
+  for my $t (@toks) {
+    $d{$t}++;
+    $tc++;
+  }
+  $f=<>;
+}
+
+my $types = scalar keys %d;
+my $ttr = $tc / $types;
+my @mfts;
+for my $k (sort {$d{$b} <=> $d{$a}} keys %d) {
+  push @mfts, $k;
+  last if scalar @mfts > 24;
+}
+my $sing = 0;
+for my $k (keys %d) {
+  if ($d{$k} == 1) { $sing++; }
+}
+my $stypes = sqrt($types);
+
+print <<EOT;
+CORPUS STATISTICS
+
+          Lines: $lc
+         Tokens: $tc
+          Types: $types
+    sqrt(types): $stypes
+ Type-tok ratio: $ttr
+     Singletons: $sing
+
+Most freq types: @mfts
+
+EOT
+
-- 
cgit v1.2.3