From 2ef170a991acda127b624fab27eb619379799899 Mon Sep 17 00:00:00 2001
From: Matthew Honnibal <honnibal@gmail.com>
Date: Thu, 16 Apr 2015 04:28:06 +0200
Subject: [PATCH] * Fix Issue #54: Error merging multi-word token when there's
 a mid-token match.

---
 spacy/tokens.pyx    | 2 ++
 tests/test_merge.py | 6 ++++++
 2 files changed, 8 insertions(+)

diff --git a/spacy/tokens.pyx b/spacy/tokens.pyx
index b61302c24..27d99a045 100644
--- a/spacy/tokens.pyx
+++ b/spacy/tokens.pyx
@@ -281,6 +281,8 @@ cdef class Tokens:
             if self.data[i].idx == start_idx:
                 start = i
             if (self.data[i].idx + self.data[i].lex.length) == end_idx:
+                if start == -1:
+                    return None
                 end = i + 1
                 break
         else:
diff --git a/tests/test_merge.py b/tests/test_merge.py
index 39693b178..370a334b8 100644
--- a/tests/test_merge.py
+++ b/tests/test_merge.py
@@ -30,3 +30,9 @@ def test_merge_heads():
     assert tokens[3].head.i == 1
     assert tokens[4].head.i in [1, 3]
     assert tokens[5].head.i == 4
+
+
+def test_issue_54():
+    text = u'Talks given by women had a slightly higher number of questions asked (3.2$\pm$0.2) than talks given by men (2.6$\pm$0.1).'
+    tokens = NLU(text, merge_mwes=True)
+