From 2f9faf95bb7a54a3d0f48a91867c9efe77676234 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Kristian=20J=C3=A4rventaus?= <kristian@clausal.com>
Date: Mon, 9 Dec 2024 12:49:24 +0200
Subject: [PATCH] [en] Form descriptions: check if a 'form' has too many words

Fixes #983: "present participle Martinize" was interpreted
as tags ["present"] and the form "participle Martinize".

This was because the code at:

```python
if (
   i > 1
   and len(parts[i - 1]) >= 4
   and distw(titleparts, parts[i - 1]) <= 0.4
   ):
       alt_related = related
       alt_tagsets = tagsets
       continue
```

would see that "participle" is very close (exactly 0.4)
in distw() Levenshtein distance to "Martinize" and
accepted that it must be part of the titleparts
list, like a word in a phrase.

This kludge just adds a test to see if titleparts is
shorter than the rest of `parts`, don't do that.

THIS WILL FAIL IN CERTAIN SCENARIOS, like when an
'inflected' form gains more words!
---
 src/wiktextract/extractor/en/form_descriptions.py | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/src/wiktextract/extractor/en/form_descriptions.py b/src/wiktextract/extractor/en/form_descriptions.py
index e39575928..b31cf10a5 100644
--- a/src/wiktextract/extractor/en/form_descriptions.py
+++ b/src/wiktextract/extractor/en/form_descriptions.py
@@ -2440,13 +2440,25 @@ def strokes_repl(m: re.Match) -> str:
                         or any("error-unknown-tag" in x for x in tagsets)
                     ):
                         if alt_related is not None:
+                            # We already had a good division, so let's stop.
                             break
+                        # Bad division, try deeper
                         continue
                     if (
                         i > 1
                         and len(parts[i - 1]) >= 4
                         and distw(titleparts, parts[i - 1]) <= 0.4
+                        # Fixes wiktextract #983, where "participle"
+                        # was too close to "Martinize" and so this accepted
+                        # ["participle", "Martinize"] as matching; this
+                        # kludge prevents this from happening if titleparts
+                        # is shorter than what would be 'related'.
+                        # This breaks if we want to detect stuff that
+                        # actually gets an extra space-separated word when
+                        # 'inflected'.
+                        and len(titleparts) >= len(parts[i - 1:])
                     ):
+                        # print(f"Reached; {parts=}, {parts[i-1]=}")
                         alt_related = related
                         alt_tagsets = tagsets
                         continue
@@ -2491,6 +2503,7 @@ def strokes_repl(m: re.Match) -> str:
                     tagsets = alt_tagsets
 
             # print("FORM END: tagsets={} related={}".format(tagsets, related))
+            # print("==================")
             if not tagsets:
                 continue