diff --git a/bip-0039.mediawiki b/bip-0039.mediawiki
index 2a6118b26a..0ee41bec11 100644
--- a/bip-0039.mediawiki
+++ b/bip-0039.mediawiki
@@ -48,6 +48,10 @@ are split into groups of 11 bits, each encoding a number from 0-2047, serving
 as an index into a wordlist. Finally, we convert these numbers into words and
 use the joined words as a mnemonic sentence.
 
+Words are joined using a single space character (U+0020), except in Japanese,
+where the ideographic space (U+3000) is used; see
+[[bip-0039/bip-0039-wordlists.md#wordlists-special-considerations|Wordlists (Special Considerations)]].
+
 The following table describes the relation between the initial entropy
 length (ENT), the checksum length (CS), and the length of the generated mnemonic
 sentence (MS) in words.
@@ -110,6 +114,11 @@ mnemonic" section is possible, this is not advised and software must compute a
 checksum for the mnemonic sentence using a wordlist and issue a warning if it is
 invalid.
 
+Computing that checksum requires splitting the sentence into words, which must
+accept either separator: NFKD normalization maps U+3000 to U+0020, so splitting
+after normalization is one way; splitting on U+0020 before normalization reads an
+ideographic-space-separated sentence as a single word.
+
 The described method also provides plausible deniability, because every passphrase
 generates a valid seed (and thus a deterministic wallet) but only the correct one
 will make the desired wallet available.