NFKD (Compatibility Decomposition)

NFKD: The Most Aggressive Decomposition

NFKD (Normalization Form KD — Compatibility Decomposition) is the most expansive of the four normalization forms. It applies both canonical decomposition (like NFD) and compatibility decomposition, breaking everything down to its most primitive components without any recomposition.

Where NFD decomposes é → e + combining acute, NFKD does that plus it also decomposes compatibility characters like ligatures, superscripts, and fullwidth variants. The result is the longest possible representation of a string in Unicode.

NFKD vs NFKC: When to Choose Each

Both NFKD and NFKC apply compatibility folding. The difference is the final step: NFKC recomposes canonical sequences back into precomposed characters, while NFKD does not.

import unicodedata

# é as precomposed (U+00E9)
text = "caf\u00e9"

nfkc_result = unicodedata.normalize("NFKC", text)
nfkd_result = unicodedata.normalize("NFKD", text)

print(len(nfkc_result))  # 4 — é recomposed to single code point
print(len(nfkd_result))  # 5 — é left as e + combining acute

# fi ligature
ligature = "find"
print(unicodedata.normalize("NFKC", ligature))   # "find" — 4 chars
print(unicodedata.normalize("NFKD", ligature))   # "find" — 4 chars (fi has no combining marks to add)

# Superscript
sup = "x²"
print(unicodedata.normalize("NFKC", sup))  # "x2" — 2 recomposed
print(unicodedata.normalize("NFKD", sup))  # "x2" — same (2 has no accent to decompose)

Practical Uses of NFKD

Diacritic stripping with compatibility folding: NFKD is the standard starting point when you want to remove accents AND fold compatibility characters:

import unicodedata

def aggressive_normalize(text: str) -> str:
    # 1. NFKD: compatibility fold + decompose
    nfkd = unicodedata.normalize("NFKD", text)
    # 2. Drop all combining marks (accents, etc.)
    stripped = "".join(
        c for c in nfkd
        if unicodedata.category(c) != "Mn"
    )
    return stripped

print(aggressive_normalize("fiancée"))   # fiancee
print(aggressive_normalize("résumé"))   # resume
print(aggressive_normalize("naïve²"))   # naive2

Database full-text search: Some search systems use NFKD + accent stripping as a pre-processing step to improve recall — a search for "resume" will match "résumé".

Fingerprinting and deduplication: NFKD provides a canonical key for detecting near-duplicate strings that differ only in how they encode the same visual text.

The Information Loss Warning

Like NFKC, NFKD is a lossy transformation. Once you strip superscripts, ligatures, and combining marks, you cannot recover the original. Only use NFKD for derived index keys or search normalization — never as your storage format.

Quick Facts