Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@normalization/languages/english/operators.py`:
- Line 195: The percent-insertion regex in operators.py is too broad and
rewrites legitimate “<number> of …” phrases, including cases that should remain
as the prior rule/article normalization handled them. Tighten the matching logic
around the existing text rewrite near the percent conversion so it only fires in
the intended spoken-percentage contexts (for example, by constraining it with
surrounding words or possessive/proper-noun patterns), and ensure the
`rule`→`article` replacement and the `re.sub(...)` percent conversion in the
same normalization flow no longer conflict on phrases like “rule 142 of the
agenda” or “5 of the members.”

In `@normalization/languages/english/replacements.py`:
- Around line 1746-1751: The VoxPopuli/parliamentary aliases in
ENGLISH_REPLACEMENTS are too broad and are rewriting valid English tokens like
puttering in all normalization paths. Move these entries out of the shared
replacements map and into a VoxPopuli-specific preset or normalization step,
keeping the general English replacement map free of corpus-specific aliases; use
the replacements module and the ENGLISH_REPLACEMENTS symbol to locate the shared
mapping.

In `@normalization/steps/text/remove_trailing_apostrophe_space.py`:
- Around line 19-20: The current remove_trailing_apostrophe_space text
normalization is over-aggressive because it collapses any 2+ letter "<word> s"
pattern, not just possessives. Update remove_trailing_apostrophe_space.py so the
trailing-s handling happens earlier in the apostrophe-stripping flow, or
restrict it to a narrower corpus-specific transform instead of the shared
text_post substitution; use the existing remove_trailing_apostrophe_space
function as the place to remove the global re.sub on "\b([a-z]{2,}) s\b" and
preserve real tokens like "letter s" and "model s".

---

Nitpick comments:
In `@tests/unit/languages/english_voxpopuli_normalization_test.py`:
- Around line 26-38: The percent normalization test coverage in
test_percent_of_patterns is missing a positive digit-`of` case, so a bug in the
literal-digit rewrite could slip through. Extend the parametrized cases in this
test to include a digit-based input like “15 of latvia population” with the
expected “15 percent of latvia population”, so the pipeline.normalize behavior
is verified for the digit-`of` branch as well.