New Brunswick data projectOne place, two official names: a bilingual data project on New Brunswick's place names
Natural Resources Canada publishes the official geographical names for each province, refreshed weekly from the national database. In New Brunswick, some features carry an English and a French official name. Our students find them with Python and then test their own rule. Every count below is ours.
Our count of the Canadian Geographical Names file for New Brunswick (cgn_nb_csv_eng.csv), downloaded 22 October 2026| What we counted | Result |
|---|
| Rows (official names) in the file | 11,663 |
| Distinct features (Toponymic Feature ID) | 11,469 |
| Features with exactly two official names | 194 |
| Of those, parks and protected areas (Concise Code PARK) | 166 |
| Pairs where both rows give the Language as Undetermined | 189 |
| Pairs where our rule marks exactly one name as French-looking | 158 |
| Pairs our rule cannot settle, left for a human check | 36 |
Here is the script, which needs only Python 3 and the unzipped CSV in the same folder:import csv
FRENCH_STARTS = ("Zone ", "Parc ", "Rivière", "Lac ", "Ruisseau", "Lieu ", "Réserve", "Unité", "Île", "Baie ", "Pointe", "Cap-")
ACCENTS = set("àâçéèêëîïôûùüÿœ")
def looks_french(name):
return name.startswith(FRENCH_STARTS) or any(ch in ACCENTS for ch in name.lower())
names_for = {}
with open("cgn_nb_csv_eng.csv", encoding="utf-8-sig") as f:
for row in csv.DictReader(f):
names_for.setdefault(row["Toponymic Feature ID"], []).append(row)
pairs = [rows for rows in names_for.values() if len(rows) == 2]
one_french = [p for p in pairs if sum(looks_french(r["Geographical Name"]) for r in p) == 1]
print("features", len(names_for), "with two names", len(pairs))
print("pairs with exactly one French-looking name", len(one_french))
for p in pairs:
if p not in one_french:
print("check by hand:", [r["Geographical Name"] for r in p])
1. Group by the shared ID
Each official name is a row with its own CGNDB ID, but a single feature keeps one Toponymic Feature ID across its names. Grouping rows by that ID turns 11,663 names into 11,469 features and reveals the 194 that have two names, such as Madawaska River and Rivière Madawaska, or Richards Lake and Lac Richards.
2. Notice the missing label
The file has a Language column, but on 189 of the 194 pairs it says Undetermined for both names, so it cannot tell a student which name is French. That gap is the problem the script solves: it guesses from the text, using accented letters and a short list of French opening words such as Parc, Lac and Rivière.
3. Measure the rule
The rule settles 158 pairs and fails on 36. Grand Falls and Grand-Sault have no accent or listed opening word; New Brunswick and Nouveau-Brunswick likewise. Students list the failures, improve the rule and run it again, which is the honest way to build any classifier, AI included.
What the counts do not show
Two names in the file means two officially approved names, not a count of how often each is used. Most of the pairs are parks and protected areas, and 61 of them are Protected Natural Areas named on the same decision date, 21 March 2003, so the pairs are not a fair sample of the province's lakes, rivers and communities.
A follow-up for a Grade 12 student: rewrite the rule as a small scoring function that weighs several clues, then compare its errors with the first version. Contains information licensed under the Open Government Licence - Canada. Data: Natural Resources Canada, GeoBase Canadian Geographical Names, New Brunswick, CSV from the NRCan download folder. The counts, the rule and the script are ours, and Natural Resources Canada does not endorse this page.
