Using gsub to separate a sting into cities and counties

I am trying to split a string into cities and countries but having difficulty when the city or country in question is more than one word (Eg, aix-en-provence or United States). The current code I am using a will work for most like Paris, France but not for ones similar to those above.

 locations
 paris_france
 miami_united states
 new york_united states
 aix-en-provence_france
 auckland_new_zealand

current code used
city = gsub("([A-z]+)_([A-z]+)", "\", locations)
country = gsub("([A-z]+)_([A-z]+)", "\", locations)

so now city will return paris and country will be france which is fine but for other stuff like auckland and zealand will be returned. Guessing its obviously a case of getting it to recognise more than one word before or after the '_'

Because of new_zealand, we have to take a little extra caution.

base R

strcapture("^([^_]+)_(.*)$", locs$locations, proto = c(city="", country=""))
#              city       country
# 1           paris        france
# 2           miami united states
# 3        new york united states
# 4 aix-en-provence        france
# 5        auckland   new_zealand

tidyr

library(tidyr)
separate_wider_delim(locs, locations, delim = "_", names = c("city", "country"), too_many = "merge")
# # A tibble: 5 × 2
#   city            country      
#   <chr>           <chr>        
# 1 paris           france       
# 2 miami           united states
# 3 new york        united states
# 4 aix-en-provence france       
# 5 auckland        new_zealand  

Data

locs <- structure(list(locations = c("paris_france", "miami_united states", "new york_united states", "aix-en-provence_france", "auckland_new_zealand")), row.names = c(NA, -5L), class = "data.frame")