I think it would be well possible to list all valid acronyms for the location property, if we limit ourself to the most useful acronyms and not try to achieve perfect Matlab compatibility here.

This is my suggestion for a list of those "useful acronyms":

# assumption: capital letters converted to small letters anyways
# strategy: put spaces only between proper words
# option: skip the "shortened" strings
strings_north = {"n", "north"};
strings_northoutside = {"no", "northoutside", "north outside", ...
   "northo", "northou", "northout", "northouts", "northoutsi", "northoutsid", ...    # <- skip those?
   "north o", "north ou", "north out", "north outs", "north outsi", "north outsid"... # <- skip those?
   };
strings_northeast = {"ne", "northeast", "north east"};
strings_northeastoutside = {"neo", "northeastoutside", ...
    "north east outside", "northeast outside", "north eastoutside", ...
    "northeasto", "northeastou", "northeastout", "northeastouts", "northeastoutsi", "northeastoutsid", ... # <- skip those?
    "north easto",  "north eastou", "north eastout", "north eastouts", "north eastoutsi", "north eastoutsid", ... # <- skip those?
    "north east o",  "north east ou", "north east out", "north east outs", "north east outsi", "north east outsid", ... # <- skip those?
    "northeast o",  "northeast ou", "northeast out", "northeast outs", "northeast outsi", "northeast outsid", ... # <- skip those?
    };
strings_northwest = {"nw", "northwest", "north west"};
strings_northwestoutside = {"nwo", "northwestoutside", ...
    "north west outside", "northwest outside", "north westoutside", ...
    "northwesto", "northwestou", "northwestout", "northwestouts", "northwestoutsi", "northwestoutsid", ... # <- skip those?
    "north westo",  "north westou", "north westout", "north westouts", "north westoutsi", "north westoutsid", ... # <- skip those?
    "north west o",  "north west ou", "north west out", "north west outs", "north west outsi", "north west outsid", ... # <- skip those?
    "northwest o",  "northwest ou", "northwest out", "northwest outs", "northwest outsi", "northwest outsid", ... # <- skip those?
    };
#west
#westoutside
#east
#eastoutside
#south
#southoutside
#southeast
#southeastoutside
#southwest
#southwestoutside
#best
#bestoutside
#none

questions:

• Do you think it is feasible to proceed with such a list of all acronym strings that we want to support? (If yes, I can generate the full list.)
• What do you think of the "shortened versions", like "northeastou", are they worth the effort of supporting them? (I tend to just skip them. They are always "...outside" strings, this concept is not usable with most of the strings anyways.)
• What would be the way to integrate this approach into the current input parsing code of legend.m? We would eventually like the location property to be "northwest", even when called via "nw" I assume. How should we do this?

Sorry, I might have written that code as a way to describe my findings from comment #2 without running it.

I thought the code would go in the update_legend_position subfunction:

diff -r e8e9f815945a scripts/plot/appearance/legend.m
--- a/scripts/plot/appearance/legend.m  Sun Apr 25 18:45:46 2021 +0200
+++ b/scripts/plot/appearance/legend.m  Mon May 10 21:43:37 2021 +0100
@@ -1417,6 +1417,24 @@
   persistent vmargin = 6;

   location = get (hl, "location");
+
+  locations = {"East", "EastOutside", "North", "NorthEast", ...
+    "NorthEastoOutside", "NorthOutside", "NorthWest", "NorthWestOutside", ...
+    "South", "SouthEast",  "SouthEastOutside", "SouthOutside", "SouthWest", ...
+    "SouthWestOutside", "West",  "WestOutside"};
+  acronyms = cellfun (@(x) tolower (x(x>='A' & x<='Z')),  locations, ...
+    "UniformOutput", false);
+  locations = tolower (locations);
+  acr2loc = cell2struct (locations, acronyms, 2);
+  if (isfield (acr2loc, location))
+    location = acr2loc.(location);
+  endif
+  location = location(~isspace (location));
+  idx = find (strncmp (location, locations, numel (location)));
+  if (isscalar (idx))
+    location = locations{idx};
+  endif
+
   outside = strcmp (location(end-3:end), "side");
   if (outside)
     location = location(1:end-7);

but that's not enough and you are probably right it should go in parse_opts. I will look into it.

No sure what is the cleanest way to implement this. What about:

location = get (hl, "location");

locations = {"North", "East", "South", "West", "NorthEast", "SouthEast", "SouthWest", "NorthWest"}; # ...
acronyms = cellfun (@(x) tolower (x(x>='A' & x<='Z')),  locations, "UniformOutput", false);
locations = tolower (locations);
acr2loc = cell2struct (locations, acronyms, 2);
try
  location = acr2loc.(location);
end_try_catch
location = location(~isspace (location));
idx = strmatch (location, locations);
if (isscalar (idx))
  location = locations{idx};
endif

Matlab (e.g. R2019b) also accepts "location" properties with spaces. The following works in Matlab, but currently throughs an error in Octave: legend('my text', 'location', 'south east');

I was about to file a new bug report of the type "incompatibility", because this Octave behavior prevented me to run a Matlab script from someone else. Shouldn't we raise the importance of the "feature request"?