Guessing a bit at what you mean by "in context", but here's another approach (Update: slightly buggy: see update below):

>perl -wMstrict -le
"my $s = 'a X bar XX c XXX d XXXX e XXXXX foo XXXXXX g';
 ;;
 my @tokens = qw(a bar c d e foo g);
 my ($token) = map qr{ \b $_ \b }xms, join q{|}, @tokens;
 my $max = 3;
 ;;
 while ($s =~ m{ ( .{0,$max} $token) (?= (.{0,$max})) }xmsg) {
   print qq{'$1$2'};
   }
"
'a X '
' X bar XX'
'XX c XX'
'XX d XX'
'XX e XX'
'XX foo XX'
'XX g'
[download]

Update: This version fixes a bug in the definition of  $token that allowed  'de' to be matched as a token (or part of one, anyway), and also makes match sub-pattern extraction clearer for demonstration purposes.

>perl -wMstrict -le
"my $s = 'a X bar c XXX d XXXX e XXX de XXX foo XXXXXX g h';
 ;;
 my @tokens = qw(a bar c d e foo g h);
 my ($token) = map qr{ \b (?: $_) \b }xms, join q{|}, @tokens;
 my $pre_max  = 3;
 my $post_max = 3;
 my $pre  = qr{ .{0,$pre_max}? }xms;  ## fixed -- see update below
 my $post = qr{ .{0,$post_max} }xms;
 ;;
 while ($s =~ m{ ($pre) ($token) (?= ($post)) }xmsg) {
   my ($before, $tok, $after) = ($1, $2, $3);
   print qq{'$before$tok$after'   :$before:$tok:$after:};
   }
"
'a X '   ::a: X :
' X bar c '   : X :bar: c :
' c XX'   : :c: XX:
'XX d XX'   :XX :d: XX:
'XX e XX'   :XX :e: XX:
'XX foo XX'   :XX :foo: XX:
'XX g h'   :XX :g: h:
' h'   : :h::
[download]

Update: Another bug: The regex
    my $pre = qr{ .{0,$pre_max} }xms;
misses token 'c' in  'b c' (for tokens 'b' and 'c') if 'c' is within the 'span' of context characters. Replace with
    my $pre = qr{ .{0,$pre_max}? }xms;