you can use trans method:

my $book1 = "Don Quixote- Miguel de Cervantes";
my $book2 = "Les Misérables -Victor Hugo";
my $book3 = "War and Peace - Leo Tolstoy";

for ($book1, $book2, $book3) -> $b {
    say $b.trans([/<wb> '- '/, /' -' <wb>/] => [' - ']);
}

wb is a word boundary.

TL;DR Another option to consider is using the <( and )> capture markers to pick out just the bit you want to replace.

A "literal" interpretation of your Q

Matching strictly per your examples:

/   \C[space]   <(   '- '   |   ' -'   )>   \C[space]   /

• The syntax \c[...] specifies one or more characters by using their Unicode names inside the square brackets (in this case the classic ASCII space character).¹

  In this pattern I've used \C[...] (uppercase C, not lowercase c). There is a range of Raku "backslash" atoms and they all have lowercase and uppercase variants, where the uppercase variant matches any character except the one(s) matched by the lowercase variant. So \C[space] matches any character other than the ASCII space character. See \c / \C for more info.

• The <( capture marker marks the start point of the regex's capture. Likewise )> marks the endpoint.

  Without them, when the pattern matches, the whole match would be captured, which would include whatever non whitespace character matches the \C[space] atom. We don't want that. So we use these markers to restrict what we capture.

  Btw, each marker is independent. The above pattern matches \C[space] '- ' or '- ' \C[space]. If the pattern to the left of the | matches, only the <( has an impact, omitting whatever matched \C[space], and capturing until the end of the match, which for this pattern stops at the |. If the pattern to the right matches, capturing starts immediately after the | and ends at the )>.

• The | is Raku's parallel (aka "longest token match" -- LTM) pattern alternation operator, an alternative to the traditional sequential pattern alternation operator (which in Raku is written ||). In this case the set of substrings that the two operators will and won't match is the same, so it makes no difference which is used. But | is shorter than ||; when the match set is the same it's typically faster; and when the match sets are different it's often | that's desirable. So I use it by default unless I know I need the traditional sequential alternation logic (try pattern on left of || first; if that fails, try the pattern on the right of the ||).

A "per its spirit?" interpretation of your Q

Matching more flexibly regarding whitespace:

/   \S   <(   '-' \s+   |   \s+ '-'   )>   \S   /           

• The \S atoms match any character that is not categorized by Unicode as being a whitespace character. (I use Raku, or tools such as this character property lookup web page, to explore what Unicode makes of a character.)

  Comparing \C[space], \S, and <wb>:

  • \C[space] matches any character, including whitespace characters, with the sole exception of an ASCII space. My guess is it'll be the fastest of the three.

  • \S matches any non-whitespace. My guess is it'll be faster than <wb>.

  • <wb> matches between characters. Also it'll match before the first character in a string, and after the last one. So @chenyf's pattern would match and change '- foo...' to ' - foo...' and '...bar -' to '...bar - ' whereas the patterns with \C[space] or \S would not match at the start/end of those strings.

• The \s+ atoms match one or more whitespace characters.

Footnotes

¹ The naming is case insensitive. Multiple characters are separated by commas. \c[...] also works in a double quoted string (but not \C[...]).

for ($book1, $book2, $book3, $book4, $book5, $book6) -> $b 
  { say $b
    .subst(/ \S <( (\-+) \h   )> \S /, {" $0 "}, :global) 
    .subst(/ \S <(  \h  (\-+) )> \S /, {" $0 "}, :global)
    .subst(/ \S <( (\-)  \v   )> \S /,   {"$0"}, :global) #fixes hyphenated words w/embedded newlines
}

Sample Input:

my $book1 = "Don Quixote- Miguel de Cervantes";
my $book2 = "Les Misérables -Victor Hugo";
my $book3 = "War and Peace - Leo Tolstoy";
my $book4 = "Moby-Dick; or, The Whale- Herman Melville";
my $book5 = "Winnie-the-Pooh --A. A. Milne";
my $book6 = "Slaughterhouse-\nFive- Kurt Vonnegut";

Sample Output:

Don Quixote - Miguel de Cervantes
Les Misérables - Victor Hugo
War and Peace - Leo Tolstoy
Moby-Dick; or, The Whale - Herman Melville
Winnie-the-Pooh -- A. A. Milne
Slaughterhouse-Five - Kurt Vonnegut

For this problem I would probably start by asking how these erroneous entries found their way into the data at hand. Was it the product of concatenation? Or informal (manual) entry? The first is fixable, the second might be a primary application of the Raku programming language (i.e. making informal, manual text entries more formal). This answer follows the excellent examples already posted, but (in contrast) uses a $0 capture to re-position the "-" field separator. In brief:

1. The first .subst(...) command globally captures one-or-more hyphens when followed by a single horizontal whitespace, and places the equivalent number of hyphens between the Title and Author (hyphens surrounded by whitespace).

2. The second .subst(...) command globally captures one-or-more hyphens when preceded by a horizontal whitespace, and places the equivalent number of hyphens between the Title and Author (hyphens surrounded by whitespace).

3. The third .subst(...) command globally captures a single hyphen when followed by a single vertical whitespace (e.g. newline), and removes the vertical whitespace. Hyphens followed by horizontal whitespace remain untouched. Note, for this the third .subst(...) command, the replacement can simply be written as "-" (i.e. no need to use $0).

Note: the first two .subst statements can be combined with | OR:

.subst(/ \S <( (\-+) \h  | \h  (\-+) )> \S /, {" "~$0~" "}, :global)

Why go to all this trouble? Well, the first reason is that a more 'pedestrian' approach is more robust to complicated input (e.g. hyphenated words). In fact, some answers already posted may not handle hyphenated book titles and/or author names, which are handled gracefully (above and below, note alternate replacement form):

~$ cat book_author.txt
Don Quixote- Miguel de Cervantes
Les Misérables -Victor Hugo
War and Peace - Leo Tolstoy
Moby-Dick; or, The Whale- Herman Melville
Winnie-the-Pooh --A. A. Milne
Slaughterhouse-
Five- Kurt Vonnegut
~$ cat book_author.txt | raku -e 'say lines.join("\n")
      .subst(/ \S <( (\-+) \h  )> \S /, {" "~$0~" "}, :global)
      .subst(/ \S <( \h  (\-+) )> \S /, {" "~$0~" "}, :global)
      .subst(/ \S <( \-   \v   )> \S /,  "-", :global);'
Don Quixote - Miguel de Cervantes
Les Misérables - Victor Hugo
War and Peace - Leo Tolstoy
Moby-Dick; or, The Whale - Herman Melville
Winnie-the-Pooh -- A. A. Milne
Slaughterhouse-Five - Kurt Vonnegut

The second reason is such an answer can be used to modify text with other separators, such as Title | Author data, wherein title is separated from author by a vertical bar. The third reason is capturing (e.g. using $0) is adapted to a wide variety of problems, such as making multiple identical separator characters like -- or || into single-character separators (note yet another way of writing the replacement, this time adding .comb[0]):

~$ cat book_bar_author.txt
Don Quixote| Miguel de Cervantes
Les Misérables |Victor Hugo
War and Peace | Leo Tolstoy
Moby-Dick; or, The Whale| Herman Melville
Winnie-the-Pooh ||A. A. Milne
Slaughterhouse-
Five| Kurt Vonnegut
~$ cat book_bar_author.txt | raku -e 'say lines.join("\n")
      .subst(/ \S <( (\|+) \h  )> \S /, {"",$0.comb[0],""}, :global)
      .subst(/ \S <( \h  (\|+) )> \S /, {"",$0.comb[0],""}, :global)
      .subst(/ \S <( \-   \v   )> \S /,  "-", :global);'
Don Quixote | Miguel de Cervantes
Les Misérables | Victor Hugo
War and Peace | Leo Tolstoy
Moby-Dick; or, The Whale | Herman Melville
Winnie-the-Pooh | A. A. Milne
Slaughterhouse-Five | Kurt Vonnegut