How to use alternation with lookbehind

Question

Aim:

I would like to match sentences with the word 'no' but only if 'no' is not preceeded by 'with' or 'there is' or 'there are' in r.

Input:

The ground was rocky with no cracks in it
No diggedy, no doubt
Understandably, there is no way an elephant can be green

Expected output:

The ground was rocky with no cracks in it
Understandably, there is no way an elephant can be green

Attempt:

gsub(".*(?:((?<!with )|(?<!there is )|(?<!there are ))\\bno\\b(?![?:A-Za-z])|([?:]\\s*N?![A-Za-z])).*\\R*", "", input_string, perl=TRUE, ignore.case=TRUE)

Problem:

The negative lookbehind seems to be ignored so that all the sentences are replaced. Is the problem the use of alternation in a the lookbehind statement?

Answer 1

You may use

(?mxi)^       # Start of a line (and free-spacing/case insensitive modes are on)
(?:           # Outer container group start
  (?!.*\b(?:with|there\h(?:is|are))\h+no\b) # no 'with/there is/are no' before 'no'
  .*\bno\b  # 'no' whole word after 0+ chars
  (?![?:])    # cannot be followed with ? or :
|             # or
  .*          # any 0+ chars
  [?:]\h*n(?![a-z]) # ? or : followed with 0+ spaces, 'n' not followed with any letter
)             # container group end
.*            # the rest of the line and 
\R*           # 0+ line breaks

See the regex demo. In short: the pattern finds 2 alternatives, either of the 2 types of lines, one with no whole word in it that is not preceded with with, there is or there are and a space after them, or a line that contains ? or : followed with 0+ horizontal spaces (\h) and then an n not followed with any other letter.

See the R demo:

sentences <- "The ground was rocky with no cracks in it\r\nNo diggedy, no doubt\r\nUnderstandably, there is no way an elephant can be green"
rx <- "(?mxi)^ # Start of a line
(?:            # Outer container group start
  (?!.*\\b(?:with|there\\h(?:is|are))\\h+no\\b) # no 'with/there is/are no' before 'no'
  .*\\bno\\b   # 'no' whole word after 0+ chars
  (?![?:])     # cannot be followed with ? or :
|              # or
  .*           # any 0+ chars
  [?:]\\h*n(?![a-z]) # ? or : followed with 0+ spaces, 'n' not followed with any letter
)              # container group end
.*             # the rest of the line and 0+ line breaks
\\R*"
res <- gsub(rx, "", sentences, perl=TRUE)
cat(res, sep="\n")

Output:

The ground was rocky with no cracks in it
Understandably, there is no way an elephant can be green

Thanks to the x modifier, you may add comments to the regex pattern and use spaces to format it for better readability. Note that all literal whitespace must be replaced with \\h (horizontal whitespace), \\s (any whitespace), \\n (LF), \\r (CR), etc. to make it work in such a pattern.

The (?i) modifier stands for the ingore.case=TRUE.

Answer 2

You just need to have a regex alternation character. Idea is to match and capture all the possible "no" sentences and match all the remaining sentences. Then replace all the matched characters with \\1 ie, characters from the first capturing group.

gsub("(?i)(.*(with|there (?:is|are)) no\\b.*)|.*", "\\1" ,string, perl=T)

DEMO

Example:

x <- "The ground was rocky with no cracks in it\nNo diggedy, no doubt\nUnderstandably, there is no way an elephant can be green"
gsub("(?i)(.*(with|there (?:is|are)) no\\b.*\\n?)|.*\\n?", "\\1" ,x, perl=T)
# [1] "The ground was rocky with no cracks in it\nUnderstandably, there is no way an elephant can be green"