Extracting matches from strings with lookaround in R - r

Question

I have textual data (storytellings) and my aim is to extract certain words that are defined by a co-occurrence pattern, namely that they occur immediately prior to overlap, which is indicated by square brackets. The data are like this:
who <- c("Sue:", NA, "Carl:", "Sue:", NA, NA, NA, "Carl:", "Sue:","Carl:", "Sue:","Carl:")
story <- c("That’s like your grand:ma. did that with::=erm ",
"with Ju:ne (.) once or [ twice.] ",
" [ Yeah. ] ",
"And June wanted to go out and yo- your granny said (0.8)",
"“make sure you're ba(hh)ck before midni(hh)ght.” ",
"[Mm.] ",
"[There] she was (.) a ma(h)rried woman with a(h)- ",
"She’s a right wally. ",
"mm [kids as well ] ",
" [They assume] an awful lot man¿ ",
"°°ye:ah,°° ",
"°°the elderly do.°° ")
CAt <- data.frame(who, story)
Now, defining the pattern:
pattern <- "\\w.*\\s\\[[^]].*]"
and using grep():
grep(pattern, CAt$story, value = T)
[1] "with Ju:ne (.) once or [ twice.] "
[2] "mm [kids as well ] "
I get the two strings that contain the target matches but what I'm really after are the target words only, in this case the words "or" and "mm". This, to me, seems to call for positive lookahead. So I redefined the pattern thus:
pattern <- "\\w.*(?=\\s\\[[^]].*])"
which says something along the lines: "match the word iff you see a space followed by square brackets with some content on the right of that word". Now to extract only the exact matches, I normally use this code, which works fine as long as no lookaround is involved, but here it throws an error:
unlist(regmatches(CAt$story, gregexpr(pattern, CAt$story)))
Error in gregexpr(pattern, CAt$story) :
invalid regular expression, reason 'Invalid regexp'
Why is this? And how can the exact matches be extracted?

Answer 1

In your code, you could add perl=TRUE to gregexpr.
In your pattern \w.* will match a single word char followed by matching any char 0+ times.
This part \[[^]].*] will match [, then 1 char which is not ] and then .* which will match any char 0+ times followed by ].
You could update your pattern to repeating the word char and the character class itself instead.
\w+(?=\s\[[^]]*])
Explanation
\w+ Match 1+ word chars
(?= Positive lookahead, assert what is directly to the right is
\s Match single whitespace char
\[[^]]*] Match from opening[ to closing ] using a negated character class
) Close positive lookahead
Regex demo
Using doubled backslashes:
\\w+(?=\\s\\[[^]]*])
As an alternative you could use a capturing group instead of using a lookahead
(\w+)\s\[[^]]*]
Regex demo