How do I combine text names within an ordered transcription of dialogue?

Say I have this data:

df <- data.frame(x = c("Tom: I like cheese.", 
                       "Tom: Cheese is good.", 
                       "Tom: Muenster is my favorite.", 
                       "Bob: No, I like Cheddar.", 
                       "Tom: You're wrong. I think cheddar is only good on burgers.", 
                       "Gina: But what about American on burgers?", 
                       "Gina: That's better.", 
                       "Bob: Yeah, I agree with Gina.", 
                       "Bob: American is better on burgers. Cheddar is for grating on nachos."))

I want to turn it into this data:

df <- data.frame(x = c("Tom: I like cheese. Cheese is good. Muenster is my favorite.", 
                       "Bob: No, I like Cheddar.", 
                       "Tom: You're wrong. I think cheddar is only good on burgers.", 
                       "Gina: But what about American on burgers? That's better.", 
                       "Bob: Yeah, I agree with Gina. American is better on burgers. Cheddar is for grating on nachos."))

Basically, I want to cut the text including and before the colon on any instance of text that already has had a recent name.

I am struggling with trying to figure out how to do it in a way that doesn't group the entire "Tom:"'s and "Gina:"'s together and remove them all but for the first instance. I want the later mentions of names to restart the loop.

We can use tidyr to split the speaker and what they say into columns, then use dplyr to combine runs of the same speaker. For example

df |> 
  tidyr::separate_wider_delim(x, ": ", names=c("speaker", "words")) |>
  mutate(instance = consecutive_id(speaker)) |>
  summarize(speaker = first(speaker), text=paste(words, collapse=" "), .by=instance)

returns

  instance speaker text                                                                                 
     <int> <chr>   <chr>                                                                                
1        1 Tom     I like cheese. Cheese is good. Muenster is my favorite.                              
2        2 Bob     No, I like Cheddar.                                                                  
3        3 Tom     You're wrong. I think cheddar is only good on burgers.                               
4        4 Gina    But what about American on burgers? That's better.                                   
5        5 Bob     Yeah, I agree with Gina. American is better on burgers. Cheddar is for grating on na…

library(stringr)
library(dplyr)

df |>
  group_by(line = consecutive_id(str_extract(x, "^\\w+"))) |>
  reframe(x = str_remove_all(str_flatten(x, " "), "\\s\\w+:"))

Output

  line x                                                                                             
 <int> <chr>                                                                                         
     1 Tom: I like cheese. Cheese is good. Muenster is my favorite.                                  
     2 Bob: No, I like Cheddar.                                                                      
     3 Tom: You're wrong. I think cheddar is only good on burgers.                                   
     4 Gina: But what about American on burgers? That's better.                                      
     5 Bob: Yeah, I agree with Gina. American is better on burgers. Cheddar is for grating on nachos.

Using data.table, split on ": ", group by relid, then paste it back per group:

df[, c("name", "text") := tstrsplit(x, ": ", fixed = TRUE) 
   ][, .(text = paste(text, collapse = " ")), by = .(name, rleid(name))
     ][, -2]

#      name                                                                                  text
#    <char>                                                                                 <char>
# 1:    Tom                                 I like cheese. Cheese is good. Muenster is my favorite.
# 2:    Bob                                                                     No, I like Cheddar.
# 3:    Tom                                   You're wrong. I think cheddar is only good on burgers.
# 4:   Gina                                       But what about American on burgers? That's better.
# 5:    Bob Yeah, I agree with Gina. American is better on burgers. Cheddar is for grating on nachos.

Here is a base R option

with(
    sort_by(
        aggregate(
            V2 ~ nr + V1,
            transform(
                read.delim(
                    text = df$x,
                    sep = ":",
                    head = FALSE,
                    strip.white = TRUE
                ),
                nr = with(rle(V1), rep(seq_along(values), lengths))
            ),
            paste0,
            collapse = " "
        ), ~nr
    ),
    data.frame(
        x = paste0(V1, ": ", V2)
    )
)

which gives

                                                                                               x
1                                   Tom: I like cheese. Cheese is good. Muenster is my favorite.
2                                                                       Bob: No, I like Cheddar.
3                                    Tom: You're wrong. I think cheddar is only good on burgers.
4                                       Gina: But what about American on burgers? That's better.
5 Bob: Yeah, I agree with Gina. American is better on burgers. Cheddar is for grating on nachos.