The Rate of Profanity Caveat

It is worth reminding ourselves that the number of episodes per series has varied over Taskmaster’s brodcast run. Consequently, to compare the profanity use between series, we cannot compare the total profanity used by a contestant (or speaker in general) in a series as contestants from longer series will likely appear as more foul mouthed.

In order to allow for valid comparisons between series, we introduce a new metric, namely the profanity rate which normalises the profanity total by the number of episodes in the series.

\[\texttt{Profanity Rate for Contestant C in series S} = \frac{\sum{\texttt{Profanity by contestant C in series S}}}{\texttt{Number of episodes in series S}} \] Profanity Rate can be thought of as the average number of times contestant C will swear in an episode (of series S).

The Next Level of Profanity

To assist our foul mouthed quest, it would be useful to create new temporary subtables which combine, transform and/or aggregate data from the various database tables we outlined above.

For example, we might want to create:

• an enhanced version of the profanity table which contains interpretable information on the speaker and series rather than numerical ids (they are people with names, not numbers).
• an aggregate of this enhanced profanity table (which is at a series, speaker, task level granularity), such that we have profanity at a series level for a contestant.

These transformations can be done in SQL or R. Based on personal preference, the former will be used for joins and aggregations, and the latter for more technical transformations (for example calculation of new statistics).

--  Stored as an R dataframe profanity_enh

SELECT 
pf.series, 
pf.episode, 
pf.task,
pf.speaker as speaker_id,
pp.name as speaker_name,
pf.roots,
pf.quote,
pf.studio,
pp.gender,
pp.hand,
pp.champion,
pp.tmi as speaker_tmi,
sp.name as series_name,
sp.episodes as num_episodes_in_series,
sp.champion as series_champion_id,
sp.special
FROM profanity pf
LEFT JOIN people pp
    ON (pf.speaker = pp.id
    AND pf.series = pp.series) OR
(pf.speaker = pp.id)
LEFT JOIN series sp
    ON pf.series = sp.id

Series Profanity

The following table takes the recently created profanity_enh dataframe and performs a number of operation which eventually results in a new dataset, series_profanity at a series, contestant level. Operations include:

1. Counting the number of profanities uttered in a given quote.
2. Aggregating data to a series and speaker level.
   1. Sum of the profanities uttered.¹
   2. Number of distinct episodes that the profanities are uttered over.
   3. Number of episodes in the series.
3. Adding a new column which calculate the profanity rate.

library(reticulate)
library(dplyr)

series_profanity <- profanity_enh %>% 
    rowwise() %>%
    mutate(num_profanity = length(reticulate::py_eval(roots))) %>%
    # To count the number of profanities utter in a quote.
    group_by(series, series_name, special, speaker_id, speaker_name, speaker_tmi, gender, hand) %>%
    # Aggregating and summarising data at a series, speaker level.
    summarise(
        speaker_episode_count = dplyr::n_distinct(episode),
        sum_profanity_series = sum(num_profanity), 
        no_episodes_in_series = max(num_episodes_in_series)
    ) %>%
    mutate(profanity_per_episode = sum_profanity_series/no_episodes_in_series)

A Close Finish?

Some of you readers may be interested in knowing wheter it was a close finish into the profanity rate race.

We can quickly determine this may changing the top_n function to consider 5 rather than 1 say, when selecting based on profanity rate.

There are no big surprises in these finishing positions although Asim Chaudhry being a close second is somewhat surprising since I don’t remember him being a particularly angry or foul mouthed incident with him³. He’s a relatively mild mannered comedy actor who just wants everyone to know he is a vegan.

Another observation is that the top 5 are all male and right handed. Make of that what you will…

Bonus Task: Foulest Mouth in Each Series

To find the foulest mouth in each Taskmaster series, we can continue use the existing data and logic we have used thus far, but introduce an additional line of logic to rank within a series; the group_by function is our friend here. The rank function provides a ranking with respect to profanity rate; for descending order ranking a minus sign is introduced on the variable we want to rank according to.

within_series_profanity <- series_profanity %>%
    filter(special == 0 & !(speaker_name %in% c("Greg Davies", "Alex Horne"))) %>%
    arrange(series, -profanity_per_episode, -speaker_episode_count) %>%
    group_by(series, series_name) %>%
    mutate(profanity_rank = rank(-profanity_per_episode, ties.method = "first"), name_prof_rate = sprintf("%s (%#.3f)", speaker_name, profanity_per_episode))

Performing the filter operation of profanity_rank = 1 will provide the foulest contestant by series:

Figure 1: Foulest Contestant by Series

Within Series Foul Mouthed Races

Similar to the overall foul mouthed analysis, we might be interested in seeing how close the profanity race was in each series. We may also find our surprising insights, less surprising by assessing the profanity race.

Figure 2: All Series Contestants Profanity Rates