A Game of Words (Part 2)

This is the second blog post in the series and again I will be analysing transcripts from the first seven seasons of GoT.

In the first post, I looked at the vocabulary and verbosity of certain characters from the show.

Having analysed the number of words and most common words they say, I’m going to go a step further and look at the nature of their words.

To do this I’ll follow up last week’s look at what the characters refer to and instead look at who the characters refer to.

Then I will apply some simple sentiment analysis to see how the emotion behind their words can describe their personality and storyline.

Sentiment analysis in this post was done in R using the sentimentr and tidytext packages.

The data is the same as described in my previous post.

Joffrey this, Joffrey thatTo start with I will continue in a similiar fashion to last week’s article, but instead of analysing every word I have refined it down to just the times each character has mentioned another character.

This was tricky as I had to make judgement calls on what to include (i.

e.

‘Khaleesi’ and ‘Mother of Dragons’ referring to Dany) and what not to include (familial references – i.

e sometimes the word ‘father’ refers to a specific person but other times it does not).

The table below shows the person that was most frequently referenced by each character.

The most referenced person by each of our main characters, along with the proportion of all references that the character accounts for.

Also the proportion of all words spoken by that character that their most referenced person accounts for.

Some of you mentioned that Dany’s most common words last week were ‘all about her’ and this pattern again holds true here.

The reason behind this is Dany’s habit of indignantly proclaiming her own name and numerous titles when she wishes to make a point.

In contrast, Theon’s apparent self-obsession is more to do with his ‘Prince Theon – Reek – Theon’ journey and his constant struggle with his own identity.

I think this contrasts nicely with the point above about Dany, who has always been certain of her identity and her birthright.

It seems that Joffrey generated the most discussion amongst other characters, but for different reasons: Cersei mostly referred to him in a motherly or defensive way; The Hound often used the young tyrant’s name when relaying orders; Sansa was the main victim of Joffrey’s sadistic tendencies; and Arya had him on her ‘list’.

Jon is interesting – he seems to not reference other (human) characters by name very often, possibly a side effect of being at the Wall away from all the other main characters for so long.

Clearly Mance was somebody who made an impression on him.

(Note: The Night King was not included as a character in this analysis).

I think this reflects incredibly well who the most important person is to these characters.

Next up I’m going to go back to analysing each character’s words but this time look at the sentiment behind them — specifically how positive or negative they are.

First — a brief technical backgroundFor those of you who don’t know what sentiment analysis is, it is a way of analysing the general emotion of some text.

You can do this by taking each word in a sentence and saying whether that word is ‘positive’, ‘negative’ or neutral (see some examples below, taken from the dataset).

Some example words with their respective sentimentYou can then aggregate by looking at each word in a sentence to determine whether that sentence is positive or negative.

This is the simplest approach and is also the most flawed.

Take for example the sentence “I don’t like you, you are not nice”, a simple sentiment analysis algorithm would see the words ‘like’ and ‘nice’ and assume this was a positive sentence.

To get around this — without going into too much detail — I used the R package sentimentr as it allows easy sentiment analysis at a sentence-level.

It has inbuilt methods for dealing with negators (“not”, “don’t”, “didn’t” etc.

 — like in the example above) as well as some other linguistic nuances such as amplifiers/deamplifiers (“very”, “barely” etc.

) and adversative conjunctions (“but”, “whereas” etc.

).

While still by no means perfect, over a large enough dataset it allows me to quickly and easily assign a sentiment score to each character’s lines.

To be clear, throughout the rest of this post you should think of ‘positive’ as meaning polite, friendly, or enthusiastic and ‘negative’ as rude, dismissive or violent.

Let’s begin:Tell me to my faceFirst, I applied the sentiment analysis approach mentioned above to look at each character’s overall ‘Sentiment score’ (average sentiment of all sentences multiplied by 100 for readability) – a Sentiment score greater than zero implies their speech is more often positive than negative, and vice versa for scores less than zero.

As well as this you can see the proportion of their sentences that the algorithm classified as ‘positive’ and those it classified as ‘negative’.

The results were…surprising, at first (stars denote particularly high or low scorers).

Table showing the overall sentiment score for each of our main characters, along with the proportion of their sentences that were positive (sentiment score > 10 and the proportion that were negative (sentiment score < 10).

Euron being the most ‘positive’ person caught me off guard initially but when you think about it, investigate a bit more and consider his words at face value without any underlying motives or feelings, it starts to make more sense.

He is generally enthusiastic about everything (to an almost maniacal degree) and much of his time so far has been spent trying to charm Cersei.

That is to say he mostly uses positive words in a negative (depending on who you are) context.

Littlefinger (Petyr Baelish) was a bit different, his high score appeared to come from his gracious, political side.

For example, the algorithm classifies “Your grace” as a very positive sentence, something which Littlefinger has said frequently.

This classification makes sense to me as it is an acknowledgement of rank, so looking purely at the words – and not the scheming and plotting behind it – Littlefinger comes in as the second most positive character.

The Hound (Sandor Clegane) was a bit less surprising.

Add his colourful language (profanity is always ‘negative’) to his perpetual pessimism and Sandor claiming the title of ‘Most Negative Character’ was never really in any doubt.

For those with a sentiment score closer to zero it is harder to say why that is or whether any conclusions can be drawn (for example Theon and Yara could quite easily have positive overall scores if one or two sentences were scored differently).

To me, these results are indicative of what you could arguably describe as a correlation between positive sentiment and dishonesty.

In general, those Game of Thrones characters who say the nicest things tend to be hiding a sinister ulterior motive, whereas those who don’t say ‘nice’ things (Hello, Hound) are simply being honest about how much they don’t like you.

This connection doesn’t always hold though, as clearly Samwell isn’t dishonest – his relatively high score is due to the fact that he is simply incredibly polite.

On a side note, Arya and Bran had the highest proportion of neutral sentences – which I think fits nicely with the identities they came to adopt (No One and the Three Eyed Raven), both of which are often emotionless.

Strong women, strong wordsNext up I’m going to look at how average sentiment has changed over the course of the first 7 seasons.

You can see this below for each of the main characters.

Note that when analysing these graphs, the values don’t matter as much as the patterns.

Graphs showing how each character’s average sentiment has changed across seasons 1–7.

As you can see, The Hound is The Hound.

He really doesn’t change.

The Theon-Reek dichotomy that was present in last week’s analysis again shows here, as well as his resurgence (of sorts) in Season 7.

Tyrion, Littlefinger and Varys – arguably the three biggest schemers – all show consistently positive sentiment scores.

I think there is an interesting parallel between Cersei and Sansa in Season 7.

It was a significant season for both of them, as they both came into positions of power and began to more openly show their strength.

Sansa as the Lady of Winterfell became more rough with her words, finally having someone she could be completely honest with (Jon), as well as finding herself in a position of power over people she used to have to rely on (Littlefinger).

At the same time, Cersei as Queen became less concerned with how she spoke to people – as she now has all the power and nothing to lose.

All of this meant that their words became more honest, more abrasive, and as result, more ‘negative’.

Analysis is comingThanks for reading and if you haven’t checked out the first post in this series please do so!.Episode 3 of this Game of Thrones analysis series is due to be coming in the next couple of weeks, but the themes (and possibly dataset) will likely differ from what we’ve seen so far…Please share with anyone you think might be interested and comment/clap if you enjoyed reading!.. More details

Leave a Reply