#AppellateTwitter wasted no time criticizing an experienced SCOTUS advocate for his recent brief’s word choice:
But wait. Nearly one year (!) after the Court invited her views, the Solicitor General has delivered what can only be described as a hot mess of a brief. In it, the government claims to have unearthed a new obstacle to the Court’s consideration of the question presented. Specifically, the government asserts that, because respondents’ claims do not “arise out of” the military’s combatant activities within the meaning of the combatant-activities exception in the Federal Tort Claims Act (FTCA), this case is an unsuitable vehicle in which to address the question.
This real-time example offered the perfect time to see whether and how ChatGPT, Bard, and Bing can measure the written word’s sentiment, emotion, and tone.
tl;dr They can!
The combined prompt:
Here is one paragraph from a court brief.
Divide the paragraph into sentences.
Treat each sentence as “Original” text that will appear in rows in a four-column table.
Here is the text:
“But wait. Nearly one year (!) after the Court invited her views, the Solicitor General has delivered what can only be described as a hot mess of a brief. In it, the government claims to have unearthed a new obstacle to the Court’s consideration of the question presented. Specifically, the government asserts that, because respondents’ claims do not “arise out of” the military’s combatant activities within the meaning of the combatant-activities exception in the Federal Tort Claims Act (FTCA), this case is an unsuitable vehicle in which to address the question.”
Reply to me with the text-based rows in a four-column table.
The first column is the original text and labeled “Original”.
In the second column, list the sentiment for that sentence. Label the second column “Sentiment”.
In the third column, list the emotion for that sentence. Label the third column “Emotion”.
In the fourth column, list the tone for that sentence. Label the fourth column “Tone”.
After the table, list and explain what “sentiment” “emotion” and “tone” each mean.
ChatGPT’s response:
| Original | Sentiment | Emotion | Tone |
|---|---|---|---|
| But wait. | Neutral | None | Informative |
| Nearly one year (!) after the Court invited her views, the Solicitor General has delivered what can only be described as a hot mess of a brief. | Negative | Frustration | Critical |
| In it, the government claims to have unearthed a new obstacle to the Court’s consideration of the question presented. | Neutral | None | Informative |
| Specifically, the government asserts that, because respondents’ claims do not “arise out of” the military’s combatant activities within the meaning of the combatant-activities exception in the Federal Tort Claims Act (FTCA), this case is an unsuitable vehicle in which to address the question. | Negative | Frustration | Critical |
Sentiment refers to the attitude or opinion conveyed by the text, whether positive, negative, or neutral.
Emotion refers to the feelings or mood conveyed by the text, such as anger, frustration, or joy.
Tone refers to the writer’s attitude towards the subject matter and the audience. The tone can be formal, informal, serious, sarcastic, or humorous.
Google Bard has thoughts:
Sure, here is the table with the four columns:
| Original | Sentiment | Emotion | Tone |
|---|---|---|---|
| But wait. | Neutral | Neutral | Neutral |
| Nearly one year (!) after the Court invited her views, | Negative | Anger | Disapproving |
| the Solicitor General has delivered what can only be described as a hot mess of a brief. | Negative | Anger | Disapproving |
| In it, the government claims to have unearthed a new obstacle to the Court’s consideration of the question presented. | Negative | Anger | Disapproving |
| Specifically, the government asserts that, because respondents’ claims do not “arise out of” the military’s combatant activities within the meaning of the combatant-activities exception in the Federal Tort Claims Act (FTCA), | Negative | Anger | Disapproving |
| this case is an unsuitable vehicle in which to address the question. | Negative | Anger | Disapproving |
Sentiment refers to the overall attitude or feeling expressed in a piece of text. It can be positive, negative, or neutral.
Emotion refers to a specific feeling that is expressed in a piece of text. It can be anger, sadness, happiness, fear, or any other emotion.
Tone refers to the overall mood or atmosphere of a piece of text. It can be formal, informal, serious, humorous, or any other tone.
In the table above, the sentiment of all the sentences is negative. This is because the text expresses a feeling of anger and disapproval towards the Solicitor General’s brief. The emotion of the text is also anger. The tone of the text is disapproving. This is because the text uses strong language and rhetorical devices to express its negative sentiment.
And so does Microsoft Bing:
Here is the table you requested:
| Original | Sentiment | Emotion | Tone |
|---|---|---|---|
| Nearly one year (!) after the Court invited her views, the Solicitor General has delivered what can only be described as a hot mess of a brief. | Negative | Disgust | Sarcastic |
| In it, the government claims to have unearthed a new obstacle to the Court’s consideration of the question presented. | Neutral | None | Informative |
| Specifically, the government asserts that, because respondents’ claims do not “arise out of” the military’s combatant activities within the meaning of the combatant-activities exception in the Federal Tort Claims Act (FTCA), this case is an unsuitable vehicle in which to address the question. | Negative | Disgust | Informative |
Sentiment refers to whether a text expresses a positive or negative opinion or attitude towards something or someone. Emotion refers to the feelings that are conveyed by a text. Tone refers to the writer’s attitude towards the subject matter or audience.
Source: Conversation with Bing, 5/3/2023
(1) sentiment analysis (opinion mining) – SearchBusinessAnalytics. https://www.techtarget.com/searchbusinessanalytics/definition/opinion-mining-sentiment-mining.
(2) Sentiment Analysis Guide – MonkeyLearn. https://monkeylearn.com/sentiment-analysis/.
(3) Sentiment Analysis: What’s with the Tone? – InfoQ. https://www.infoq.com/articles/sentiment-analysis-whats-with-the-tone/.
(4) What is Sentiment Analysis? – Sentiment Analysis Explained – AWS. https://aws.amazon.com/what-is/sentiment-analysis/.
Asking AI to help on the rephrase
Fortunately, AI tools can help improve the original sentiment, emotion, and tone. Consider this prompt:
Rephrase this paragraph for a court brief so that it is more informative and more persuasive and less negative. Here is the text: “But wait. Nearly one year (!) after the Court invited her views, the Solicitor General has delivered what can only be described as a hot mess of a brief. In it, the government claims to have unearthed a new obstacle to the Court’s consideration of the question presented. Specifically, the government asserts that, because respondents’ claims do not “arise out of” the military’s combatant activities within the meaning of the combatant-activities exception in the Federal Tort Claims Act (FTCA), this case is an unsuitable vehicle in which to address the question.”
ChatGPT suggests:
After almost a year since the Court’s request, the Solicitor General has submitted a brief which presents an obstacle to the Court’s consideration of the question presented. The government claims that this case is not a suitable vehicle to address the question because respondents’ claims do not “arise out of” the military’s combatant activities as defined by the Federal Tort Claims Act’s combatant-activities exception. It is our position that this claim lacks merit since the respondents’ claims are indeed related to the military’s activities, and we ask the Court to consider the merits of the respondents’ claims in addressing the question presented.
Bard has a similar approach:
Nearly one year after the Court invited the Solicitor General’s views, the government has submitted a brief that raises a new obstacle to the Court’s consideration of the question presented. The government argues that the respondents’ claims do not “arise out of” the military’s combatant activities within the meaning of the combatant-activities exception in the Federal Tort Claims Act (FTCA). This argument, however, is unpersuasive. The respondents’ claims are clearly related to the military’s combatant activities, and the FTCA does not bar the Court from considering these claims.
Bing also found a way to make the point with a lighter touch (though the argument seems incomplete):
The Solicitor General has submitted a brief nearly one year after the Court invited her views. In it, the government claims to have discovered an obstacle to the Court’s consideration of the question presented. Specifically, the government asserts that this case is an unsuitable vehicle in which to address the question because respondents’ claims do not “arise out of” the military’s combatant activities within the meaning of the combatant-activities exception in the Federal Tort Claims Act (FTCA).