Author: Rama Akkiraju, Distinguished Engineer, Master Inventor, IBM Watson User Technologies To read a message and to judge the tone conveyed in the message comes naturally to humans. But, at times, the tone may be overlooked, undesired, or not conveyed well by the author. Can a computer detect the tones conveyed in a message accurately and automatically? Helping humans assess and refine tone in written communication is an interesting challenge in the Artificial Intelligence and Cognitive Sciences fields. At IBM Watson, we are beginning to answer this question. IBM Watson Tone Analyzer, now available in experimental mode, is a service to help you assess and refine the tone in textual communication. This latest addition to our Watson Developer Cloud APIs and SDKs is available on Bluemix. Building on similar linguistic analyses that power IBM Watson Personality Insights, Tone Analyzer analyzes given text and provides insights about the emotional, social and writing tones reflected in that text. Such insights can be used for a number of purposes including personal and business communications, self-branding, market research, public relations management and automated contact center management. The constituent parts of the three dimensions of Tone Analyzer Scorecard are explained below. Emotional Tone: Many psychology models exist in literature to capture human emotions such as anger, fear, anticipation, surprise, joy, sadness, trust, and disgust. We have developed a model for inferring emotions from written text. Tone Analyzer captures the salient three among these that we found to be relevant for Tone analysis based on our user studies. These include: cheerfulness, negative emotions and anger. Cheerfulness refers to positive emotions such as joy, optimism, contentment, inspiration, and happiness. Negative emotions include feelings of fear, disgust, despair, guilt, rejection, and humiliation. Anger is a type of negative feeling with strong intensity such as annoyance, hostility, aggression, hurt, frustration and rage. Social Tone: Social tone includes aspects of social propensities in people’s personality. Tone Analyzer currently uses three social tones namely: openness, agreeableness, and conscientiousness adopted from the Big-five personality model. Specifically, openness is the extent to which a person is open to experience a variety of activities; agreeableness is a tendency to be compassionate and cooperative towards others; and conscientiousness is a tendency to act in an organized or thoughtful way. We use these three dimensions to illustrate the openness, agreeableness and conscientiousness of the writer as reflected in the text. Writing Style/Tone: Writing tone provides feedback on how analytical, confident and tentative one’s writing is. Analytical tone shows a person’s reasoning and analytical attitude about things. Confidence tone indicates the degree of certainty exhibited by an individual towards something. Tentative tone shows the attitude of inhibition. In addition, the Tone Analyzer service explains which words in the provided text contributed to which tone. Furthermore, it offers alternate word suggestions to refine the text to reflect desired tones. There is no minimum word requirement for the text input in order for the Tone Analyzer service to perform its functions well. The following figures 1, 2 and 3 show how the service works.

Tone Analyzer demo 1

Figure 1: Shows how the input is supplied to the IBM Watson Tone Analyzer demo application

Tone Analyzer demo 2

Figure 2: Shows the Tone scorecard along with explanation on which words contributed to what tone

Tone Analyzer demo 4

Figure 3. Shows how word suggestions can be obtained to refine text The ability to assess tone automatically becomes especially important in situations where: (a) it is impractical for humans to process large volume of articles/text to derive insights or (b) humans need the aid of a tool to act as their inner critic or tone assistant in helping refine a message being drafted or (c) virtual agents interacting with humans could benefit from understanding the tone of users to respond to them appropriately. Application developers can write email and other social media connections tool plugins (e.g. for Facebook posts, Blog posts and dating applications) to help authors be aware of the tones they reflect in their writing and to help them refine their messages to reflect the desired tones before sending the message, thereby, potentially avoiding regrets and miscommunications. When combined with the IBM Speech to Text service, the Tone Analyzer service can be a powerful tool in helping humans communicate more effectively. The experimental service is currently available for English text input, and is free for exploration. More details about this service, the science behind it, how to use the APIs, and example applications are available in the documentation. You can also try out the service at this demo link. IBM is placing the power of Watson in the hands of developers and an ecosystem of partners, entrepreneurs, tech enthusiasts and students with a growing platform of Watson services (APIs) to create an entirely new class of apps and businesses that make cognitive computing systems the new computing standard.

18 comments on"IBM Watson Tone Analyzer – new service now available"

  1. Hi Rama,

    As always, you and your teams are involved in very interesting initiatives. I appreciate the time when we worked together. Watson is becoming one of us and this is not bad at all but…and here comes my problem. You already know me as a very emotional type of person. Maybe I’m not fully understanding yet the utility of this new feature but for sure there will be a lot of useful applications. Anyway, what I want to say is that the feelings, the emotions which are “displayed” by a human being, either by writting or in verbal communicaiton, are his/her signature. “Enhancing” the written message with the “proper” wording does not mean anything else than eliminating the nature of that person and even distorting the real message. Nobody should be afraid about his/her message, about his/her feelings even if those might not be liked by some of the people they address. Nevertheless, good or bad news, the message transmitted by the written communication shall still remain true and not “politically correct”. I realy hope that Watson will be engaged in scenarios where he should really help solving critical problems and not only fancy stuff. The humanity needs solutions (mostly avoided by the big decision makers…we know why) for very critical problems which are affecting us on a daily basis. It seems we have forgotten that the life should be a sum of small, lot, continuous and happy moments ! In the end, please do not consider my comment as a criticism. I really appreciate the initiative and I’d like to know how wouuld Watson intreprete this message. Keep in touch and have fun with Watson !

    Best and sincere regards,
    Sorin Vlad

    • Rama Akkiraju July 17, 2015

      Hi Sorin, It’s so good to hear from you. I also very much appreciate the time we worked together. I hope is all is going well for you.

      Thanks for your comments on the IBM Watson Tone Analyzer service. I understand the point-of-view you are putting forth. Let me clarify a few things about the service and its potential use cases. Tone Analysis is not about prompting individuals to put on a persona that is not theirs. It’s about helping individuals avoid regret. Most people say they had sent emails that they regretted sending. Tone Analyzer is a tool for humans to refine their message so that it serves its purpose effectively. We envision use cases in Marketing and Public Relations. In Marketing, marketers need to fine-tune their messages to reach out to their target audience effectively. In Public Relations, people need to process large amounts of articles written on a topic/company/product to determine which ones to focus on and act or respond. In this latter case, it is not humanly possible to read all articles to infer the tone conveyed in a message in a timely manner to launch damage control campaigns. In such cases, we envision a tool such as Watson’s Tone Analyzer helping public relations managers in quickly identifying those articles that they should focus on for further action. Other use cases include analyzing the call-log records in call-center management domain to identify opportunities for improvement in customer interactions. I hope this clarifies things.

      Thanks again for taking the time to share your point-of-view.

      • Hi Rama,

        Thank you for your clarifications. I have run the above initial message into in order to see how would my new friend Watson treat my message :-). It’s amazing how many areas can be pretty correctly identified/covered. Anyway, the Conscientiuosness (my god what a difficult word :-))) ) and Agreableness areas are much away from the reality…at least in my case. You’ll know exactly what categories I mean when you’ll look on the generated numbers :-)). Have fun further !

  2. Good idea, but there is a completely different approach to do this.
    It is called BooksAI.

    The main idea is that the words don’t matter when it comes to analyzing a literary work.
    It is all about analyzing non- verbal elements in a language.

    Check it out

    Just a couple of quotes:

    “style is a very simple matter; it is all rhythm. Once you get that, you can’t use the wrong words. Now this is very profound, what rhythm is, and goes far deeper than any words. A sight, an emotion, creates this wave in the mind, long before it makes words to fit it”.
    Virginia Woolf on Writing and Consciousness

    “rhythm is something much more significant; rhythm probably means the dissolving of word meanings, their merging into a continuous, inwardly indissoluble stream of images. In other words, rhythm provides an opportunity for non-Bayesian reading of the texts”
    V. Nalimov (Russian philosopher and mathematician)

    I hope you will find it interesting

    • Rama Akkiraju July 19, 2015

      Thank you Andrew for your comments and pointers. We will look into these pointers.

  3. I read with interest and excitement this article about the Watson Tone Analyzer. Advances in AI can be significant aides to human activity when used with care. They can also be used as tools of oppression and control. Let’s work diligently to ensure that this tool is used, as Rama Akkiraju says, for “helping individuals avoid regret” and not for making individuals “regret the existence of the tool”.

    To “avoid regret” we should attend to who uses the tool, the author of the text, someone editing or publishing the text or the recipient of the text. Judging tone may come “naturally” to humans but there is a wide spectrum of ability in this skill. Such judgment is an interaction between the author and the judge, not solely dependent on the author as implied in Akkiraju’s opening remarks.

    My interest is in learning and teaching — of normal/average humans and of those who bring learning deficits or extraordinary strengths to the learning environment. The tone analyzer could have beneficial applications to help students who lack normal ability to judge tone in written communications. It could also be very useful for gifted writers who want to address a particular audience with which they may not be familiar. Another educational use might be as an automated tool to help teachers detect emotional issues they had not previously noticed in their students.

    This last application of tone analysis has some less-than-savory implications for privacy. Not every writer wants their emotions detected. Careful thought should be given to who is given access to this powerful tool and under what conditions it is used. It would probably not be beneficial for university professors to begin grading student essays on the basis of the emotional and social tone conveyed rather than the academic merit of the work.

    Having outed my reservations, I used the Analyzer on an earlier version of this text and a revised version. By added references to Akkiraju’s phrase “avoid regret” and substituting “Watson Tone Analyzer” for “it” in a couple of places I increased the “Negative” rating and the “Analytical” significantly. I look forward to the development of increased sophistication in the Tone Analyzer and many creative, positive uses to which it can be put.

    • Rama Akkiraju July 20, 2015

      Thank you LizaLoop for your observations and for sharing your ideas around additional use cases for Tone Analyzer. We will continue to improve our Technology.

  4. Shann Dornhecker July 20, 2015

    Having read this, one of the biggest problems will be that your majority of users will most likely be those who are ESL (English as a second language). Because of this, tone analysis will be off largely due to grammatical and syntax errors that have not been compensated for by inquiring about the user. Also, as there is a marked difference in written language skills by age and gender among native English speakers, that also plays into tone.
    As well, whether the person being written to is known intimately, professionally, or by third person also plays a part in the communication. I did a study on a small sample of students aged 18-50 on their text language and how formal or informally they wrote and used emojis to their friends/peers how closes those friends and peers were, business associates senior, equal and junior, parents and significant other. They changed syntax, language usage accordingly.
    Different parts of the country have different ways of using formal and informal language.
    Also, knowing the age/gender it is possible to tell the education level/literacy level of the communication which goes a lot towards tone.
    If your algorithms are not taking in input from the user on these items then at the moment you don’t have much more than a common English usage device that is being used for plagiarism and machine learning to find documents.

    • Rama Akkiraju July 21, 2015


      Thank you for sharing your observations and the results of your study. You correctly pointed out that writing patterns, grammatical structure, formality etc. could all vary based on the person who is writing. Our work is presently focused on analyzing words in written text. Research is ongoing on understanding sentence level, paragraph level and article level context.

  5. Shann Dornhecker July 20, 2015

    One more question, having worked in PR and Marketing, how many PR executives did you work with when designing this? As there is a set format and PR people use data analytics and responses to adjust messaging, how is tone going to impact that? 99% of the PR Execs I know, instantly know what and how to say it during a crisis management problem. It’s not going to stop people being human and sending an email that may be scathing or regretful. However before a press release or a response goes out it has been reviewed a number of times with a high consequence for things that go out wrong. As well the highest likelihood that this happens is more on instant media like Twitter where mistakes are hard to retract.
    The biggest application would be in the business communication or as an educational tool.

    • Rama Akkiraju July 21, 2015


      Thanks again for making another important point. In Public Relations scenario, we envision two use cases. One is to help humans identify the tone reflected in large volumes of articles written about a company automatically. The second one is for refining messages. I see that your point is more on the latter use case. A tone check even by an editorial body before sending out all that important marketing message may be a good idea but I do agree that in this latter case, the utility value may be low. In Watson portfolio, there is another cognitive service that can help refine marketing messages to suit specific target audience. It is called Message Resonance. You can find more information on that service here.

      Your observation about biggest use cases in business communication domain or an educational tool is very helpful.

      Thank you.

  6. I enjoy what you guys are usually up too. This sort of clever work and reporting!

    Keep up the amazing works guys I’ve incorporated you guys to my blogroll.

  7. Hi Rama
    Congratulations to you and your team for delivering this invention in the area of tone analysis; This area itself looks very interesting to me. This tool will be useful to everyone in future, not only for PR and Business Communication. May be non-native-english speaking people can also leverage it well in their communication. I am happy to hear that you are working in tone analysis area. Keep up the good work !

    Best Regards

    • Rama Akkiraju July 23, 2015

      Hi Kesava,

      Thanks for your note and your comment on the additional use cases where Tone Analyzer can be applied.

  8. Giridhar Athanki September 24, 2015

    Rama the idea and the Watson Tone Analyzer solution are great. The value of this asset can be further enhanced by integrating with lotus notes. This can become a strong selling point internally. By extending this to other company e-mail service IBM can enhance their client team capability.

    Let me know your thoughts on this.

    Thanks for your time.

    • Rama Akkiraju September 25, 2015

      Hi Giridhar,

      You are absolutely right. Tone Analyzer can be a value add to email applications.

Join The Discussion

Your email address will not be published. Required fields are marked *