Perla is a conversational agent who conducts a structured interview with a user and estimates the presence of depression symptoms.

Early diagnosis of depression is an aspect of great importance for the psychological health of the population. Perla provides a new interactive method, based on Artificial Intelligence, which allows the automatic detection of depression symptoms.

Perla’s face (generated with StyleGAN2)

Perla was born as a conversational research agent for psychological evaluation (P.E.R.L.A. – Psychological Evaluation Research Lab Agent). The current version of Perla is based exclusively on the Spanish language and its programming includes instrument PHQ-9 (Patient Health Questionnaire).

In the following demonstration video you can see how its text-based interface works, although Perla is also available though a voice interface.

Perla uses Artificial Intelligence to analyze user responses and estimate the frequency in which different symptoms of depression appear.

For some people, such as teenagers and people with visual disabilities, using an interactive agent (chatbot or voicebot) is much easier, more attractive and motivating than filling out a questionnaire.

As part of our research, we are comparing the data obtained using the traditional PHQ-9 questionnaire and the estimates made by Perla based on interviews with the same users.

Our preliminary results (see research summary) indicate that Pearl has a reliability and validity equivalent to that of PHQ-9 itself.

To talk to Perla and see how it works you can visit our technology demonstrators section or go directly to Talk to Perla (only in Spanish).

Perla is able to:

  • Understand user’s responses in natural language (Spanish).
  • Calculate an interviewee’s PHQ-9 score and decide if it exceeds the cut-off point (depression screening).
  • Provide information to the interviewee about the results obtained.
  • Provide links and additional information to each person based on their results.
  • Conduct the interviews through multiple online channels (Skype, Hangout, Facebook Messenger, Google Assistant, Web, telephone, Telegram, Twitter, etc.).
  • Store the data of each user in an anonymized and secure way, including the total score and the responses to each item.
  • Connect with third-party APIs to send the data and generate a report with the results of a screening campaign.

To learn more about Perla’s design and the validation studies that we have conducted, please see the following paper:

Arrabales, R. (2020). Perla: A Conversational Agent for Depression Screening in Digital Ecosystems. Design, Implementation and Validation. arXiv (versión en español).

Want to know more about Perla?

If you are interested in these applications, please Contact Us for more information about our AI products.