💻 computer science

What Does a Meow Mean? In Search of Intuitively Understandable Communication by a Nonverbal Companion Robot

This paper presents a user-centered study on designing nonverbal communication signals for a cat robot assisting older adults, finding that while combining visual and auditory cues yields the highest accuracy in understanding the robot's intentions, auditory signals alone are often less effective or even counterproductive unless they convey strong sentiments like purring.

Original authors: Vivienne Bihe Chi, Claudia B. Rébola, Bertram F. Malle

Published 2026-05-05

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Vivienne Bihe Chi, Claudia B. Rébola, Bertram F. Malle

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you've just bought a new, high-tech robotic cat to keep you company. It can't talk, and it doesn't have a screen that says "I'm hungry." Instead, it tries to tell you what it needs using a mix of meows, flashing lights on its collar, and simple pictures (like a pill or a ball of yarn) on a tiny screen.

The big question the researchers asked was: "If this robot cat meows or flashes a light, will an older adult actually understand what it's trying to say?"

Here is the story of how they found out, broken down into simple parts.

1. The Goal: A Robot That "Speaks" Without Words

Many older adults live alone and face challenges like loneliness or forgetting to take medicine. Robots could help, but if the robot is too confusing, people will just turn it off. The researchers wanted to build a robot cat that communicates in a way that feels intuitive—like talking to a real pet, not a computer.

They built a robot based on a popular toy cat, but they added a "smart collar" with:

Ears: To play sounds (meows and purrs).
Neck: To flash LED lights.
Collar Screen: To show simple icons.

2. The Testing Process: From Guessing to Knowing

The team didn't just guess what signals would work; they tested them in three steps, like a chef tasting a soup before serving it.

Step 1: The Pilot (The Taste Test): They showed 105 people GIFs of the robot doing things. Most people guessed correctly what the robot wanted (e.g., "It wants to play!" or "It needs charging"). This gave them confidence that their basic ideas were on the right track.
Step 2: The Focus Group (The Kitchen Talk): They sat down with a small group of older adults (some with hearing or mobility issues) and showed them the robot's signals.
- The Surprise: The group loved the visual icons and lights. However, they got confused by some of the sounds. One sound that was supposed to mean "I need medicine" made them think, "Oh no, there's a medical emergency!" Another sound that was supposed to be "happy" sounded too frantic. They also thought the blue flashing light looked like an ambulance siren.
- The Fix: The researchers went back to the drawing board. They swapped the fake robot sounds for real recordings of actual cats and changed the light patterns to be less alarming.
Step 3: The Big Experiment (The Banquet): They invited over 300 older adults (ages 65+) to an online study. They showed them 9 different scenarios (like the robot wanting to play, reminding them to take pills, or saying "hello" after being petted).

3. The Three Rules of the Game

In the big experiment, they tested three different ways the robot could communicate:

The Full Package: Sound + Picture + Lights.
The Silent Cat: Picture + Lights (No sound).
The Mute Cat: Sound + Lights (No picture).

Think of it like a movie: Sometimes you have the picture and the sound (Full). Sometimes you have the picture but the volume is off (Silent). Sometimes you have the sound but the screen is black (Mute).

4. What They Discovered: The "Visual vs. Audio" Surprise

The results were fascinating and a bit counter-intuitive:

The Picture is King: When the robot showed a picture (like a pill icon), people understood it almost perfectly, even if they couldn't hear the sound. In fact, removing the sound actually made people understand the robot better in some cases. It seems that for older adults, a clear picture is much more reliable than a meow.
The Sound Trap: When the robot only made sounds (no picture), people often got it wrong. A "meow" is too vague. Is it hungry? Is it lonely? Is it in pain? Without a picture, the meaning gets lost.
The Exception (The "Purr" Factor): There was one time the sound worked great: When the robot was being petted. If the robot purred while you were stroking it, everyone knew it meant "I love this." In social, emotional moments, the sound worked well. But for tasks (like "take your pill"), the picture was essential.
Initiating vs. Responding:
- If the robot started the conversation (e.g., "Hey, let's play!"), you needed the picture to know what it wanted.
- If the robot responded to you (e.g., you petted it, and it purred), the sound was enough to tell you it was happy.

5. The "Practice Makes Perfect" Effect

The researchers noticed something encouraging: People got better at understanding the robot the second time they saw the signals.
Even though the first round was a bit tricky, by the second round, everyone understood the robot much better. This suggests that older adults are very capable of learning a new "language" if it's designed well and they get a little practice.

6. The Bottom Line

The paper concludes that for a non-verbal robot cat to be useful for older adults:

Visuals are non-negotiable. You need icons or pictures to explain what the robot is talking about (especially for tasks).
Sounds are for feelings. Meows and purrs are great for showing emotion (like happiness or urgency), but they are bad at explaining specific tasks.
It works. Even without a manual or training, older adults could figure out what the robot wanted, especially when the robot used both pictures and sounds together.

In short: If you want a robot cat to tell your grandma to take her medicine, don't just make it meow. Show her a picture of a pill. If you want it to say "I love you," a purr is perfect.

Technical Summary: What Does a Meow Mean? In Search of Intuitively Understandable Communication by a Nonverbal Companion Robot

Problem Statement
Older adults living alone face significant challenges, including isolation, loneliness, and safety risks. While social robots offer potential assistance through reminders, activity initiation, and comfort, older users often struggle with technology due to anxiety and feelings of incompetence. A critical barrier to adoption is the lack of intuitive communication. Specifically, for zoomorphic (animal-shaped) companion robots, verbal communication is often inappropriate or uncanny. The authors address the need to design and evaluate nonverbal communication signals (auditory and visual) for a cat robot that are intuitively understandable by adults over 65 without prior training or manuals.

Methodology
The research employed a mixed-methods, user-centered approach involving three distinct phases to iteratively design and evaluate communication signals for a modified "Joy For All" animatronic cat equipped with a smart collar (e-ink display, LEDs) and audio output.

Pilot Study: An online survey ( $N=105$ ) tested eight initial signal "packages" (combinations of sound, LED patterns, and icons) across various use cases (e.g., medication reminders, play invitations). Participants viewed GIFs of the robot and provided open-ended interpretations followed by ranking researcher-provided options.
Focus Group: A qualitative interview with four older adults (including one with severe hearing impairment) and a caregiver. Participants reviewed revised signal packages via GIFs and audio, providing feedback on the clarity of auditory cues (meows, purring) and visual cues (LED colors, icons).
Large-Sample Online Experiment: A preregistered study with 315 participants (median age 70) tested the revised signals. The design utilized a $2 \times 2 \times 2$ $2 \times 2 \times 2$ factorial structure (Direction: Initiate/Respond; Purpose: Social/Functional; Priority: High/Low) plus an emergency scenario. Participants were randomly assigned to one of three signal conditions:
- Full Signal: Visual icons + Sound + LEDs.
- No Sound: Visual icons + LEDs only.
- No Display: Sound + LEDs only.
  Participants interpreted nine scenarios in two rounds. Accuracy was measured via open-ended coding and a likelihood judgment task (rating the intended message against five distractors).

Key Contributions

Signal Design Framework: The paper details the development of a multimodal signal set for a nonverbal robot, distinguishing between "Initiate" (robot starts interaction) and "Respond" (robot reacts to user) contexts, and "Functional" vs. "Social" goals.
Empirical Evaluation of Redundancy: The study systematically tests the hypothesis that redundant multimodal signals (visual + auditory) are necessary for effective communication with older adults, specifically examining how the removal of one channel affects comprehension.
User-Centered Iteration: The work demonstrates the value of moving from synthetic sounds to natural cat recordings and refining visual icons based on direct feedback from the target demographic (older adults).

Results

Overall Comprehensibility: Participants were generally able to infer the robot's intentions without training. In the "Full Signal" condition, the median accuracy for open-ended interpretations was 62%.
Channel Effectiveness:
- Visual Dominance: When visual signals (icons) were present but sound was absent, accuracy actually increased (median 70%).
- Auditory Limitations: When visual signals were absent (sound only), accuracy significantly decreased (median 41%). Auditory signals alone were often insufficient for conveying specific intentions, particularly for initiating actions.
- Contextual Nuance: Redundancy was most critical when the robot initiated an interaction; icons were essential here. When the robot responded to a user (especially in social contexts like being petted), auditory signals (e.g., purring) were sufficient to convey sentiment even without visual icons.
Learning Effect: Accuracy improved significantly in a second round of testing, particularly for the "Full Signal" and "No Sound" conditions. The "No Display" condition showed the least improvement, suggesting that visual icons are harder to learn without exposure.
User Acceptance: While the "No Display" condition resulted in lower accuracy and perceived ease of understanding, user acceptance (liking and excitement) remained relatively stable and did not significantly drop compared to other groups, though overall acceptance was modest ( $M=5.9$ ).

Significance and Claims
The paper claims that human-centered design can enable older adults to spontaneously and successfully interpret nonverbal communication from a companion robot. The primary insight is that visual and auditory signals are not strictly redundant; their utility depends on the communicative intent. Visual icons are essential for conveying specific, object-referenced, or initiating messages, while auditory signals are effective for conveying emotional states during responsive interactions.

The authors conclude that while the current prototype has limitations (e.g., failure in specific "greeting" scenarios, limited gesture vocabulary), the study establishes a foundation for real-world deployment. They emphasize that older adults are capable of learning these signals over time, and that a design focusing on intuitive, multimodal cues can facilitate adoption without requiring complex training. The work highlights the necessity of testing signals with the actual stakeholder population, as designers' assumptions about signal clarity (e.g., a "hello" meow) often diverge from user interpretation.