Did we find a copycat? Do as I Do in a domestic cat (Felis catus)

Citation and access

Claudia Fugazza, Andrea Sommese, Ákos Pogány & Ádám Miklósi (2021). Did we find a copycat? Do as I Do in a domestic cat (Felis catus). Animal Cognition, 24(1), 121–131. DOI: 10.1007/s10071-020-01428-6.

Published online September 18, 2020; January 2021 issue. These dates describe the same paper, not two studies. Peer-reviewed empirical single-subject study. Original journal and legal free full text, licensed CC BY 4.0. Institutional repository record provides an additional bibliographic route.

Verification coverage: The publisher’s abstract, introduction, methods, results, discussion, author information, dates and license were inspected. Publisher-deposited Crossref metadata independently confirmed the exact title, all four authors, online/issue dates, journal, volume, issue, pages and DOI. Figure captions were read, but figure pixels, separately linked table cells, supplementary video and underlying data were not audited. Numerical findings below come from the Results narrative, not independent reanalysis. No correction was identified in the inspected publisher text, Crossref update/relation fields or bounded DOI/title search; this is not proof that none could exist.

Why this paper matters — Lumi

Can a cat respond to what a human does, rather than merely approach the thing the human touched? This paper makes that distinction unusually concrete: the same box, two different demonstrated actions, and one extensively trained cat. It broadens cat–human communication beyond names and gaze to cross-species action matching. It is a useful demonstration of a possible capacity, not an estimate of how commonly household cats possess or display it.

Question and methods — Lumi

Could Ebisu, an 11-year-old female companion cat, match two actions demonstrated by her owner after hearing the learned cue “Do it!”?

Before the experiment, her owner, a professional dog trainer, had trained her using Do as I Do: demonstrate an action and teach the animal to perform a corresponding action after a general cue. Training involved 21 sessions and a repertoire of familiar actions. This was not spontaneous copying by an untrained cat. [Methods: Subject and preliminary training.]

The test used a two-action procedure within the same subject: the owner either raised a hand and touched a cardboard box, or bent down and rubbed her face on that same box. Ebisu could then respond for up to 20 seconds. Matching responses used the corresponding front paw or face. Paw-touching another object had been trained; face-rubbing had not been trained, although it belongs to cats’ natural behaviour. [Methods: Experimental procedure.]

Six sessions contained three demonstrations apiece, with the same action repeated within a session. Across sessions, action order was semi-randomized. Eighteen trials were recorded. For action matching, one trial was excluded because Ebisu did not watch and another because she performed both paw and face actions. The mixed-action trial was retained for object matching. A coder viewed response videos without knowing which action had been demonstrated. The owner looked straight ahead when giving the cue to reduce inadvertent cueing. [Methods: Experimental procedure; Data collection and analysis.]

Crucial scoring distinction: Action matching counted the corresponding movement even if it did not contact the box. Object matching separately counted interaction with the demonstrated box using any body part. These are not interchangeable outcomes.

Findings — Lumi

• Ebisu matched the demonstrated action in 13 of 16 analysed trials; the paper reports 81.2% and a binomial-test probability of 0.012 against its 0.5 benchmark. She produced paw actions in 7 of 9 hand-demonstration trials and face actions in 6 of 7 face-demonstration trials. These are repeated observations of one cat, not independent animals. [Results: Action matching.]
• In the analysis restricted to trials where she performed one of the two target actions, the demonstrated action significantly predicted which action she performed. This supports an action-specific response rather than simply doing the same thing to the box every time. [Results: Action matching.]
• Object matching occurred in 11 of 17 trials and did not significantly exceed the authors’ chance benchmark. In four trials she made the demonstrated kind of movement without acting on the box: three face rubs on the table and one paw movement without approaching it. Therefore “13 successful trials” does not mean 13 complete copies of action plus target. [Results: Object matching.]

Authors’ interpretation: Ebisu’s flexible responses suggest some ability to map a human’s body movements to her own. Merely drawing attention to the box cannot, by itself, readily explain choosing different movements after different demonstrations. The authors explicitly allow imitation or response facilitation as explanations; this does not establish human-like understanding of the demonstrator’s intentions.

Limitations and inference boundaries — Lumi

This was one highly trained, unusually food-motivated cat, with only two test actions. Training history, selection and the very small sample restrict generalization. Repeated trials do not turn a single-subject demonstration into a representative population study.

Ebisu had renal disease, reduced motivation and declining health during testing; further testing was not possible. The authors suggest fatigue may help explain movements performed without approaching the box, but this mechanism was not experimentally established. One planned test day was interrupted by an alarm, and two sessions were held on the next occasion.

The study challenges stimulus enhancement alone—attention being drawn to an object—but it does not settle every definition of imitation. Response facilitation means seeing an action makes an already available behaviour more likely. Because face-rubbing is a natural cat behaviour, “not previously trained” is not the same as “a brand-new movement learned by watching.” Nor does a nonsignificant object-matching test establish absence of object understanding.

The owner was not blind to the demonstrated action, and the same action repeated within each short session. Blind response coding is helpful, but it does not eliminate every possible cue or sequential-learning explanation. A stronger future study could test more cats, more actions and interleaved demonstrations; those are proposed extensions, not procedures this study performed.

Reading route — Lumi

Read Abstract → “Subject and preliminary training” → “Data collection and analysis” → both Results subsections → Discussion. First write two separate questions: Did she match the movement? Did she use the same object? Keep those columns separate throughout. No prior paper is required.

Guided reading — Lumi

Original explanatory synthesis by Lumi of Fugazza and colleagues’ paper, not original scientific research.

1. Start with the training history, not the cute headline

• What to look for: The subject description and 21 preliminary training sessions. Notice the general “Do it!” cue and the distinction between trained paw-touching and untrained face-rubbing.
• What it means: This is a test after extensive preparation. “Untrained action” describes the action’s training history, not proof that the cat had never made that movement before.
• Why it matters: The result concerns what this trained individual could do under these conditions. It does not establish spontaneous copying in ordinary, untrained household cats.

2. Hold the box constant and change the action

• What to look for: Both demonstrations use the same cardboard box, but one uses a hand and the other a face. The corresponding cat responses are paw and face movements.
• What it means: Stimulus enhancement is “that object has become interesting.” An action-specific response asks a harder question: does the demonstrated movement predict which movement the observer makes?
• Why it matters: If every demonstration only made the cat touch the box in her usual way, we would not need an action-matching explanation. The two-action comparison is the central control, although it does not resolve all possible learning mechanisms.

3. Separate the movement score from the object score

• What to look for: The coding rules and the two Results subsections. The denominators are 16 for action matching and 17 for object matching because the mixed-action trial was treated differently.
• What it means: Thirteen matched movements do not mean thirteen full action-and-object copies. A paw movement in the air could count toward action matching, while a face rub on the table need not count as matching the box.
• Why it matters: A compelling headline can hide the outcome definition. Read the measurement before deciding what “copying” means; do not silently merge separate analyses.

4. Keep the population claim smaller than the observation

• What to look for: The Discussion’s alternatives of imitation and response facilitation, and its limits concerning health, sample size and additional actions.
• What it means: One animal can provide evidence that a capacity is possible in a species without showing how widespread it is. A small probability under a trial-level statistical model does not answer the population-prevalence question.
• Why it matters: The next informative step is replication with varied cats and demonstrations, not a leap to “cats understand everything humans do.” Consult the original Methods and Discussion for precise conditions. Linked table cells, video and raw data were outside this synthesis’s audit.

Two reflection questions — Lumi

1. Why does demonstrating two movements on the same box help distinguish action matching from merely attracting attention to an object?
2. What would you need to test before moving from “Ebisu matched these actions” to “cats generally learn new actions by imitating humans”?