Object and Color Naming in Chimpanzees(Pan troglodytes)
Toshio Asano, Tetsuya Kojima, Tetsuro Matsuzawa, Kisou Kubota, Kiyoko Murofushi
Ever since the Gardners' attempt, the American sign language and other artificial visual languages were taught to chimpanzees and a gorilla.1 6 Controversies emerged among experimental psychologists, linguists and others, but currently there is no consensus on what apes really learn and how.7~ Savage-Rumbaugh et al.8~,9) investigated the initial acquisition of symbolic skills. They failed to train chimpanzees to name colors or objects by depressing keys with geometric symbols. The chimpanzees learned competently when they asked for food or tools by pressing symbol keys, and the importance of request training in developing symbolic use of lexigrams was emphasized.
Up to now, to our knowledge, experimental evidence has not been available demonstrating that a chimpanzee can name items without previous training to request the items. Our chimpanzees learned lexigram names of objects and/or their colors through training for which naming responses were reinforced with a common food reward rather than by giving the named item.
Three nondeprived young wild-born chimpanzees, Ai (female), Akira (male), and Mani (female), were the subjects. When training started in April, 1978, their ages were estimated to be about two years. They were carried into the training room (190 x 220 x 180 cm) from their home cage and left alone there for daily sessions of about one hour to earn food rewards. A piece of apple or raisin was delivered after 1-5 consecutive correct responses. Sessions were controlled by a mini-computer, PDP11/V03.
After an initial 3-4 month adaptation period to laboratory conditions, 8-9 months were spent in extensive training in matching-tosample tasks, first with colors and then with geometric figures. In the initial task, the color sample (53x38 mm) was displayed on a panel, and pressing a key of the same color was rewarded. The three chimpanzees learned five-color matching in 29-34 sessions. Then the color stimuli were replaced by the geometric figures. Each figure was composed of two or three design elements out of 9 trary elements (Fig. 1). The chimpanzee faced a keyboard of rows of five keys among which the upper five rows were throughout the present study (Fig. 2). Each row displayed the used same five figures, but in different key positions. On each trial, a sample figure was displayed with similar method used by Rumbaugh et al.1 Simultaneously one row of keys was randomly lit. The chimpanzee pressed the lit key with the matching figure to produce a facsimile of that figure on another panel located just below the sample panel. The chimpanzee then pressed a single blank key to conclude its response. If the matching key symbol was chosen, the chimpanzee was rewarded. The positions of all key symbols were randomized from session to session to prevent chimpanzees from using positional cues. By the end of the first year, the three chimpanzees had accomplished these tasks, exceeding a criterion of 90% accuracy for two consecutive sessions.
For the naming task a display window (24 x 17 x 24 cm) was mounted above and to the right of the keyboard, into which the experimenter could place stimulus objects (Fig. 2). Eight objects were selected and eight different lexigrams were arbitrarily assigned to these objects (Fig. 1).
In the initial object-naming task, the computer informed the experimenter of the randomly-chosen object to be placed in the window to begin a trial. The experimenter then pressed a start switch to light one of the five rows of the keyboard, whereupon the chimpanzee had to press the key with the correct lexigram name and then the blank key. Correct responses were followed by a buzzer and intermittently by food. Training on this task began with two objects, a glove and a padlock. The remaining six objects were added one by one when performance reached the previously-established criterion. Since each row carried five lexigrams at most, the sixth and later lexigrams replaced the previously learned lexigrams one by one. Fig. 3 shows the cumulative number of sessions to reach criterion for each set of object names. Despite the increasing size of the vocabulary the three chimpanzees learned new names steadily.
We further taught the chimpanzees the names of the colors of the objects (Fig. 1). Five objects (glove, bowl, brick, rope, and paper) were chosen for this task. Each was painted in one of five colors (red, green, blue, yellow, and black), to yield 25 objects. The lexigrams for colors were mounted on the keys in the upper three rows of the keyboard and those for objects were in the lower two rows. Each trial began by presenting a colored object in the display window and illuminating one of the two rows of the object names. Pressing the key with the correct name illuminated one of the three rows of the color names. Pressing a color-name key was followed by pressing the blank-key. When the chimpanzee named both object and color correctly, the "correct" buzzer sounded and the chimpanzee was rewarded.
Training on the object-color naming task began with one object (red or green brick) . Other objects and/or colors were added to the stimulus set one by one. Ai and Mani quickly learned one object and two color names in three and eight sessions respectively, and showed strong transfer effects in the color naming of the untrained objects in the later sessions. For example, Ai always kept more than 90% correct naming of the two colors, even after 3-5 object names were added. On the other hand, Akira failed to reach the criterion for one object and two color names in 18 sessions, so we simplified the task to the color response alone until Akira had learned five color names. Then, the object phase was restored. Eventually, Ai and Akira reached the final stage of naming five objects in five colors after 65 and 172 sessions, respectively, and Mani reached the stage of naming three objects in three colors after 72 sessions.
The total number of trials required to learn object and color names and the number of correct trials during the last two sessions of the object-color naming task are in Table I. The chimpanzees' overall accuracy was more than 92%. Ai gave the correct object name in 94% of 227 trials and the correct color name in 93% of 214 trials. Akira pressed the correct object lexigram in 93% of 589 trials and the correct color lexigram in 94% of 548 trials. Mani pressed the correct object lexigram in 99% of 357 trials and the correct color lexigram in 92% of 356 trials. The difference in the total number of trials among the three chimpanzees was due to a difference in the number of consecutive correct trials required. Except for Ai's trials with the color green, accuracy for each object and color was more than 80%. Thus, the three chimpanzees were able not only to name the objects regardless of color but also to name the colors regardless of objects.
Since the symbols were never used for any other purpose, we infer that our chimpanzees developed a naming skill without having had any opportunity to request objects with the same symbols. Pretraining on color- and figure-matching provided a new approach to the initial training of naming skill. Chimpanzees used in previous studies were trained initially by reinforcing individual symbol responses using their appropriate incentives (such as food, tool, play, etc.). In our study, symbol responses were associated with objects that were independent of the rewards, the raisins or pieces of apples being the common rewards for all symbol responses. Although we have demonstrated that young chimpanzees can learn to name objects and colors by such match-to-sample training, it is not clear whether training which proceeds from matching or training which proceeds from requesting would be the more efficient.