Why study animals' response to speech sounds?

1. Species-general versus species-specific mechanisms

Most theories of speech perception emphasize the aspects of speech that appear to be unique from other sounds. As such, it is expected that other species will not perceive speech sounds in the same manner as humans. We investigate this claim directly by assessing nonhuman animals' response to speech sounds.

Using animals, it is possible to examine the contributions of audition to speech perception while factoring out potential effects of experience (e.g., Kuhl & Miller, 1975; Kuhl & Miller, 1978; Kluender, 1991; Kluender & Lotto, 1994). From these studies, we have learned that animals respond to speech categorically (Morse & Snowdon, 1975; Waters & Wilson, 1976; Kuhl & Miller, 1975; Kuhl & Miller, 1978), exhibit phonetic context effects (Dent et al., 1997; Lotto et al., 1997) and are sensitive to acoustic trading relations (Kluender, 1991; Kluender & Lotto, 1994). In the opposite manner, animals have also assisted in directly assessing putative roles of experience with speech (Kluender et al., 1987; Kluender et al., 1998; Lotto et al., 1999; Holt et al., 2000), allowing rather precise characterization of effects of experience that can be hard to garner with human adult or infant listeners (see Holt et al., 1998 for a discussion) and demonstrating that animals exhibit learning-dependent hallmarks of speech perception such as phonetic categorization (Kluender et al., 1987; Kluender et al., 1998; Lotto et al., 1999). With these qualities, animal models offer a well-suited population with which to probe perceptual ramifications of experience with speech.

These data suggest that mechanisms supporting speech perception may have bases in general auditory processing rather than speech-specific or species-specific mechanisms.

2. Understanding the role of experience in shaping speech perception

Many studies demonstrate that how we perceive speech is shaped by our experience with speech sounds. Japanese infants have little difficulty discriminating the 'r' sound from the 'l' sound (as in "rock" and "lock"). However, as adults, Japanese listeners have great difficulty hearing a difference between these two sounds. Experience with the Japanese language (which does not distinguish between 'r' and 'l') shapes Japanese listeners' perception. Understanding how this happens is a central goal of speech research. However, it is very difficult because it is nearly impossible to exercise complete experimental control over experience with speech. Even very young babies have had a great deal of experience. It would be very advantageous to have a population of listeners that is entirely inexperienced with speech. Among these individuals, it would be possible to exercise complete experimental control over experience and test various theories of how learning shapes speech perception.

Nonhuman animals are just such a population. In our laboratory, we use gerbils (Mericones unguiculatus) to investigate these questions.

But why gerbils?

The gerbil has been identified as well-suited for study of auditory physiology, neuroanatomy and development (e.g., Ryan, 1976; Nordeen et al., 1983; Dolan et al., 1985; Caird et al., 1991; Scheich, 1991; Sinnott, 2000) and is quickly becoming an important, well-developed model of mammalian audition. Unlike many other small mammals, they have excellent low-frequency hearing owing to enlarged middle ear structures (Lay, 1972).

Gerbils have begun to be developed as models of sound localization (Heffner & Heffner, 1988) and other psychophysical tasks (Schulze & Scheich, 1999; Wetzel et al., 1998). Furthermore, preliminary accounts suggest gerbils perform very well in speech perception experiments. For example, gerbils’ threshold functions for detection of vowels are similar to those of nonhuman primates for most English vowels (Sinnott, 1995). Thus, at normal experimental presentation levels (e.g. 60-70 dB), gerbils can be expected to readily detect speech. Gerbils appear to have more refined abilities too. For example, they are able to discriminate vowels from one another with very high accuracy (Sinnott, 1995). Thus, there is evidence to suggest that gerbils’ possess the psychophysical capacity to detect and discriminate speech sounds.

However, these general abilities are of little importance to understanding speech perception if gerbils cannot master more complex behavioral paradigms. Fortunately, there is good reason to believe gerbils are up to the task. In fact, gerbils appear to be an especially good nonhuman animal model of speech perception research. One reason for this is that their auditory cortex is substantially more complex than that of other rodents (Scheich, 1991) and has been described as "...rather comparable to species such as cats or primates" (Wetzel et al., 1998, p. 30). In addition to the obvious benefit this confers in working with an animal that is better-suited to more complex tasks, this fact has encouraged growth of an informative physiological literature on learning and plasticity within gerbil auditory cortex. There have been exciting advances in mapping learning-dependent changes in gerbil auditory cortex. Scheich and Zuschratter (1995), for instance, have begun using imaging techniques [flouro-2-deoxyglucose (FDG) mapping] to explore the relationship between experience with auditory stimuli and concomitant changes in cortical receptive fields. Already, this work has begun to be applied to important issues in speech perception. For example, these techniques have been combined with electrophysiology to investigate the cortical mapping of vowel formants in the gerbil (Ohl & Scheich, 1997).

How do you find out what a gerbil is hearing?

Animals used in the present experiments are in no way exposed to injury, discomfort, or pain. A joint CMU/University of Pittsburgh IACUC has approved methods of animal care and use.

The apparatus is modeled after that of Sinnott et al., 1997. Testing takes place in a sound-attenuated chamber. Within the chamber, gerbils are in a 20-cm3 mesh cage attached to a microphone stand. An automatic feeder to deliver reinforcement pellets (20-mg banana flavor) connects to a feeding cup mounted on the side of the cage. The gerbil cage also contains a cuelight, a water bottle (ad lib access) and an inverted metal cup fixed to the bottom of the cage to serve as platform from which gerbils jump on and off during testing. A photobeam apparatus connected below the platform senses gerbils' actions. Outside the cage, a loudspeaker is mounted facing the gerbil. Stimuli are presented over this speaker via a Tucker-Davis Technologies (TDT) D/A converter, programmable attenuator, and low pass filter under the control of a Pentium III microcomputer. Responses from the photobeam are routed through a customized I/O apparatus and fed through the TDT input device for storage on the computer by custom software.

At the beginning of a trial, the cage cuelight begins flashing. This signals to the gerbil to jump on the platform. At that time, the cuelight stops flashing, initiating a variable interval (1.5-6 sec). The gerbil must remain on the platform for the duration of this interval. Leaving the platform (false alarm) results in a "timeout" of 4 sec during which another trial cannot be initiated. At the end of the variable interval, a 2-sec response interval commences. During this time, a training stimulus is played twice at a pulse rate of 1/sec. If the stimulus is a positive stimulus and the gerbil leaves the platform during the response interval, s/he is reinforced with a 20-mg banana-flavored pellet. If the stimulus is negative, the gerbil must remain on the platform for the duration of the response interval to receive a pellet. An incorrect response in either case results in a 2-sec timeout during which the cuelight is extinguished and no new trials may be initiated. Gerbils work in 15-20 minute daily sessions consisting of 65 trials. Performance during training is indicated by percent correct and d' (an unbiased index of sensitivity; Macmillan & Creelman, 1991) measurements. Reaction time to jump off the platform is also recorded on each trial, providing a continuous variable with which to assess behavior.

---------------back---------------