This article is based off of a paper titled Development of temporal structure in zebra finch song written by Christopher M. Glaze and Todd W. Troyer published in The Journal of Neurophysiology, Volume 109 in 2013.
The zebra finch is a bird native to central Australia that is often used as a model organism in neuroscience. They learn their song repertoire, and thus act as a basis for study on learning, memory, and sensorimotor integration. In this study, finches are used to examine the neural basis of sequence learning, with temporal precision and tight links with timed bursts in forebrain neurons.
Sequence learning is an important topic in modern neuroscience. Since the presentation of material plays such an integral role in learning, scientists are working to understand the neural connections involved. Zebra finches work as an excellent model organism in this regard.
Song learning occurs in two overlapping processes: “sensory acquisition” and “sensorimotor learning.” In sensory acquisition, a young bird is exposed to songs sung by a tutor and forms a template for the songs. Sensorimotor learning occurs when the bird sings the songs learned by sensory acquisition. Each song is highly stereotyped, and each is unique in its timing and sequence of syllable production.
There are four aspects of song temporal structure in the development of song timing: mean duration of song syllables and the silent gaps between them, timing variability linked to song tempo, timing variability expressed independently across syllables and gaps, and transition probabilities between consecutive syllable pairs.
Most studies to date have focused on how the songs are learned, but little has been done to focus on the overall temporal structure of a song that is being learned. It would be interesting to know how syllable-based representations are formed and how they come into play during song learning.
In this study, a timing variability model was used that had been developed in a previous study (Glaze and Troyer 2012) to analyze changes in the temporal structure of zebra finch song from 65 days to 1 year of age in seven male songbirds. Specifically, the songs were gathered at four ages: 65-70 days, 85-90 days, 125-135 days, and 365-375 days. By using the aforementioned model, they were able to separate components of temporal variability and track them over the course of the zebra finch’s development. Using manually selected sound clips, a match score was used to calculate how much like the sound clip it was, and, if the match score was greater than 0.5, it was grouped into categories from 0.5 to 1.5 where audio was randomly selected for judging. Then, a scientist would manually determine whether or not the song was classified correctly by syllable, with a gap being a period of silence amounting to two syllables. A transition was deemed valid as long as both syllables had a match score above the optimal threshold of 0.5, and the transition time was less than 100 ms.
They disproved a previous study that found average song tempo increases as a function of age, when in fact it seems to decrease. Next they showed that variability in tempo decreased over late development with silent gaps decreasing in tempo variability between 65 and 365 days old, when syllables did not (Figure 1).
In order to see if the changes in timing correspond to syllables in the gaps, developmental changes in transition probabilities were determined. These probabilities were used to show a developmental trajectory that correlated with the timing parameters (Figure 2). These results were disproportionately occurring between 65 and 85 days old. Any tempo increase happened in the gaps between syllables, and never on the syllables, and on average, gaps shortened during the time periods studied, and syllables did not.
The birds were also interrupted from their singing by a quick flash of light. Behavior data from the adult birds suggest a chaining hypothesis in which “the parts of the chain corresponding to gaps remain more weakly connected than the parts of the chain corresponding to syllables.” Since the timing is more variable in gaps than syllables, being disrupted by flashes of light seemed to interrupt songs at transitions more frequently. This hypothesis was physically linked to the high vocal center (HVC) neurons as a chaining nucleus that helps to control the timing of zebra finch song.
Synchronization is posited to be a result of bilateral connections from the midbrain and brain stem that end up feeding back to HVC. However, another locus is needed to account for the changes in timing variability found in this study. The lateral magnocellular nucleus of the nidopallium (LMAN), an output nucleus located in the basal ganglia specifically involved in song learning, was determined as a possible source for the neural connections contributing to independent variability. Previously, song variability was found to be driven by LMAN, but this influence decreases over time. Since these two loci (HVC and LMAN) are not linked, they thought to attribute the communication to the robust nucleus of the arcopallium (RA), a premotor nucleus that receives signals from LMAN, to the HVC. New neurons are also formed in these areas during the first year of life, indicating that they may play a role in the decrease of variability in song.
Strengths and Weaknesses
An obvious weakness of this experiment is the manually selected clips by the scientists. It appears to be inefficient, and though their data was not apparently detracted from, it may be in their interest to develop computer algorithms that draw from a database to recognize similar sound patterns. However, the timing variability model they used in this study was novel and allowed them to accurately separate the variation within the song in a structured way.
Considering these results, it would be interesting to determine whether these timing characteristics and developmental patterns apply to other systems. Studies have already been proposed for action learning in mammals (Graybiel 1998; Yin et al 2009; Costa 2011), with timing playing a large role. This system of learning in finches may correspond to mammalian learning patterns, and science may be able to explain the neural patterning behind learning and memory, thus furthering the understanding of human learning and memory.
Brainard, MS and Doupe, AJ. Auditory feedback in learning and maintenance of vocal behavior. Naure Reviews Neuroscience 1: 31-40, 2000.
Costa RM. A selectionist account of de novo action learning. Curr Opin Neurobiol 21: 579–586, 2011.
Glaze, C and Troyer, T. Development of temporal structure in zebra finch song. J. Neurophysiol 109: 1025-1035, 2013.
Glaze CM, Troyer TW. A generative model for measuring latent timing structure in motor sequences. PLoS ONE 7: e37616, 2012
Graybiel AM. The basal ganglia and chunking of action repertoires. Neurobiol
Learn Mem 70: 119–136, 1998.
Graybiel AM. The basal ganglia and chunking of action repertoires. Neurobiol Learn Mem 70: 119–136, 1998.
Yin et al. Dynamic reorganization of stiatal circuits during the acquisition and consolidation of a skill. Nat Neurosci 12:333-341, 2009.