top of page
Search
  • Writer's pictureJiah Hwang

The Potential for More: A Review of “[Speech & language detection in dogs]"

Updated: Jun 18, 2023

Summary

The hypothesis for this experiment was that “If dogs can extract auditory regularities of speech and a familiar language, then there will be distinct patterns of brain activity for natural speech vs. scrambled speech, and also for familiar vs. unfamiliar language.” The participants of this test include 18 adult family dogs (9 females, 3-11 years old with a mean age of 6.6 — 5 golden retrievers, 6 border collies, 2 Australian shepherds, 1 labradoodle, 1 cocker spaniel, and 3 mixed). As for the human participants, 16 adults filled out a 7-point Likert scale* survey to rate the naturalness of the stimuli (mean age = 30.9 years, 7 males) without knowledge of either Hungarian or Spanish (native languages of the participants: 4 French, 3 English, 2 Hebrew, 2 Italian, 2 Polish, 1 German, 1 Portuguese, and 1 Swedish).


The “stimuli” mentioned here consist of extractions of full sentences from each recording of Hungarian and Spanish speech samples from readings of the XXI chapter of “The Little Prince,” recorded by 2 different female speakers with similar vocal characteristics, one in each language with a lively tone. These would be played to test if the canine participants' brain activity changes when the sounds are separated into different properties. In order to compare the two sounds, they used the quilting algorithm, which scrambled the original speech fragments into 30ms of slices, which resulted in 24 different speech fragments with the same volume and quality of sound. The low-level properties, such as pitch and duration, were maintained in these recordings. In contrast, higher-level properties, such as specific sounds, structures, and rhythm, were disrupted into 4 different condition blocks: Natural speech in a Familiar language (NF), Natural speech in an Unfamiliar language (NU), Scrambled speech in a Familiar language (SF), and Scrambled speech in an Unfamiliar language (SU). Each condition block was presented three times in a pseudo-random* order, with three additional silence blocks at certain points of the recordings.


Results


Lastly, the study used fMRI MVPA (multi-variate pattern analysis)* in order to detect the neural representation of speech and language familiarity in the dogs and found that the detection of speech naturalness was associated with the bilateral near-primary auditory cortical regions* of the brain, while language familiarity was associated in the ventral parts of the auditory cortex. Data analysis involved preprocessing the functional images and conducting whole-brain general linear model analysis and multivariate pattern analysis.


The data from the results of the brain scans underwent t-tests* and were assessed by their distinct cerebral* patterns for natural vs. scrambled speech and familiar vs. unfamiliar language in the dog brain, along with the difference in the ratings of naturalness and neural dissimilarity by the 16 human participants. From this statistical test, a negative correlation was found between the neurocephalic index* and the performance of the classifier in a near-primary auditory region with no correlation with age. Yet, there was a positive correlation between the dissimilarity index for familiar vs. unfamiliar speech and age in 2 clusters: the left postcruciate gyrus* and the left mid-suprasylvian gyrus*. This correlation shows that dogs can differentiate natural/scrambled speech and familiar/unfamiliar languages. Yet, individual differences in the neurocephalic index and age may influence brain activity related to language familiarity.



Review


  1. What was the reason for choosing family dogs/specific dog species as the participants in this experiment?

As the beginning of this research paper mentioned, the test included “8 adult family dogs (9 females, 3-11 years old with a mean age of 6.6 — 5 golden retrievers, 6 border collies, 2 Australian shepherds, 1 labradoodle, 1 cocker spaniel, and 3 mixed),” yet a lot of these dogs already have much experience and training with humans as the majority are classified as hunting breeds. As they are breeds equipped for agility, larger jaws, sharper teeth, and more, they either have an ancestry or presently spend more time working with humans other than simply being taken care of by them. Therefore, they have likely listened to and were trained to understand more subtle tones and instructions. At one point, the article even stated that they “were trained previously to remain still inside an MRI scanner,” proving that these dogs have much experience with training by the instructions of humans in the first place compared to the average dog. Although the research article introduces that this extensive experience with communicating with humans is part of the reason that dogs are the top choice for this test, the concern here is that the tone by which the reader of the two languages were speaking, which was described as “recorded with a lively, engaging intonation,” may have reminded the dogs of a specific memory or training command with their owner, causing the proceeding stimulus in their brain.


  1. What was the reason for choosing The Little Prince?

The Little Prince is at a 4th grade reading level with some complex metaphors and characters; unlike anything these dogs have likely experienced before. Therefore, as this test were trying to test the auditory regularities of familiar/unfamiliar language through the testing of natural and scrambled speech, would there not have been more accurate or prominent results if they were read a book or passage of a lower reading level? The explanation for the chosen book and specific chapter was lacking compared to the extensive explanations of the ethics of the experiment.


  1. Wouldn’t the tone and pitch of the female speakers affect how their brain perceives the languages?

Again, the article lacked extensive explanation on why they chose female speakers only to record for the chosen passage of The Little Prince for each language. This may have been a missed opportunity to see how the data of brain activity for each language differentiates when there is a male and female speaker, especially if we consdier which gender the owner of the dogs are, as most dogs tend to be attached to one specific person out of a family (typically the one who feeds them). If the person that the dog that the most attachment to was a woman, then the recording coming from a female speaker might have made more of an impact on the brain activity, compared to a dog who had a male owner for most of its life, in which case, the recording might have held less interest or less of an effect on their brain.


Key


Likert scale

A rating system designed to measure people’s attitudes, opinions, or perceptions, most prominently for questionnaires.


Pseudo-random

Using mathematical formulas to produce sequences of seemingly random numbers.


Multi-variate pattern analysis

Refers to a diverse set of methods that analyze neural responses as patterns of activity that reflect the varying brain states that a cortical field or system can produce based on the MRI scan.


Primary auditory cortical regions

Located on the left lateral side of the brain, responsible for temporal sequences of sound.


T-test

A statistical test appropriate for when you've collected random sample values a small “population” and want to compare the mean from your sample to another value.


Neurocephalic index

The cephalic index is an objective parameter for determining the skull shape, as well as assessing their effectiveness in correcting cranial deformations.


Postcruciate gyrus

A prominent gryus (a ridge-like elevation on the surface of the cerebral cortex of the brain) on the lateral paritel lobe of the brain; a sensory receptive area.


Suprasylvian gyrus

A gyrus related to the motor function of the cervical muscles and the muscles of specific areas of the head.




Bibliography


“Cephalic Index in the First Three Years of Life: Study of Children with Normal Brain Development Based on Computed Tomography.” NCBI, 26 December 2013, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3933399/. Accessed 17 June 2023.

Cuaya, Laura V., et al. “Speech naturalness detection and language representation in the dog brain.” NeuroImage, vol. 248, 2022. ScienceDirect, https://www.sciencedirect.com/science/article/pii/S105381192101082X.

Haxby, V. “Multivariate pattern analysis of fMRI: The early beginnings.” NCBI, 9 March 2012, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3389290/. Accessed 17 June 2023.

Jamieson, Susan. “Likert scale | Social Science Surveys & Applications.” Encyclopedia Britannica, Encyclopedia Britannica, 24 April 2023, https://www.britannica.com/topic/Likert-Scale. Accessed 17 June 2023.

“Pseudo-Random Number.” ScienceDirect, Elsevier, https://www.sciencedirect.com/topics/mathematics/pseudo-random-number. Accessed 17 6 2023.

“Suprasylvian Gyrus.” ScienceDirect, Elsevier, https://www.sciencedirect.com/topics/medicine-and-dentistry/suprasylvian-gyrus. Accessed 17 6 2023.


4 views

Comments


bottom of page