In the modern technological age, machine learning algorithms have been applied to almost every field of study, and linguistics is no exception. In this post, I’ll give a brief explanation of how neural networks work and then demonstrate how they can be used to assist linguistics research.
Neural networks, or NNs for short, are machine learning algorithms that imitate human thinking in order to process information. In a broad sense, all of their parts correspond to parts of the human brain. A NN consists of a series of layers, each of which plays an important role in processing data. The first layer is the input layer, a relatively simple section of a neural network that takes in preprocessed data and feeds it onwards through the model. The data then goes through one or more hidden layers. Composed of weighted neurons, these hidden layers are responsible for categorizing the data. The preprocessed data is run through each neuron and, depending on how well the data fits the conditions of the neuron, the weight of the neuron is adjusted. A neuron with a high weight accurately fits many data points. This process is referred to as training, as it teaches the model how to classify data. After the data runs through all of the hidden layers, it reaches the output layer, which classifies each data point based on how well it corresponded to each neuron.
From a more mathematical perspective, each neuron can be thought of as a line. If the dataset fits the line well, then the weight, or importance, of the neuron will increase. The output layer classifies data by creating a graph that averages all of the neuron lines based on their importance. A datapoint’s placement on the graph gives it its classification. For a fantastic visual representation of this process, I highly recommend you check out TensorFlow’s neural network playground.
Now that we have a basic understanding of how neural networks function, let’s discuss how they can be used to advance linguistics research. In order to perform a large portion of linguistics research projects, linguists must identify and study patterns across entire languages, each consisting of hundreds of thousands of words that follow hundreds of rules. Seems like a daunting task right? Thankfully, the classification power of neural networks makes finding such patterns significantly easier. Furthermore, recural neural networks, NNs that are specialized for machine translation, can be used to mimic human language acquisition, providing valuable insight into how the human brain understands language.
Much linguistics work is still done by hand, as neural networks are still an advancing and complicated technology, and sometimes patterns can be identified more easily by an experienced linguist. However, NNs certainly play a large role in modern linguistics work, making classification, pattern identification, and language acquisition modeling significantly easier.