Language Acquisition: From Children to Machines

I found this study rather interesting. It focused on English speakers and the English language but the idea is present across languages. I am currently learning Spanish which also has certain suffixes that also have suffixes that act like –er. In both languages, the suffix is not the only suffix that makes a word mean ‘a person that’ does the root verb. This may make children who speak many languages infer what the meaning of a word is even if they have never heard it before just based on the suffix. This may be valuable to study for learners of new languages. Specific suffixes and prefixes give helpful hints to what a word can mean. In addition to people learning languages, this would also be very helpful for machine learning. If machines broke words down by their parts – prefixes, roots, and suffixes – they would more easily learn. In the experiment, younger children were able to answer questions better for agents than instruments. With instruments, the children often gave longer explanations to what the instrument may do. This could speak to how words they already know are structured. “Farmers,” for example, just farm. However, “openers” don’t just open anything. Also, “hammer” is a weird term to focus on the ‘-er’ because “hammers” do not ‘ham’ things. Perhaps, if this were to be used for machine learning, the root would have to be determined to be a verb before it is used. Another possible issue is that the computer would not intuitively know whether the word referred to an agent or an instrument. This would have to be prompted somehow just as it would have to be if a human were asked what a word meant. If a computer does not know, maybe it can learn to react as a child does. Instead of saying “builder” it may say “build-man.” It could use a number of suffixes it knows to create a word that can be understood adequately. As the machine gathers data, it can go from using ungrammatical patterns like “pull-things” for a person who pulls things to eventually creating the correct word for agents and instruments. This would have a machine mimicking how humans learn language – first relying on simple compounds, then learn how to use the correct suffixes for agents, and lastly learning how to use correct suffixes for instruments.