Chatterboxes, The Language Instinct, Steven Pinker

This post is a summary of the chapter "Chatterboxes" from the book "The Language Instinct" written by the eminent linguist Steven Pinker. This chapter deals with the statement in the title of the book - Is language an instinct for humans?

Language has never been a cultural phenomenon. Language exists and has existed in every culture on earth. No society can claim the title "the cradle of language." This is one of the arguments for claiming that language is innate in human beings. However, for the skeptics, the universality of language may not single-handedly prove the innate nature of language. For instance, Coca-Cola or Facebook are available almost everywhere on earth. Does this mean Coca-Cola and Facebook are innate too? (According to me, the desire to have a drink and to stay connected are innate in human beings. So the universality of language is a conclusive proof.)

Let's look at another argument for the innate nature of language. There are evidences to show that children reinvent language, not because they are asked to do so, but because they have to. Before going into the argument, it is important here to burst two myths about child language acquisition: 

    (1) The first myth is that children learn to speak from their parents (e.g. Motherese - dogggie, pappie, ...). This is not true because parents do not explicitly teach children the rules of the grammar. Chomsky reasoned that this argument of poverty of input is the primary justification for the saying that language is innate.

   (2) The second myth is that children learn to speak by imitating their parents. If this is true, then children should not be making any mistakes when their learn to talk. However children do make mistakes when they learn language (e.g. കാവളവണ്ടി, വെക്കള് )

Now it is time to look at two real-world cases where children reinvented language. One of them is the development of creole from pidgin languages such as pidgin English. Second one is the development of sign languages. In both these cases, phrases and crude sentences of a pseudo-language were converted to a bona fide language by the second generation of users i.e. the children of plantation workers and deaf children respectively.

Finally, another argument for the innate nature of language is that language is different from intelligence (or cognition). Those who suffer from Broca's aphasia are language-retarded. However they have sound cognitive skills. Those who suffer from chatterbox syndrome are language-savvy. However they have negligible cognitive skills. These two cases show that the ability to speak and the ability to, say, cook food are managed by different parts of the brain and hence different faculties.

Talking Heads, The Language Instinct, Steven Pinker

This post is a summary of the chapter "Talking Heads" from the book "The Language Instinct" written by the eminent linguist Steven Pinker. The previous chapter in the book was a discussion on syntax. This chapter focuses more on semantics and pragmatics.

Parse trees are different from parsing.  Parse trees define the syntax or structure of sentences in a language. The process of parsing defines the processing of a sentences and thus is related to cognition and semantics of a language. Chomsky demonstrated this feature of language processing using the classic example "colorless green ideas sleep furiously". This sentence is syntactically right. However it is rife with absurdity and does not make sense at all. In other words, the sentence is semantically not right.

The chapter then discusses the differences between a human being and a machine and how they interpret a natural language sentence. There are various types of sentences: (1) onion (or Russian doll) sentences, (2) garden-path sentences, and (3) ambiguous sentences.

The first two types are hard for human beings because humans are not good at memory as compared to machines. By memory, we mean short-term memory, something similar to a stack for a machine. On the other hand, the last type is easy for human beings because humans are good at decision-making.  The decision making employs different kinds of knowledge such as background knowledge, commonsense knowledge and world knowledge. A classic example that demonstrates commonsense reasoning is given below:

             Woman: I'm leaving you.             Man: Who is he?

The converse is true for machines. Machines can process onion sentences and garden-path sentences because of the availability of memory. However they perform very badly when it comes to ambiguous sentences. In addition to this, machines are too meticulous in parsing a sentences and identifies interpretations that a human being will never detect (e.g. pigs in a pen). 

In real-life, the task of text processing is worsened by the fact that dialogues are filled with short utterances, lots of pronouns, and fillers such as 'uh' and 'hmm'.

The chapter concludes with a discussion on pragmatics. Parsing a sentence involves more than simply understanding the sentence syntactically. A conversation between two parties can either be co-operational or adversarial. In co-operational conversation, the assumptions made by the speaker are also made by the listener. This phenomenon is absent in adversarial conversation. Legal documents demonstrate a form on adversarial conversation by clearly specifying each and every nuances of a contract. The following anecdote demonstrates the difference between the co-operational and adversarial conversation. Two psychoanalysts meet in the morning. The first psychoanalyst greets the other "Good Morning." The other wonders what he really meant by that statement.

Neural networks, explained - Janelle Shane, Physics World

This post is a summary of  the article published in the 2018 issue of "Physics World." This article is written by Janelle Shane.

Advantages:

1. They are excellent at recognizing patterns in multivariate data.
2. They are suitable for problems that are not very well-understood. Traditional systems were either rule-based or feature-based. However manually coming up with rules or features is intellectual challenging and infeasible in many cases such as face recognition. Neural networks are good at feature engineering. 

 Limitations:

1. Interpretability is an issue with neural networks. A neural network acts like a black box because humans cannot easily interpret the the features learnt by the model.
2. It is necessary to review the results by human experts because neural networks might learn features that are not at all relevant to the task at hand.
3. Neural networks might suffer from class imbalance in training examples. This is a major issue in the case of rare events, for which it is hard to generate sufficient number of training examples. 
4. Neural network might suffer from overfitting to training examples. Overfitting can be resolved by testing the network on unseen examples.

"Neural networks can be a very useful tool, but users must be careful not to trust them blindly. Their impressive abilities are a complement to, rather than a substitute for, critical thinking an human expertise."