Linguistics and its Relation to Machine Learning

Posted by Somsukla Banerjee on Jan 16,2024

Text Data - Example

The term data (plural of the Latin datum which means a given entity) refers to qualitative parameters and/or quantitative values within collected information. Such information can be factual, descriptive and quantitative as per the following example

Mt Everest is the world’s highest peak - factual

Mt Everest is shaped like a three-sided pyramid – descriptive

Mt Everest is over 29,000 feet in height - quantitative

All three of the above sentences provide information about Mt Everest.  The first two sentences contain qualitative (factual or descriptive) data while the third sentence provides quantitative data about the world’s tallest peak.

In-depth Analysis of Qualitative Text for Implicit Meaning

Data vs. Insights – They’re Not the Same, Not Even Close- Benjamin Spiegel

The analysis of qualitative text data using logic, while a relatively mature discipline, has tended to be limited to explicit statements such as the ones above. Rarely is analysis conducted to derive deeper implicit meaning from such data. For example it is possible to derive from the sentence “Mt Everest is shaped like a three-sided pyramid” a conclusion that a pyramid can have more than three sides.

Such implicit derivations an have high value in a business scenario. For example test driver feedback comments such as smoother, faster acceleration etc in the context of the customer’s age, gender, nationality, income bracket, earlier car used and residential address can help the dealer to derive conclusions on the probability of the client making a purchase.

Such analysis of implicit meaning is possible by applying skills from a field of expertise known as linguistics. In the following sections this will be elaborated upon. A new field of data analysis called linguistic analysis is hereby identified for the current digital era of high connectedness, instant mass communication and exploding big data.

In the following sections we will introduce the discipline of linguistics and explore how insights from this area of study can have a huge impact in an era where we have massive amounts of data at our disposal.

What is Linguistics?

A linguist often encounters the following question “How many languages do you know?”

For most individuals, linguists are no different from polyglots - people who have mastered many tongues of the world. However a linguist may or may not be a master of multiple languages. Let us obtain some clarity about linguistics and the role of a linguist.

Five modes of human communication exist, corresponding to the five human sense organs which act as channels of information: sound, sight, touch, smell and taste. The information exchange and its interpretation using these modes, constitutes our overall communication. However the five modes are not equally relevant in the transmission and reception of meaning. In fact two of them, smell and taste modes hardly play any role in human communication. Unlike some animals, we do not routinely emit smells to communicate with others. There are very limited amounts of information from the outside world which we can receive via the media of smell and taste. In contrast the use of sound - the auditory-vocal mode is fundamental to the notion of language. The tactile and visual modes lie between these two extremes. They are often described as channels of non-verbal communication that includes body language and touch. However a significant bulk of communication in this age occurs using the media of speech and writing.

Language is one of the defining characteristics of human beings and its use lies at the center of most human activities and interactions. Of all the skills that human beings possess, language is the most quintessentially human. Practices and institutions that have been around for centuries, such as law, religion and science would not be possible if communication of meaning using language were not relatively easy. No other animal can claim to have any capability even approximating human language. In spite of the fact that language is a ubiquitous part of our lives and every day we are constantly surrounded by its spoken and written forms, most of us are completely unaware of its true nature and structure. As language is so closely integrated with our existence we use it unconsciously and tend to take it for granted. But this easy, largely unconscious skill is very deceptive, for all human languages are highly complex systems of communication, with greatly elaborated structures and rules.

Linguistics is the discipline that takes language as its particular object of study, to uncover its structures and rules and to understand how these are used in human acts of communication.

The Relevance of Linguistics in our Daily Lives

Language use is an inherently social phenomenon. How you speak depends on such factors as your upbringing, your racial and ethnic identity, whether you are a woman or man and your level of education. In other words you employ variation in language as a creative means of expressing who you are (and who you are not). By studying this variation, linguists enhance their understanding of language as well as their understanding of social processes, and discover the social factors that influence our linguistic choices and how these choices are perceived by others. Linguists who study the social aspects of language also investigate such topics as how and why languages change over time, how new languages are created when speakers of divergent languages come into contact, how language attitudes are used to maintain forms of discrimination, how conversations are social transactions, the relation between language and power, and the use of language in communication media.

Language use is an inherent human ability and is reflected in almost everything we do. By reviewing the work of linguistic researchers, one can see the questions raised by language use in everyday life. Do we speak the same way with our parents as we do with our peers? Do we write an academic paper in the same way we would write a text message? Why do some people speak differently from others?

These are some of the questions that we examine when we examine language used by certain people and by others around them.

What Linguists Study

There are different approaches to the study of language, eg, studying language structure and studying language use. Language can be studied purely from the perspective of its structure. Language is structured at different levels. For example we can look at language from the perspective of sounds: how sounds combine to form words, the internal structure of words, how words are combined to form sentences and what words and sentences mean. When a linguist considers the above mentioned areas as the object of his enquiry, he considers language as an abstract entity and studies language in isolation, divorced from the way it is used.

However, a linguist can also study the way language is used for communication, how people interact with each other and the relationship between language and society. As humans we hardly communicate with ourselves, our communication is with other individuals and groups who form human society.

The following diagram shows the various branches of Linguistics:


Much of linguistic study is centered about three broad questions:

  • What is the nature and structure of language?
  • How is language physically embodied and cognitively processed?
  • How is language used in society for purposes of communication?

When linguists study language as a structured, formal system, they investigate many distinct subsystems: the physical characteristics of speech sounds (phonetics); how sounds function together as part of a linguistic system (phonology); how words are formed and new words created (morphology); how words and phrases are combined to form a potentially infinite number of sentences (syntax); and meaning (semantics) and how context can add another dimension to meaning (pragmatics).
 Some linguists who focus on these aspects of language spend many years in the field investigating previously unstudied languages, many of which are now on the verge of extinction. By studying the properties of languages from around the world, linguists hope to better understand the properties shared by all human languages and also the ways in which languages differ. Thus their goal is to understand the nature of human language - how language "works."

Language is a universal human characteristic, and a component unique to the human mind/brain.  Therefore studying the nature of human language provides important insight into human cognitive abilities.  The celebrated linguist Noam Chomsky, who has revolutionized many areas of humanities and social science, was interested in questions such as: What do you actually learn when you "know" a language? And what do you know that enables you to translate the symbols you are now reading into meanings? How do children acquire language and why is learning a second language often difficult? Why is it so challenging to program computers to understand language? How might language have evolved in humans? How do our language abilities compare to other cognitive abilities? Linguists who explore language is a cognitive process, often conduct experiments in such areas as speech perception and production, language processing, and child language`acquisition, in order to better answer these questions.

In the world of big data and the digital transformation era, the third question “How is language used in society for purposes of communication?” is highly relevant. This is because a high percentage of the data that people and businesses are using today for digital communication is language data.

Thus while language structure, cogntitive processes around language and language use, are all important in themselves, in this paper we are interested primarily in language use.

In the following sections we will explore how insights linguistics in general and certain branches of linguistics such as Pragmatics and Discourse Analysis can have a huge impact in an era where we have massive amounts of data at our disposal.

Linguistics as a discipline can give enormous insights in understanding the natural language data around us. So far the potential of linguistics in understanding and solving matters pertaining to real life language use has been largely underutilized. Although there has been some interface between linguistics and communication in the corporate scenario, the full implications of a discipline as powerful as linguistics remains unexplored. This is true because most activities in an organization revolve around communication. A large part of our communication activities uses language as a medium and linguistics is the scientific study of language. Therefore an in-depth and sophisticated linguistic analysis can give us valuable understanding about the data around us. These insights will be useful for both profit making and not-for profit enterprises.  There has been utilization of certain basic principles of linguistics in the domain of sentiment analysis. In the following section, we give a brief description of sentiment analysis and the role of linguistics in sentiment analysis.

Sentiment Analysis and Linguistics

Sentiment analysis is an area where linguistics plays a crucial role. Sentiment analysis or Opinion mining is defined as the task of finding the opinions of contributors about specific entities. These days before buying a product, a typical consumer searches for reviews and opinions written by other people on various offerings and tries to detect the overall reaction towards a particular brand in terms of positive or negative comments. If the content has more positive than negative keywords it is concluded that the reaction of the contributor is inclined towards positive. Accurate sentiment analysis tools take into account the relation between words in a text and the semantic relation between them. We cannot always assume that a sentence that contains a positive sentence word is always positive. For example, the sentence, “the tool is everything butuseful” contains a positive word (useful) but indicates a negative sentiment. Therefore sentiment cannot be accurately presumed by word-by-word analysis, but rather on a level thatallows semantic interpretation. However language anaylsis is a complex exercise and there are certain factors that stop us from blindly relying on tools that are available for sentiment analysis. Some of these factors are:

  1. Context: a positive or negative sentiment word can have a different meaning depending on the context (Example: My students do a great job when it comes to copying from each other).
  2. Sentiment Ambiguity: a sentence with a positive or negative word does not necessarily capture a positive sentiment (Example: Can you recommend a good library?) Likewise sentences can express sentiments without using a sentiment word (Example: the browser uses a lot of memory). The first example does not express any sentiment while using the positive sentiment word “good”. The second example expresses a negative sentiment without using a sentiment word.
  3. Sarcasm: a positive or negative sentiment word can switch sentiment if there is irony or sarcasm in the sentence ( Example: I am delighted that my computer has crashed right before the examination).
  4. Language use: a word can change sentiment and meaning depending on the language use. This is often seen in dialects, slangs and other language variations.

The problems listed above indicate that there is a need for a deeper and more insightful linguistic analysis. In day to day communication we often understand what a speaker has not explicitly said. We as humans are endowed with the ability to understand the speaker’s intention from following a series of steps. Human communication involves the process of inferencing in a very crucial way. Therefore when we analyse linguistic data, we need sufficiently sophisticated tools that will be able to derive the implicit from the explicit.

A branch of linguistics known as Pragmatics is the study of language from the perspective of language users and the context of language use. In the following section we shall define and describe Pragmatics in some detail. We will explore how the tools of Pragmatics are very useful to understand and analyze language data as a component of Big Data

Pragmatics

Let us consider the sentence

The cat is on the mat.

If we just consider the words that make up the sentence, all of us will agree that it talks about a four legged feline creature who is sitting on a square piece of fabric.

However we will see how this simple sentence can mean different things for different language users in different settings.

For a family who owns a pet cat who is usually fed on the mat, if someone utters this sentence it means that it is time to feed the cat. The cat usually sits on the mat when it is hungry.

In the same family, if a member is looking for a mat to begin her floor aerobics, the sentence may mean that she has to remove the cat in order to procure the mat.

The same sentence will have a different meaning if we consider a third context. Imagine two police officers are waiting for a criminal and communicate with each other with the sentence “the cat is on the mat” about the arrival of the criminal.

The above example has shown us how meaning is actually dependent on the users of language who constitute the context of a linguistic utterance.

Pragmatics is the study of "speaker meaning" (or "utterance meaning)" rather than sentence meaning. As such, it is the study of how speakers and hearers use context.

Let us consider another sentence:

"All men are mortal"

This sentence can be understood if you know what "men" means [people], and what "mortal" means [do not live forever].

But in order to understand

"All you ever do is complain"

We need to know more than what the sentence says: we have to know who "you" are and you have to know that the sentence doesn't literally mean that all "you" ever do is complain.

So we need to know the context in which the sentence was spoken - what was said before, perhaps what was said after - and also when and where it was spoken, and by whom it was spoken. We also need to know what they had in mind in saying that, and what their relationship was with whoever "you" is. (The same remark can be understood as a gentle joke or an aggressive barb, depending on who is saying it to whom, where, and when.)

This is the kind of "mutual knowledge" that human communication depends on; language itself cannot do all the work. When two people are having a conversation, each one needs to know what the other one knows, and that the other one knows that they know, etc.

To have this kind of mutual knowledge about what each speaker has in mind it is necessary to know that the other speaker has a mind in the first place.

The Philosopher C.W. Morris coined the term Pragmatics in the 1930s. Pragmatics was developed as a field of linguistics in the nineteen seventies. Pragmatics focuses on what is not explicitly stated when two people communicate with each other and the interpretation of situational contexts.

Let us elaborate with the help of an example:

Jane: I heard that the new café round the corner makes great beef pasta.

Mary: I am a vegetarian.

When we look at the conversation exchange between Jane and Mary, It is possible to interpret that Jane was asking Mary to join her to try out the beef pasta that has been recommended to her. However Jane did not explicitly say so. However Mary understood Jane’s intention and conveyed her inability to join her. Mary’s reply was implicit and needed interpretation on Jane’s part. Jane had to infer that since Mary is vegetarian, she is not game to beef pasta.

Communication is an exchange where the interlocutors constantly interpret one another’s intention by making various inferences.

At this point, it is perhaps necessary to elaborate upon two dominant models of communication: the Message Model and the Inference Model.  The importance of Pragmatic analysis in linguistic communication will become clearer with the description of the two alternate models.

Traditionally, communication was regarded as an almost automatic coding-decoding activity. In the code model of communication, the hearer interprets the words uttered by the speaker in an almost machine like way. In this model, the speaker encodes a message and transmits  the message through a code ( language)  and it is the hearer who has to decode it. Thus according to this model, sentences of a language are just complex signals that encode messages.

This model makes a number of assumptions about the message encoded by the speaker. Some of the assumptions are as follows:

  1. The message is unambiguous
  2. The meaning of what the speaker is referring is determined by the meaning of the
  3. The speaker speaks literally
  4. The speaker speaks directly

However if we consider the conversation exchange between Jane and Mary discussed earlier, It is clear that in actual communication situation many of the above assumptions are violated. In order to understand what the speaker intends to communicate, more than a common language is needed. A shared system of beliefs and inferences must be in operation to function as communication strategies for effective and successful communication. In other words, the participants in the communication act engage in making a series of inferences.

Most of us are familiar with the term “inference”. To define it formally an inference is the act or process of deriving logical conclusions from premises assumed to be true.  It is a mental operation that forms an important part of most of our day-to-day lives and the decisions that we take. Readers of crime thriller fiction who are familiar with Conan Doyle’s immortal character Sherlock Holmes, are familiar with his techniques of drawing inferences from certain available premises. Here is an illustration of his inimitable manner of deriving a logical conclusion from a set of premises:

Holmes (to Watson): ...Now, it was not difficult, by an inspection of the groove between your left forefinger and thumb, to feel sure that you did not propose to invest your small capital in the gold fields.

Watson: I see no connection.

Holmes: Very likely not; but I can show you a close connection. Here are themissing links of the very simple chain:

  1. You had chalk between your left finger and thumb when you returned from the club last night.
  2. You put chalk there when you play billiards, to steady the cue.
  3. You never play billiards except with Thurston.
  4. You told me, four weeks ago, that Thurston had an option on some South African property which would expire in a month, and which he desired you to share with him.
  5. Your checkbook is locked in my drawer, and you have not asked for the key.
  6. You do not propose to invest money in this manner.

(Excerpted from the Adventure of the Dancing Men in the Return of Sherlock Holmes by Sir Arthur Conan Doyle, first published in 1905).

The above excerpt exemplifies the process of deriving a logical conclusion from a set of given premises. In what follows we will see how linguistic communication involves certain inferential techniques similar to one that w have described above.

According to the Inferential Model of Communication, linguistic communication is successful when a speaker recognizes the communicative intention. Let us revisit the conversation between Jane and Mary to illustrate this point;

Jane: I heard the new café round the corner makes great beef pasta

Mary: I am a vegetarian

Jane: I suppose they will have some nice vegetarian fare too

Mary: Shall we try it this evening?

In case of the above conversation exchange, both participants have successfully recognized the communicative intentions of one another. Jane recognized that since Mary is a vegetarian, eating beef pasta is not an option for her. However Jane still wants to visit the café with Mary and therefore signals about the availability of vegetarian dishes in the café. Mary too recognizes that Jane’s primary intention is to have Mary’s company. The choice of the meal is secondary. Having recognized that intention, Mary suggests that they try out the café in the evening. Therefore we see that for a successful communication to happen between two or more parties one needs more than encoding and decoding of meaning of words and sentences.

According to the inferential model, an utterance is a piece of evidence of the speaker’s meaning. Decoding the linguistic sentence meaning is just one part of the process of comprehension – a process that relies both on linguistic meaning and on the context in order to identify the speaker’s meaning.  The philosopher H. P. Grice was first to suggest this point of view, approaching the relationship between sentence meaning and speaker meaning in an unique way. According to Grice, speaker’s meaning is a complex communicative intention that must be recognized by the hearer in order to be fulfilled. It is an intention to achieve a certain effect on the mind of the hearer by means of the hearer’s recognition of the intention to achieve this effect.  Communication depends upon the ability of human beings to attribute mental states of others. Humans are able to interpret one another’s behavior not simply as physical actions but as the belief-guided fulfillment of intentions.

The inferential model connects the message with the meaning of the uttered expression by a sequence of inferences. This model involves a series of inference strategies that take the hearer from hearing the expression to the speaker’s communicative intent. Inference in this context is a mental operation that individuals perform in their assessment of others’ communicative and informative intentions on which they base their own utterances.

The role of language in inferential communication is to provide the communicator with evidence, as exact and complex as she wishes, of the content she wants the hearer to accept. For this, it is not necessary that the utterance encode this content extensivelyand unambiguously. Quite commonly, a fragmentary, ambiguous and loose coding is sufficient, in the context, for indicating a complete and unequivocal meaning. In this respect, inferential comprehension is not different from any other cognitive process of non-demonstrative inference that draws relatively reliable conclusions from fragmentary evidence open to multiple interpretations by relying upon both empirical regularities and context.
The main task of pragmatics is to explain how such a process of inference is carried out in the particular case of linguistic communication: what empirical regularities guide the process? How are the linguistic properties of the utterance, on the one hand, and contextual information on the other, put to use? Although different pragmatic theories give different answers to these questions, they agree on the two basic considerations: comprehension is inferential, and, by drawing on both the sentence meaning and the context, it aims at discovering the meaning intended by the speaker.

The notion of inference plays a very central place in pragmatics. If we were told that all my friends are smokers and if Susan is my friend, we can go on to conclude that Susan is a smoker. In deriving this conclusion we are effectively making explicit, the information that is contained implicitly within what we have been told. This extraction of new implicit information from given explicit information is the basis of the process of inference. The same process will be shown to underlie many of the pragmatic phenomena that we discuss in a subsequent section.  Before going on to discuss the various theories of Pragmatics, we undertake a brief discussion of how the notion of “context” is understood in Pragmatics.

Context

 We often come across the term context in discussion and elaboration of various theories of Pragmatics. In pragmatics we distinguish between three types of contextual information. These are as follows:

  1. Physical Context of the utterance: This encompasses what is physically present around the speakers and hearers at the time of communication. It includes various objects and people who are visible to the participants at the time of the communication.

          If a speaker says, I want that book or look at her, it is is accompanied by a physical act of pointing.  In this case the objects referred to are visible to both participants in the conversation.

  1. Linguistic context involves what has been said in the conversation so far. Both the participants in the conversation are aware of a certain topic being discussed.

When a speaker says I can’t believe she said that, here that refers to something that has been discussed earlier in the conversation.

  1. Social Context: The social relationship of the people in the conversation will determine the style of conversation. The choice of formal or informal style of communication is often determined by the social relationship between the speakers.  An act of communication is always anchored in a certain social context.

Pragmatic Theories

In the following section, we discuss the Pragmatic phenomena of Entailment, Presupposition, Implicature and the Theory of Relevance.

Entailment and Presupposition

Entailments and Presuppositions are two aspects of what is communicated through language but not explicitly stated.

Presupposition is an implicit assumption about the world or background belief relating to an utterance whose truth is granted in discourse. A presupposition is something that the speaker assumes to be the case prior to an utterance.

The utterance “John’s brother is waiting for him” presupposes that John has a brother.  When a hearer listens to this utterance, he automatically assumes that John has a brother without the speaker explicitly committing to the proposition.

To take another example if one says “John no longer writes fiction”, it presupposes the fact that John once wrote fiction.  The study of presupposition in in fact not only the study of natural language but the study of interactions between people and how they communicate with each other.  Stalnaker (1973) has defined presupposition as “A speaker presupposes that P at a given moment in a conversation just in case he is disposed to act, in his linguistic behavior, as he takes the truth of P for granted, and as if he assumes that his audience recognizes he is doing so”.  In any language there are some words or phrases which can act as the sources of presuppositions. These are called presupposition triggers.  Verbs such as realize, regret etc. are known as factive verbs. These act as a type of presupposition triggers. In the sentence  “John realized he was late” the verb realize presupposes the fact that John was late.

If we consider the relationship between two sentences:

Mary’s cat is cute (p)

Mary has a cat (q)

p>>q=p presupposes q

Entailment is defined as what logically follows from what is asserted in a sentence.  Consider the following sentences

Mary has a cat

Here Mary has a cat entails that Mary has an animal as cat is a type of animal.  Entailment is defined as a relation between a pair of sentences such that the truth of the second sentence necessarily follows from the truth of the first. One cannot assert the first and deny the second.

Consider the sentence

Susan’s sister bought a house

The above sentence presupposes that Susan exists and that Susan has a sister. It entails that Susan’s sister bought something. The entailments are communicated without being said and are not dependent on speaker’s intention. Presupposition seems to contrast with entailment. If X presupposes Y, the negative counterpart of X also presupposes Y.

For example The King of France is bald presupposes there is a king of France. The sentence the king of France is not bald also presupposes there exists a king of France.

If X entails Y, the negative counterpart does not entail Y. The president of Germany is a bachelor entails the president of Germany is unmarried. However if we say the president of Germany is not a bachelor, it does not entail thepresident of Germany is unmarried.

 Implicature

Entailments and Presuppositions are a certain kind of inferencing.

Under Implicature we look at another type of inferencing and at how speakers cooperate in a conversation to achieve a shared meaning for utterances. As we have seen speakers can communicate more than they can actually say. Philosopher H. P Grice under the label of conversational implicature has explored a special and interesting type of communication. By Implicature we mean what is not explicitly said but implied in a conversation situation. When people say more or less than what is expected they produce extra meanings beyond the literal meanings of words and sentences. This extra meaning is dependent on the context of the conversation. Let us consider the following dialog:

A: Are you going to John’s party?

B: I heard that Mary is going

In the above dialog, speaker B is implying something. He means that Mary is a factor to affect his decision about attending John’s party. If speaker knows about the relationship between John and Mary, he can infer correctly what B meant to say. Implicature is an indirect way of saying something.

Consider the following examples:

A: I have run out of gas

B: There is a gas station round the corner.

(implicature: B can fill gas in the gas station).

A: Have you finished the both the reading lists?

B: I have finished one

(implicature: B has not finished both the reading lists).

According to Grice implicatures are of two types – generalized or conventional implicatures and particualarized implicature.  If we consider the sentence John went into the house and saw a cat, from the convention of language it will imply that John went into a house that was not his. This is derivable from the phrase “a house” . We do not have to anchor the sentence in a particular context to derive this implicature.  The examples that we have cited above are examples of particularized implicature where a hearer needs to understand the utterance in a particular context.

Grice proposed that participants in a conversation mutually understand  each other’s implicature and they do so by cooperating with each other. In other words, participants in a conversation follow a Cooperative Principle (CP).

What does cooperation mean? Grice suggests that cooperating in a conversation amounts to obeying (implicitly) certain conversational maxims such as:

  • Quality: Do not say what you believe to e false. Do not say for which you lack adequate information.
  • Quantity: Make your contribution sufficiently informative for he current purpose of communication. Do not make your contribution more informative than necessary.
  • Relevance: Make sure whatever you say is relevant to the conversation at hand.
  • Maxim of Manner: Do not make your contribution obscure, ambiguous or difficult to understand.

These maxims are not always observed in communication. People may quietly violate a maxim or openly flout a maxim. Violating is quiet in the sense that it is not obvious at the time of the utterance that the speaker lied, supplied insufficient information or has been ambiguous or difficult to understand. Violations might hamper communication and they do not lead to implicature. It is the practice of flouting a maxim that leads to implicature. It is obvious to the hearer at the time of the utterance that the speaker has deliberately and quite openly failed to observe one or more maxims.

To take another example:  Consider the following excerpt from a recommendation letter for a prospective graduate student of Philosophy seeking admission in a university

“X has attended all my classes regularly and submitted all his assignments in wonderful handwriting. “

In case of the above example, The maxims of relevance and quantity has been violated. The referee has not supplied adequate information about X’s credentials as a Philosophy student. The person who reads this letter will be able to see this and perhaps decide against X’s admission.

Unlike entailments and presuppositions, implicatures cannot be made from isolated utterances. They are dependent upon the context of the utterance and the shared knowledge between the speaker and the hearer. In Grice’s analysis, the speaker’s flouting of a maxim combined with the hearer’s assumption that the speaker has not really abandoned the cooperative principle leads to an implicature.

Relevance Theory

Relevance theory is another  inferential theory of communication which aims to understand how the audience infers the communicator’s intended meaning.  The Relevance theory was proposed by Dan Sperber and Deirdre Wilson.

The relevance theoretic explanation of these inference processes is rooted in an account of cognition. An inferential theory of communication explains communication in the following terms

The speaker intends to communicate some information I, and produces a stimulus which enables the addressee to identify this information by recognizing the speaker’s intention to convey it. Communication involving the production of ostensive stimuli is called ostensive communication.

The basic point about ostensive inferential communication is that the communicator intentionally provides evidence that he intends the audience to arrive at certain conclusions. In other words, the communicator is attributed two intentions, to inform the hearer of something which is called the informative intention, and the intention to inform the addressee of this informative intention, This is called the communicative intention. Let us consider an example:

I am walking on a road that leads to the railway station. I see my friend John who is walking towards the station carrying a suitcase. I wave at John and he waves back and while looking at me points to his watch with a worried expression. I understand that John is in a hurry to catch a train and does not have time to talk to me.

My conclusion was based on:

  1. Someone walking to the station with a suitcase has a train to catch
  2. John is pointing at his watch and looking worried
  3. John is rushing to catch the train
  4. John may miss the train if he is delayed
  5. John cannot be delayed and this is not the right time to talk to him.

I was successful in understanding John’s intention to communicate by drawing a series of inferences and making use of the appropriate contextual information. Sperber and Wilson argue that ostensive inferential communication exploits a general cognitive principle: the mind tends to attend primarily to information that is RELEVANT. At every moment the mind is confronted with more information than it can attend to. To operate efficiently, it should attend to those information that helps it to improve upon its existing knowledge base. The human cognitive system should pick out information which connects existing assumptions in such a way as to improve the individual’s overall representation of the world by making it richer, better evidenced  and more likely to be true. In Sperber and Wilson’s terms such information produces positive cognitive effects and are of three main types: contextual implication, contextual strengthening and contextual contradiction and elimination. Human cognitive system should pick out and process that information which achieves the greatest positive effects for the smallest processing effort. Sperber and Wilson call this information which has this property RELEVANT information and formulate the cognitive principle of relevance which is as follows:

The Cognitive Principle of Relevance

Human cognition tends to be geared to the maximization of relevance.

The mind is set up in a way that it automatically attends to the most relevant seeming information. The cognitive principles of relevance has important consequences for communication. Communication involves production of an ostensive stimulus, some intentional behavior which requires some processing effort by the addressee. If the communicator wants to be understood then  she must produce ostensive stimulus in such a way that it will seem relevant to the audience under the intended interpretation. Thus it can be said that every act of ostensive communication creates in the audience a presumption that it will be relevant for the audience’s attention. According to the communicative principle of relevance, every ostensive stimulus in effect conveys to the audience the claim that it is optimally relevant without having to waste any processing effort.

Let us illustrate this with the help of an example: Mary and Sam are about to go on a weekend trip by car and they are busy packing. Sam says to Mary:

The car is open

Intuitively Sam conveys that the car is unlocked and Mary can start loading luggage without waiting for Sam to unlock it. While the first proposition “ the car is unlocked” is expressed explicitly, there is an implicature in it. Mary can start loading the luggage. Let us see how relevance theory explains how Mary arrives at this interpretation.

Ostensive stimulus by Sam – The car is open

According to relevance theory this information should be optimally relevant to Mary with the least processing effort. Mary has easy access to the information that they own a car, they are going on a trip, the car is normally locked when it is parked etc.

From the available knowledge of the context, Mary can draw the possible interpretation

If the car is unlocked, Mary can start loading the car without waiting for Sam. The contextual implication has other cognitive effects such as

If Mary can load the car without waiting for Sam, then the loading of the car will be faster

If the loading of the car is faster, Sam and Mary can start for the trip early. Hence Mary is able to gain many cognitive effects for little processing effort.

Within Pragmatics, the communicative principle of relevance is specifically applies to verbal communication. However this principle can be applied to many forms of communication in the present times such as emails, blogs, letters, posts on social networking forums. The addressee undertakes an interpretive task which aims at selecting the most appropriate interpretation from the range of interpretations that the utterance might have in a given context.  Hearer’s construct the most relevant interpretation by following the general inferential procedure:

  1. Follow a path of least effort in constructing an interpretation
  2. Stop when the expectations of relevance are satisfied.

Other Branches of Linguistics

In this section we will examine the importance of Discourse Analysis, Critical Discourse Analysis, Sociolinguistics, Sociology of Language and Cyberpragmatics and highlight the importance of these disciplines in the present context. Discourse Analysis is a general term for a number of approaches to analysing written, vocal or sign language. The objects of discourse analysis- speech, written text or a communicative event are defined and analysed in terms of sentences, propositions and turns-at-talk. Contrary to traditional linguistics, discourse analysis studies language use beyond the boundary of the sentence and prefer to analyse naturally occurring language use and not invented examples. Discourse analysis aims to uncover the socio-psychological characteristics of the persons involved in a communicative act from the structure of the text. Related to Discourse Analysis is Critical Discourse Analysis. Critical Discourse Analysis or CDA is a form of discourse analysis that studies the relationship between the discourse and ideology (set of beliefs, attitudes and behaviors that constitute a perspective on the world. It focuses on critiquing social injustice and is strongly linked with language and power. By analysing a spoken and written interaction, one can understand the dynamics of power that exists between the interlocutors. The methods of Critical Discourse Analysis can be utilized to uncover implicit information from explicit text. It is possible to decipher crucial iinformation from the lexical or grammatical choices made by the author of the text. Most times, linguistic choices are not random and have been used on purpose to convey a particular idea. Here are a few examples:

  • Use of Active or Passive Voice: The use of an active verb gives a clear idea of who performed an action and to whom, for example: Police attacked the protesters, The use of passive voice puts the focus on the action rather than the actors. If we say “ Protesters attacked”, the focus is on the victims. Sometimes the use of passive voice can be used to shift attention from the actor and is often used when the author is unsure about who performed the action or wants to suppress this information.
  • Naming: The ways in which people are named can perpetuate ideologies, mindsets and attitudes. For example, the newspaper headline “Five Asian men involved in robbery” creates a very different impression from “ Five young men involved in robbery”.
  • Indirect quotes: This is particularly common when the results of polls are being used, for example, poll shows 70 % oppose gay marriage, however there may be no example of reported speech saying this.

In a subsequent section, we will examine real life language data from hoax e-mail messages and examine the importance of discourse analysis in the study of language.

Sociolinguistics is the descriptive study of the effect of various aspects of society including cultural norms, expectations and context on the way language is used. Sociolinguistics differs from Sociology of Language in terms of its focus. While Sociolinguistics studies language and its effect on society, Sociology of Language studies the effect of language on society. Sociolinguistics overlaps considerably with Pragmatics as “context” plays an important role in both these fields of enquiry.

Sociolinguists are interested in how language varieties differ between groups separated by certain social variables such as ethnicity, gender, social class, education and age. Sociolinguists can possibly determine through study of social attitudes that a particular variety of language use, for example the use of slangs would not be considered appropriate in certain professional domains. Insights from sociolinguistics can help us categorize individuals according to the social dimensions such as ethnicity, age and level of education among others. We can use text data sources to determine important information about the individual/s who created the text. In a way this is one form of deriving implicit information from explicit text data.

Semiotics is the study of meaning making with signs and symbols. Semiotics is closely related to the field of linguistics which for its part studies the structure of linguistic symbols. As different from linguistics, however semiotics also studies non-linguistic signs. These days a substantial portion of text data that we encounter is exists in conjunction with other non-linguistic data such as pictures and other visual symbols, a semiotic analysis will be helpful to determine the overall meaning of a text. We should consider and examine text data along with other non-linguistic symbols present in the context, wherever applicable.

Cognitive Linguistics refers to the branch of linguistics that interprets language in terms of concepts,underlying structures and schemas that underlie external forms of language such as words and sentences. Cognitive Linguistics studies linguistic cognition that underlies the use of language and claims that the underlying forms of language that exists in the mind of its users is a consequence of language use. A central claim of Cognitive Linguistics is that language and cognition mutually influence each other and are embedded in the experiences and environments of the users.

Cyberpragmatics , a term  coined by Francisco Yus, has recently come into being, expanding extensive earlier work on Internet-mediated communication (Crystal,2001).  Cyberpragmatics draws important insights from the fields of Pragmatics and Cognitive Linguistics. Cyberpragmatics is defined as the cognitive pragmatic analysis of Internet interactions. Yus claims that Sperber and Wilson’s Relevance Theory (discussed in an earlier section of this paper) is a useful framework for explanation of user-to-user communication on the Internet.  Yus has examined the virtual community and presentation of self-identity in virtual settings and provided important details about the nature of language use in asynchronous web environments such as blogs, social networking sites and Twitter.

Towards Machine Learning: Language Science and Computational Linguistics

As a key outcome of our discourse, we introduce a new discipline we call “Language Science” which we define as the systematic study and analysis of language, its usage, effects and impacts.

Language Science draws from a number of fields of study and applies the knowledge to real life situations in order to solve real life problems pertaining to language. Some of these fields are named below.

  • Sociolinguistics is the study of language use in society and social networks; 
  • Cognitive Linguistics is the study of how the mind represents and uses language. 
  • Pragmatics studies how context contributes to meaning. 
  • Semiotics is the study of meaning making through signs and symbols
  • Discourse Analysis reveals the socio-psychological characteristics of a person by analysing written, spoken and sign language. 

Of the above Pragmatics and Pragmatic Theories have been discussed at length in previous sections as a key field for conducting in-depth inferential analysis of implicit meaning which even today’s Large Language Models (LLMs) are unable to address.

Language is the medium of most of our communication activities. Without language it will be impossible to share our thoughts and ideas with others. Language is inextricably fused with our beings but we rarely give it a second thought. Communication constitutes 80 per cent of a Business Executive’s role. Most of such communication is via the medium of language. A chance remark or ill-chosen phrase is enough to make or break the reputation of a business.

However if we understand and analyse how language works, we can draw conclusions from business communication data, written and verbal, with clients and internal. Such conclusions can be of enormous business benefit. Some of the benefits of applying language science to your business are

  • being able to draw inferences of implicit meaning and “hearing what isn’t said”.
  • being able to comprehend how and why certain uses of language are likely to be more effective than others.
  • being able to use language to shape people’s perceptions and influence their subsequent actions .
  • being able to minimize miscommunication
  • being able to overcome cultural barriers through effective use of language
  • being able to build better relationships
  • being able to build and enhance your social capital by effective use of language.

In an effort to achieve automated comprehension and processing of language computer scientists have worked for several decades to develop a knowledge discipline known as Computational Linguistics.

Language Automation, Language Models using Deep Learning Algorithms

With the advancement of the discipline of Computational Linguistics there has evolved today the so-called ‘Natural Language Processing' (NLP) automated software tools. Well-known examples of such tools are those used for

  • Conducting “sentiment analysis” (mostly at ‘polarity’ levels of positive, negative and neutral) upon text data of client surveys and
  • Detecting fraudulent email communication using keyword-matching and rules.
  • Generating text based on natural language ‘prompt’s – these are called ‘Large Language Models’ (LLMs) which are built using Deep Learning algorithms. LLMs are in high focus in the current era.

While a more detailed blog on LLMs and Generative AI will be available later in this series it should be noted here that there is much more work that needs to be done on LLMs before they can achieve the capabilities of comprehending and processing the more subtle aspects of language described earlier within this discourse.

Related Posts:

Jul 06,2023

What is AI?

Artificial intelligence (AI) is human-like intelligence exhibited by machines or software. It is also the name of the academic field of study which s

By Tirthankar RayChaudhuri

Dec 12,2023

Do We Need Ethical Considerations for AI?

There is much paranoia and pessimism about AI misconceptions being promulgated in an irresponsible manner by individuals who are ignorant about what A

By Tirthankar RayChaudhuri

Jan 15,2024

What You Need to Know About MLOps

Most heavy-duty industrial operations today have monitoring systems in place which gather data records (logs) of the daily operational cycle. Such da

By Tirthankar RayChaudhuri

Nov 13,2023

All the False Stories You Have Been Hearing about AI. Science Fiction and Smart Weapons

A number of negative statements which are misleading in nature about the discipline of Artificial Intelligence have been written and circulated in the

By Tirthankar RayChaudhuri