Accounts, Generalizations, and Theories
In addition to content, we can look at research on language from the perspective of what it is trying to accomplish and how it does that. First, we need to go back and consider again what science is all about. As we've seen, a scientist starts with a phenomenon of interest, gathers some data on the phenomenon, and attempts to come up with a description or explanation of it. This may take the form of a discussion in some language (such as English or Chinese), a set of equations, or an algorithm, that is, a precise description of a set of processes to be carried (by a person or a computer). Whatever form it takes, this result of the scientist's research may be referred to as an "account" or an "analysis" of the phenomenon. A scientific account is expected to include some sort of generalization about the phenomenon, that is, to go beyond simply listing the data.
A generalization can be relatively specific, applying only to a vary narrow range of data. For example, an anthropologist studying kinship relations in some ethnic group might conclude that the relation between a mother-in-law and daughter-in-law matters more than other in-law relations in the culture. However, most often, an anthropologist wants a particular culture to be seen only as an example of a more general phenomenon. That is, the goal is not just to describe or explain one culture but to say something general about human culture. So the anthropologist might want to state a generalization about in-law relations across all ethnic groups belonging to some type or to all ethnic groups. In the most general cases, the generalization is usually stated in terms of a particular theory, which is a general set of principles for understanding phenomena of a particular type. A theory is supposed to offer an explanation of the phenomena, not just a description. For example, kinship theories start with a set of basic categories that are supposed to be sufficient both for describing and explaining the role of the different possible kinship relations in all societies. A theory is like an "account" or an "analysis", only more general.
How Scientists Describe and Explain: Generalizations and Theories
To take a linguistic example, a researcher might be interested in describing the way present-tense verbs are negated in English dialects like the one referred to by A and B in the box in the last section:
- I/you/he/she/it/we/they don't go
The linguist could make the low-level generalization that for all but a few verbs in these dialects, the present tense negative is formed by putting don't before the base form of the verb (go, mean, eat, etc.). This generalization would apply only to the dialects being investigated.
Or the researcher could note that there is a distinction in the affirmative in these dialects that is not made in the negative, the same distinction that is made in standard English dialects:
- I/you/we/they go
- he/she/it goes
Explaining Languages, Explaining Language
Going beyond these English dialects, the linguist might then discover that something similar happens in many languages, for example, among the languages that this book focuses on, in Japanese and Amharic, making the generalization that languages tend to make more distinctions in the affirmative than the negative. The researcher could then go even further and try to place negation within a set of other forms in terms of how likely they are to favor distinctions; for example, the researcher might discover that fewer distinctions are made in verbs in subordinate clauses (for example, in the places they go) than in independent clauses (for example, in they go many places) in many languages. (Don't worry if you don't understand what "subordinate", "independent", or "clause" means.) Even more abstractly, the linguist could try to explain why there would be such tendencies across many languages. For example, they might propose that it is more difficult to produce and understand negative forms than affirmative forms and that this puts pressure on the users of the language, and hence on the language itself, to compensate by making some other aspect of the grammar simpler. This kind of proposal would be a theory that is designed to describe and explain a set of grammatical phenomena across many languages.
To summarize this section so far, we see that a scientist trying to understand language makes generalizations about data. These generalizations can apply only to one language, or they can apply to language in general. A linguist or other language scientist often works within the context of a particular theory of language. I'll have more to say about the role of theories in this section. But we've said nothing so far about what the goal of a description or explanation is. There are two kinds of possibilities, related to the two fundamental ways of looking at language that are mentioned in the overview of the book: focus on the product (sounds, words, sentences, etc.) and focus on the process (speaking, reading, understanding, etc.).
Consider this line from the English comedy show "The Two Ronnies":
- She left her husband for the garbageman.
The joke, in case you missed it, is based on the ambiguity, that is, the multiple possible interpretations, of the verb left. The sentence could mean that the woman abandoned her husband in favor of the garbageman, or it could mean that she put him outside for the garbageman to carry away. It turns out that ambiguity is quite common in language and is the basis of puns such as this one. What sort of a problem does this present for a description or explanation of how language works? What aspects of language would an explanation of ambiguity need to refer to?
All instances of language are obviously the result of processes: speakers, writers, and signers (in the case of sign language) produce something that we call language and hearers, readers, and sign observers attempt to understand something that has been produced. Over a longer time scale, what a person knows about a language (that is, how to produce and understand it) changes; this is the process that we call learning, development, or acquisition. Over an even longer time scale, every language also changes; the English of today is not the same as the English of 1900. This slower process is called language change. Finally, over the longest time scale that is relevant for language, we know that at some point in the distant past, for example, 200,000 years ago, the ancestors of modern humans did not have anything like what we call language; the process that resulted in the kind of system we have now is called language evolution.
Most linguists choose to ignore all of these processes and focus instead on the products, the words, sentences, and entire discourses that are produced and understood by people at a particular point in historical time. In their research they attempt to describe and explain these products, to generalize about what must appear, what may appear, and what may not appear. Other language scientists, especially psycholinguists and computational linguists, focus instead on the processes themselves. In their research they attempt to describe and explain these processes, to generalize about what is going on during language behavior, language change, or language evolution, for example, when a speaker pronounces a word, when a child learns how to combine words into sentences, when a population of agents "invents" grammar in the process of evolving a communication system. Their accounts and theories often take the form of algorithms and are often implemented in the form of computer programs. They may be called processing accounts or computational models.
There are at least two important differences between these two ways of looking at language. First, the product-oriented perspective deals with static objects; even though it took time for the words or sentences to be produced or understood, the things being studied have no real time in them, except in that certain parts come before other parts. In the process-oriented perspective, on the other hand, time cannot be ignored. Processing happens in real time, and the implementations of processing accounts as computer programs obviously run in real time. These accounts differ a lot in terms of how serious they are about the time course of human language processing, but they are all in some sense dynamic.
Production and Comprehension
Second, processing accounts are directional. At its most basic, language processing is either production or comprehension. Even processing accounts that are concerned with the slower processes of learning or evolution are based on some idea of how production or comprehension takes place. I'll have more to say about production and comprehension later in this chapter. For now the important idea is that these processes occur in opposite directions.
In production, a speaker, writer, or signer starts with something to be communicated (perhaps to themselves) and then goes through a process that results finally in an instance of language (a spoken, written, or signed utterance of some kind). When I produce the sentence take the garbage out, I start with something that isn't language at all, something that doesn't include the words take and garbage but is more like my mental representation of some situation in the world, either one I'm experiencing (seeing the garbage piling up in the house) or one I'm imagining (seeing the hearer taking the garbage out). Then somehow I get from this thought to the utterance itself. In comprehension, a hearer, reader, or sign observer starts with an instance of language and then goes through a process that results in some sort of approximation to what the speaker, writer, or signer wanted to communicate. When I hear somebody say the sentence take the garbage out, I start with some sounds and from these eventually figure out what the speaker wants and what I'm supposed to do, and of course what I'm supposed to do isn't language at all; it's an action. The product-oriented perspective ignores this directionality, treating it as irrelevant to what makes language work.
So who is right? Does the product-oriented perspective or the process-oriented perspective give us more insight into how language works? To some extent, the answer depends on what we're after. If we want a way to describe languages in as efficient and understandable a way as possible, then it may be that we can confine ourselves to product-oriented research. The outcome of this research could be an archive of many languages in a form that would allow researchers from different fields and people who want to teach or learn the languages to consult the archive. Of course if our goal is to understand what people are doing when they are actually using language or if we want to write programs that allow computers to use language, then we will need to rely on process-oriented research.
But what if we are interested in the more abstract and theoretical question of why language is the way it is? Which perspective is the right one, or do we need both? There is a lot of controversy within the language sciences on this point, with one camp, associated especially with the famous linguist Noam Chomsky, claiming that we can learn what makes language special by studying products alone. The idea is that processing is something separate, something to be understood in its terms, but not something we need to refer to to understand how language works. The opposing camp, associated with the theories in linguistics known as cognitive linguistics and functionalist linguistics and with many language researchers in fields outside of linguistics proper, takes the view that the nature of language is intimately tied to the way it is used, that if we want to understand why language is the way it is, we need to refer to processing. As I discuss more in this section, this book belongs more in the second camp. I will assume that both product-oriented and process-oriented perspectives can help us understand how language works.
Let's return to the issue of ambiguity, illustrated in the box above, to see how thinking about process as well as product can help us figure out what is going on. For the sentence in question, a product-oriented approach might simply include the information that left (or leave) has at least two meanings, though of course it would have to be more precise about what is meant by "have two meanings". A process-oriented approach would look at the processing of the sentence from both directions. This would make clear that ambiguity is a "problem", that is, a potential challenge for a person or a computer, in the comprehension direction but not necessarily in the production direction. It is such a problem in fact that a very large body of research from the processing perspective has looked at what is called disambiguation. The problem is that when a word is ambiguous, a listener (reader, sign observer) has to figure out which meaning is intended (or, in the case of a joke like the one in the box, to see that the sentence has a possible interpretation for each of the meanings). It is easy to imagine how a person hearing a sentence like the one in the box would know that the word left has more than one meaning. What is hard to explain is how the person knows which meaning is the right one, or, in this case knows that both are right. Theories of disambiguation are designed to explain the process. We'll return to disambiguation later; the important point for now is that it only becomes an issue when we look at language from a processing perspective.
In this section and the last, we've seen that the business of linguists and other language scientists (and the subject of this book) is trying to describe and possibly explain particular languages, language in general, and how people use languages. But this is not the only way we might treat language. Some people who are not scientists and may not be particularly interested in the scientific study of language do talk about language as a part of their work. And one of their concerns may be deciding what people should and should not say and trying to enforce these decisions. This kind of work, some of its consequences, and how linguists sometimes comment on it are discussed in the next section.