Summary Presentation by Prof. Frawley on Language, Meaning, and Minimal Content

Summary prepared Jacy Ippolito and Jonathan Welch (dditions and comments by Frawley)

We began the lecture by drawing the connections and differences between a cognitive neuroscience structure and a computational structure. We went from looking at specific machines which were created to resemble what we might envision the human neuron to be "about" to talking about the specific neurocognitive "language machine." Specifically, we were asked the question, "What is the nature of the neurocognitive machine?" We then answered this question in two ways: with a computational answer and a neurocognitive answer.

Computational: We talked about computational machines with distinct memories, input/output relationships, and hardware/software, connections. These take the form of the models which Prof. Case demonstrated. Could our mind/brain be functioning in this manner?

Neurocognitive:The neurocognitive mind/brain would include (as Dr. Frawley points out) domain-specificity, modularity, encapsulation, innateness, shallow/rich code, and some minimal knowledge. This lead us into the discussion about the language module and its subunits.

The language module is thought to be comprised of distinct, separate parts: meaning, syntax, and phonology. Meaning is the interpreting module or component which is used to decipher the meaning behind expressions. Syntax is the module which allows us to combine forms into sentences. Phonology is the module which contains the articulation of expressions (phonemes, etc.) and which, combined with syntax, allows a person to produce expressions for others to interpret. These three parts connect through specific interfaces and work together in order to produce the spoken language as well as to interpret what is heard.

The language modules achieve success mainly due to the fact that they produce rules when given essentially positive input. This is a factor of its learnability or ability to acquire information. The fact that the language module continually revises its "algorithm" or general rules from the given positive information -- only up to a point -- would suggest that the "machine" is dependent on a rather rich code. It is important to note, however, that learning process is critical-period dependent, and that after a certain age, learning languages with the proficiency of a native becomes significantly harder, if not impossible.

The language module uses an abstract code and is universal on a core level; in other words, all humans initially begin with a core/schematic base which would in theory allow them to learn any language. The language module could also be thought of in part as a data structure with certain operations and separate domains subject to the learnability constraints imposed by critical periods and physical limitations. How do we know that the language module is domain-specific and encapsulated as well as based on a rich code? Well, one of the ways in which cognitive scientists try to prove these theories is by pointing to specific cases where the brain has been damaged in such a way as to detach the encapsulated modules from one another. When there is a lesion in Broca's area of the brain, subjects will produce broken sentences without articles or verbs and function words. In this case, only the nouns seem to be left. In this instance, it seems that the subject can understand and interpret what is heard, but he can no longer join the parts of sentences into completely coherent phrases. Another instance is when the Wernicke's area is damaged. This produces a speech pattern that is fluent and yet almost nonsensical, as if the meaning is no longer attached to the form of the sentences. Dr. Frawley also gave us the example of the subject whose frontal lobe had been damaged and could relay the story of the golden egg, but he had no coordination of his sentences within the conversation -- telling a story and then interrupting himself to relay unimportant facts. These precise, clear results from trauma cases suggest that the language module is indeed modular, but we then must ask ourselves: what kind of programs does the language "machine" output?

One way of looking at these outputs is by looking at how our expressions get meanings. The two meanings we discussed in class were semantic meanings and pragmatic meanings. Semantics is the study of meaning as truth. As Dr. Frawley puts it, semantics depicts the world: truth. An example of this would be a boy pointing at a cow and stating, "The cow is brown." This is a statement (assuming that the cow is indeed brown) that is true and hard to refute. Pragmatics is the study of meaning as use, or, as Dr. Frawley puts it, pragmatics brings about the world: significance. An example of this kind of statement is the one that Dr. Frawley used in class: a man yelling at a car that almost ran him over, "My brother's a lawyer!" This statement, whether true or not, conveys different meanings than if stated simply as a factual statement: it conveys a threat, anger, as well as an intention of a law suit.

Semantics andpragmatics are another way of conveying "aboutness." How are our statements "about" the world? Semantics studies the factual statements that connect us to the world. Pragmatics study the meanings and contextual differences between statements, informaiton that is variable and can possibly be cancelled. It would be hard to cancel semantic information, however. Pragmatics is based on the context and intentions of the speaker/hearer and cause the world to be a certain way. Semantics is based instead on matching the world. Green (in the textbook) defines pragmatics as "how to take the context of an utterance into account."

Pragmatics:

When a conversation is in progress, pragmatics is used constantly. There are four kinds of mutual knowledge that are taken into account by all the participants: community membership, physical copresence, linguistic copresence, and indirect copresence. Community membership is all the things a community knows and assumes others know. Physical copresence is things that we assume others know because they were physically present when an action occurred. Linguistic copresence is the things we assume others know because they previously communicated. Indirect copresence is the things we assume others know indirectly about another person. All four of these things are essential in communication between two parties. If these inferences are not made, the information would always have to be recommunicated. This would not be evolutionary advantageous.

Semantics:

Semantics is also very important in our conversations. It involves the actual true meanings of the words presented. The combination of semantics and pragmatics allows us to understand each other in conversations. The pragmatic meaning in a sentence and the semantic meaning can be confused, though. When someone says something sarcastic, this is an example of semantic meaning being used in a totally different fashion than is stated. Sometimes the listener is confused as to the true meaning of the sentence and takes it in its semantic or pragmatic meaning. This is why sarcasm is considered a higher level of communication. There are two theories as to how we are able to receive meaning from a word: one through rudimentary computational intellegence and one through intentionality. Computational intellegence is a list of all the properties that make up a certain object or idea and comes from logic. If you were talking about a cow, then these would be the cow "properties".These are all contained in a set. In order to find the meaning of a word used in a sentence, you take the words in the sentence and find where they overlap. An example of this can be found in the sentence "The cow ran." The idea of computational intellegence says that you would take the properties of a cow and the properties of run. You see where they intersect, and that is the meaning of the cow ran.

More technically, the computational view of meaning is that interpretation is a formal processs and can be handled by the mechnanisms of set theory and related devices only. There is no need for any content per se. So the meaning of "cow" would be the set of cow properties, whatever they might be. The meaning of "the cow ran" would be the intersection of cow and run properties and so on. Note how this gets us a minimal interpretation mechanism without committing us to a specific picture of the world.

More technically, you could have a minimal interpretation mechanism that relies on "just enough" stuff from the world in order to interpret expressions: boundedness, animacy, dimensionality, etc. If this is the core of interpretation, then it is a kind of rudumentary intentionality -- minimal knowledge of the world. This certainly has empirical validity in that all languages seem to use these concepts to interpret expressions. But it again falls into the trap of reductive definitions (see Landau's lecture): where does this information stop? How is it defined?