It is so frustrating: You are talking to a friend when suddenly you cannot remember the name of something. It is on the tip of your tongue but no matter how hard you try, you cannot say what it is. Even more maddening, you know so much about it: It is an animal that lives in South America. It gives wool that is sometimes made into sweaters. You met someone in California last year who raises them. It is an, .. er, um…

To scientists who study the brain, this is a tip-of-the-tongue experience. It even crops up in a slightly different form among users of sign language. They call it a tip-of-the-finger experience.

"Humans love to talk," says Dr. Willem Levelt, director of the Max Planck Institute for Psycholinguistics in Nijmegen, The Netherlands. "Most of us spend large parts of the day in conversation. If we are not talking to others, then we are talking to ourselves."

Dr. Levelt’s interest is in the problem of how people go from thinking about something to actually saying it. In the course of studying the component systems involved in generating spoken words from thought, he has developed a theory about what is happening inide the brain when a speaker blocks on a ord.

The identification of these systems is base on advanced techniques for imaging the brain, he said. These two techniques allow the researchers to watch the process whereby thoughts are transformed into a seamless flow of words, to see where the process breaks down, and to show how people correct errors on the fly.

Dr. Levelt and his colleagues base their research on the assumption that the human brain contains distinct modules for processing thought into language. Dr. Gary Dell, a professor of psychology at the University of Illinois, said that this assumption is widely accepted among psycholinguists.

The exact anatomy of the proposed modules is not yet known, Dr. Levelt said in a telephone interview, but their existence can be demonstrated experimentally. They are not little boxes in the brain, he said, but widespread networks of interconnected neurons that cooperate, with precise timing, to carry out specific tasks.

Dr. Levelt calls the three modules the lexical network, the lemma network, and the lexeme network. Essentially, the lexical network handles thoughts, the lemma network handles syntax, and the lexeme network manages spoken sounds.

In speaking, the first module to be activated is the lexical network. "Conceptualizing is deciding what to express, given our intentions," Dr. Levelt said. "As speakers, we spend most of our attention on these matters of content" and how to order thought sequentially.

Once a message is thought-out, he said, "we must capture it by some lexical concept." To do so, we dip into our stored vocabulary -- typically tens of thousands of words. Speakers can retrieve two to three words per second containing 10 to 15 speech sounds.

Imagine you want to say the word llama, Dr. Levelt said. Perhaps you saw a picture of a llama, or you thought of the animal while talking to a friend. The mind first activates the lexical node for llama, which contains everything you know about llamas. It is an ungulate with a long neck, it is used as a pack animal, and so forth.

When the lexical node for llama is activated, nodes for words of similar meaning are also stimulated. These might include the nodes for sheep and goats, nodes for beasts of burden in general, nodes for hoofed animals, and so forth. At this point, you still don't have the word for llama. But you have activated a great deal of information about llamas and similar animals.

The next stage in processing is handled by the second module, the lemma network. When the lexical concept for llama and other activated concepts are passed to this level, two things happen.

First, the lemma assigns proper syntax to each incoming concept. These are the rules of the speaker's language, including word order, gender if appropriate, case markings and other grammatical features. Also at the lemma level, verbs, nouns, and modifiers are put in their proper place in a word string.

Second, the various activated lexical concepts engage in a competition. Most of the time, the most highly activated concept (llama) will win. But sometimes there is interference from other lexical concepts. The more that are activated, the longer it takes to generate the desired word.

Timing experiments done in Dr. Levelt's lab show how this works. Subjects are shown picture and asked to name them as fast as they can. The average naming time is 700 milliseconds. Then the experimenter adds a distraction -- such as muttering the word horse when a picture of a cow is presented. People need 800 milliseconds on average to name cow when the lexical concept of a horse is also activated.

The third part of the process is to turn a chosen lexical concept into a spoken word. This is called the lexeme level. "Accessing the lexeme is harder than you think," Dr. Levelt said. The mind has to find the correct sound and match them to the syntactic elements in the lemma network. This is where the process of generating thoughts into speech can fail, Dr. Levelt aid. Many things can go wrong.

One is the tip-of-the-tongue phenomenon. You are at the lemma level but the word refuses to come. You know a lot about it. You might even know it has two syllables with the stress on the first syllable, which suggests that part of the lexeme information is accessible.

People who speak languages with masculine and feminine words almost always know the correct gender of the missing word, said Dr. William Badecker, a research scientist at Johns Hopkins University who works with brain-damaged patients who have trouble naming things.

Why do words become blocked?

"We don't know for certain," Dr. Levelt said. But one idea is that a given lexical node is not sufficiently activated to spread to the lexeme level where speech sounds are stored. Thus llama might not win out over sheep and goats, leaving the speaker fumbling for the word. But as most people have discovered, waiting for a few minutes will help retrieve a word on the tip of your tongue.

Dr. Dell explains what happens: "Say you are trying to remember the name of that funny stuff inside a sperm whale, the stuff used in perfume. You may think it sounds like amber but you know it's not. But you keep thinking amber, amber, which makes the amber part of your network activate. Eventually, you give up and think about something else. Later, when you think about it again, the word may suddenly appear -- ambergris."

Other kinds of speech errors can occur in the transition between the lemma network and the lexeme network, Dr. Dell said. Sometimes people exchange one word for another (Fill up my gas with car) or mix up speech sounds (queer old dean instead of dear old queen).

So-called Freudian slips of the tongue are also common. Freud thought that they represent deep sexual urges but they are more innocent. Dr. Levelt said. While talking, people are often thinking about other things,  which can caue an unrelated lexical node to become activated.

But if all goes well and a word is retrieved correctly, it goes to the next level of processing which is articulation, Dr. Levelt said. This is the process whereby the syllables are mapped into motor patterns generated in the tongue. lips, mouth, larynx, and lungs.