Stephen van Vlack
Sookmyung Women`s University
Graduate School of TESOL
Introduction to Linguistics
Fall 2006
Week 5 - Questions for discussion
(Intro to Ling, pp. 120-141)
1. What is categorical perception and how does it work?
Categorical perception is one of the cornerstones of language (and certainly human cognition) and it occurs on all levels. What it basically says is that humans are programmed to make judgement calls between but not within categories. This means that all humans can distinguish between the sounds /p/ and /b/ (as long as they have them as phonemes in their language), but have terrible difficulty deciding what a good /p/ or a good /b/ might be as opposed to a bad one. This might sound a bit confusing, but it really not so difficult. It is important that we go over this because the ability to break things down into categories is the one of the most basic skills needed for any kind of perception, and by extension, cognition.
From a basic cognitive viewpoint the ability to form and identify categories is the ability to generalize. Everything we know and perceive as humans is broken down into one category or another based on a set of features which we have stored in our heads. In some cases, these features might be very much idiosyncratic, based entirely on the experience of the individual, but in the vast majority of the cases categories are broken down into features that are very much universal, similar or even identical across time and cultural dimensions. The physical constraints of life on this planet restrict our minds and perceptions in a profound way and force us to come up with predominantly similar, but certainly not identical, categories.
The first ground-breaking work on categorical perception was carried out by Brent and Berlin back in the 1970s and early 80s in the area of color terms. After many years of vigorous cross-cultural and cross-linguistic research, they came to the same conclusion as has been reported by Radford et al. (1999); namely, that it was universally hard for people to identify the boundaries of a certain color, but they were able to distinguish colors from each other quite readily, but not equally. They found that there was such a thing as a good example of a color and a bad example of a color. This distinction lead to a vast array of research in the area resulting in Rosch`s prototype theory and ultimately Lehrer`s semantic field theory. Prototype theory says that in a person`s mind there is a prototype (a perfect example) for each category. This prototype is made up of all the best possible examples of all the features which make up the prototype. For semantics, these features can be posited to create a semantic field around which the category is enclosed. It is the field itself which creates the fuzzy border around the category, but borders are hard to maintain and identify because categories overlap on top of each other to some extent and are therefore fuzzy. Before we get too caught up in this, let`s go back to speech sounds.
Each phoneme is its own category. In the generative model the features which make up the category are the distinctive features, which we discussed in class last week. Each phoneme has its own specific configuration of those distinctive features, but there is also some amount of overlap. This overlap makes it hard to perceive the borders, but the central concept of each sound remains intact and easy to identify. This works the same in each language although each language has a different phonemic inventory.
This ability to identify categories also seems to be a general cognitive trait of humans and is not restricted to language alone. The difference with language, as Chomsky would argue, is that the features we need to create the categories, no matter what the level, are born with us. That means, we are all born with a set of distinctive features in our heads and other concepts that allow us to conceptualize the world more easily. Linguists have called these semantic universals or primes and they are believed to be the underlying features of which concepts are created and, according to standard generative thought, we are born knowing them. Categorical perception is absolutely essential for being able to understand language basically because when other people speak they speak differently than we do. Everybody has her own different way of speaking. Since we're mostly concerned with sounds here we will focus our attention on sounds specifically. Because everybody creates sounds in a slightly different way this makes it impossible for us to perceive their sounds the same way that we perceive our own. In order to make sense of what people are saying we, therefore, need to process their sounds not as individual sounds but as categories. As a result, it should be clear that when we listen to language we listen in categories. The actual nuances and tiny differences made between speakers are necessarily generalized for the necessity of fast processing/comprehension.
2. What are some of the most basic types of speech errors and what do they tell us about the mental lexicon?
As was mentioned briefly last week, the mental lexicon is sensitive to syllable information. We saw this in reference to the phonological rules we were creating to account for variation in language. Now, we have the evidence from speech errors to back up and substantiate our claims about the construction of the mental lexicon. The most important thing such mistakes can tell us is that linguistic environment (syllable structure) is somehow encoded in the lexicon. Such errors also can substantiate the hypothesis that words actually exist in the brain as units.
There is tremendous regularity in the mistakes that native speakers make when speaking. Not only are consonants not ever substituted or switched with vowels, but like elements within the structure of a word are also only subject to such phonological operations. Such mistakes also tell us that the brain both remembers what has come (for a certain period of time) and also plans in advance what will come.
Some of the most common types of speech mistakes are:
Spoonerisms (metathesis)
Here, classically, the first segments of words in an utterance are switched.
the dear old queen = the queer old dean
Metathesis also indicates any kind of switching of segments on any level as in the following examples:
a language learner needs = a language needer learns
Freudian slip = Freudian shrip
cooked a roast = roasted a cook
Additions
Here a segment is or segments are added.
Have a good fight. = Have a good flight.
Omissions
Here, segments are deleted.
rose plants = rose pants
Preservations
Here, some aspect of a preceding segment is preserved.
John praised the man = John praised the pan.
water down = water dawn
Anticipations
Here, some aspect of the segment to come is imposed on an appropriate preceding segment.
walk in the park = pork in the park
see you tomorrow = see to tomorrow
It can be seen from the above examples that the basic patterns of phonology are never broken. This means that the mental lexicon (*and we will try to define what this is a bit later in the course) must have access to a vast array of phonological knowledge. This knowledge would have to include information about not only the segments themselves and the features bundles which make them up, but there must also be information about supra-segmental properties, which implies that information about syllable structure as well as word and clause information just be present as well.
3. What is the scan-copier model and why is it so influential?
The scan-copier model was created to account for speech mistakes like anticipations. Such mistakes show that some amount of preplanning is necessary. The scan-copier model claims that we plan our utterances out slightly in advance of time. In doing so we create two representations of what we want to say; a full version and an empty one. As we go to actually speak, the process of actually doing so, entails moving segments one by one from the full template to the empty one. As we do so, sometimes the process will go wrong and one segment will be moved into the incorrect slot. In the end, a segment is left over which must be placed. It will be placed in the only space that is available.
This is a very simple model, but it is powerful because if we look at it carefully, it substantiates several important observations we have already made about the lexicon via an analysis of speech errors. First, only like segments are replaced. Also, they are always inserted into a similar linguistic environment in the template. The most important thing is the claim that morphological operations come after general segment placement. This final observation raises a number of very interesting questions which we are not yet equipped to answer, so let`s leave it at that
4. What is cohort theory?
The cohort theory was one of the most widely accepted models of language perception and still carries a tremendous amount of influence although it is not without its problems. What the cohort model says is very simple. It states that finding the right word (target lexeme) in our mental lexicon is really just a process of segment elimination progressing from the front of a word to the back. When we hear a word, we first perceive the coda. We start searching through our lexical reserve for words which share the same coda. Words with other codas are not searched for. Once the coda is identified all the words which share that coda will be locked on (primed as it were). As we hear more segments of the word in question we refine our search looking only at those words which match the segments we have heard. This continues until we have heard the entire word or found the word based on the segments we have heard, whichever comes first.
This theory is powerful because it predicts that we can actually access words in our mental lexicon before the speaker has finished uttering them. This seems to fit in with what has been documented as regards speed of perception. It might help to think about it this way. When you teach your students or talk to your children, sometimes they are very slow to get words out and you, as the teacher or parent, might actually finish their words or sentences for them. That is, you are able to access the word in your lexicon before they have heard it. The cohort model explains how this might be possible.
The configuration of the cohort model also brings potential problems. Because the first segment is the key to unlocking the rest of the word, it is very hard in this model to access a word if the first segment is not heard. We know that in real life situations often segments are not heard properly due to outside noise or slurred speech, among a multitude of other reasons. Also, the simple fact that word boundaries are not always clear makes it very hard for a speaker to be able to know which is actually the first segment of a new word. These are major problems for the cohort model. Some simple answers to these problems are to be found in the next couple of questions.
5. How do frequency and recency effects work?
How frequency and recency effects help to access words can be explained together because they are prey to the same system constraints. We should know by now that language is in the brain and as such is constrained by the way the brain works. We also know that the brain is really nothing more than a very sophisticated chemical/electrical appliance. In short, the brain runs on electricity and quite a bit of it actually. A simple metaphor claims that whenever we recognize or access a word with/in our mental lexicon we do so by shooting a certain amount of electricity into the space that word occupies (sic.). The effects of all this movement of electricity are quite important. Studies have shown that the more frequently a word is accessed the more easily it will be accessed in the future. For example the word the is the most frequently used word in the English language. The word the is thus much easier to access than the word off. Frequently used words just seem to pop into our heads without us having thought about them. Recency works the same way. Words that have recently been accessed (fired) retain a certain amount of electricity for a certain period of time. Another term for this is priming. Primed words are much more easy to access than words which have not been primed. We can also think of frequently used or heard words as words which are perpetually primed.
How all this electricity relates to speech recognition is quite simple. Words we have just heard or are frequently used do just pop into our heads, even if we have failed to hear all the segments, including the first one. If you are having a conversation with your friend, for example, about what to have for dinner and your friend wants to eat hippopotamus steak (your friend is strange) and you want to eat fried porcupine rump (you are equally strange) then you will be able to access the words hippopotamus and porcupine as well as some other words easily because they will be repeated several times in the conversation. All conversations are like this. They revolve around a particular theme (which does shift so be careful), which also corresponds to a fairly limited set of appropriate or established vocabulary items. In this way the recency and frequency effects of the brain help us to access words even though we have not heard them completely or properly.
From a practical teaching point of view, this argues for organizing lessons around certain themes with as much exposure to and use of the TL as possible.
6. .How does context work in recognition?
The idea of how context works in speech recognition takes the same idea we saw in frequency and recency effects and extends it. As we will see in greater depth later in this course, words are arranged in a vast array of different networks in the brain, one of the moat salient of which is a semantic network. This means that words which are related in meaning and use are linked together in the brain. So the word doctor is semantically linked to the words nurse, patient, hospital, wound, surgery, pain, evil, rich, expensive, bossy, etc.. These words form a semantic network. Semantic networks are idiosyncratic in that they vary slightly from person to person within a linguistic/cultural group and probably more across linguistic/cultural groups. Now, when one word in a semantic network is fired (is accessed by having electricity pass through it) all the other words in the network become primed. As we discussed above, priming a word makes it easier to access. This is semantic context.
Another type of context is the syntactic context. This often works together with the semantic context, but works in a different way. In syntax, all words are assigned to a certain grammatical category. Some grammatical categories are noun, verb, and preposition to name a few. We can also use this syntax categorical information to help us decide on the intended word if, for some reason, the word is not able to be heard.
Context, both semantic and syntactic, in coordination with frequency and recency effects make it almost impossible for a native speaker to fail to access the right word in a normal conversation and, as such, support the basic idea of cohort model.
7. What is the minimal attachment theory and how does it work?
This is a simple theory which says that a person will always take the simplest (and most familiar) route possible when analyzing language. This is something we will discuss more in relation to syntactic processing later in the course, but it is important to note that the same theory applies to phonological analysis, particularly in non-native speakers. For non-native speakers often the simplest and certainly the most familiar analysis will be one based on the first language. Often, in the initial stages of second language acquisition, learners will not be able to perceive at all of the underlying systems of the TL. Everything is interpreted automatically through the systems of the L1. This is a simple cross-linguistic effect of the minimal attachment theory. A simple example would be in phonology.
Some researchers argue that non-native speakers actually hear the sounds of the L2 differently than a native speaker because they are only able to use their L1 systems to perceive the sounds. This is sometimes called psychological reality. As time goes on and they are shown specifically how the particular sound system of the TL is composed in relation to specific sounds, the learners will begin to create a new system in their head which differs from the L1 system. This generally happens in a series of small steps rather than one big leap. This is an interesting and often overlooked phenomenon in the realm of SLA research.
Further blab and stuff
This chapter is divided into two sections; speech perception and speech production. The former is hardly discussed in the chapter, not because there is nothing to say about but because much of what can be said was already covered in the section on sound change. There are, however, some claims which can reiterated. The first of these is that vowels behave differently (remember the push and drag chains) and are perceived differently then consonants. The boundaries between vowels in a language are much more fuzzy than the more linear boundaries we find for consonants. Remember last week when we were looking at allophony at there didn`t seem to be much to find in the way of vowels. Well, this might very well be due to the way the vowels are perceived in speech and are stored in the brain. The basic idea is that vowels are much harder to differentiate. Consonant sounds, on the other hand, seem to be more distinguishable both from vowels and the categories (such as voicing) which make them up. As mentioned before, none of this is earth-shattering. We encountered much of it before in the form of sound change.
The standard way of studying speech production is by looking at errors adult native speakers make when they speak. A careful analysis of such speech errors (whether Spoonerisms, anticipations, preservations, substitutions, additions, or omissions) reveals much about how words are stored in the brain. The chapter argues that words are stored in syllables with some reference to their role within the syllable, i.e., onset, nucleus, coda. There also seems to be evidence that we plan what we are going to say before it actually comes out. This is claimed because many of the errors found in speech production are dependent on the structure of the intended message. Errors are systematic and constrained by the structure of what we want to say as dictated by the scan-copier-model. As this model operates on the syllable level, the authors also point out that certain aspects of phonological processing occur at the segmental level as well. The example they give for English is rhyme, but they offer not theoretical explanation of this and instead go on to explain how the Roman alphabet (used in English) developed and conveniently forget that Korean also uses and developed its own phonetically-based alphabet even though the letters are arranged in syllable configurations.
(If you want a copy of this old but interesting and still valid article (Cole 1981) please come to my office. I have it on reserve there.)
Cole in his paper on speech perception has quite a few interesting things to say. He makes three claims about how words are recognized. The first of these says that words are recognized through an interaction of their sound as related to the contextual situation. His second claim is that words are recognized sequentially one after the other. This is done in connection to both syntactic and semantic constraints. His last claim is that phonetic information is used to help recognize words. Personally, I don`t see a need for this as it is already implied in the first claim. While these observations seem to be true it does not mean that they are true in every situation.
The results of Cole`s experiments show that children are able to perceive the differences in initial sounds (onset) much more than final (nucleus) or medial sounds (coda). This matches, at least in part, what was reported in Aitchison (1987). He also says that some sounds (stops) are more easily differentiated than others by children. This is supported by what we read earlier in Macken and Ferguson (1981) about the order that phonemes are acquired in children. The net result of all this is language perception of adults as well as children are based on the phonological system of the speaker. This much has been made clear by the data.
CogLing, Unit 4: Radial Categories, pp. 53-69
8. What is the nature of radial categories and how do they differ from other category types?
The original idea of categories has been around for more than 2000 years. The ancient Greeks believed that categorical membership was based on a set of specific features which all members of the category had to have in common. Therefore, membership in the categories such as bird required certain features, such as feathers, the ability to fly, stick-like feet, and certainly a beak. In order to be a bird, in this model, an entity would need to have all of these different features. Of course the category of bird also entailed other features which were maybe less central or less important or essential to being denoted as a bird. Generative linguistics still basically uses this Platonic view of categories and how it describes language. The radial categories of cognitive approaches differ dramatically in that, while there is a central core to each concept, concepts tend to be extended out and directions from the central core in many different ways. In radial categories, therefore, there are no particularly essential elements to the category. If we look at a simple example began related to the category of bird, birds are used in the central core to denote a certain type of animal would share certain features. In language is however we extend this category of bird to people on a fairly regular basis and in a fairly regularized way but in quite different senses. In Great Britain for example, an attractive young woman is often referred to as a bird obviously relying on some sort of association between some aspect of birdiness and attractive young women. A person can also from referred to as bird depending on how little they eat. Thus, we can say 'she eats like a bird'. Based on some aspect of birds which we hold culturally be true, namely the belief that birds don't eat very much, which is actually not true at all and certainly not true for all birds. We can therefore see how radial categories extended out from central core of which is generally a prototype, and should be clear to us by now that prototypes occur as a result of frequency and to certain extend recency, but the two of those are closely linked anyway in the real world. It is however relatively unpredictable exactly how we will extend things out from the central core. All elements/units of language are in effect radial categories. Even sounds are arranged in radial categories, albeit simpler ones than words, and some words have much more complicated radial sets than others.
In short, radial categories extend out from a central core in multitude of different directions which indicate different senses associated with the concept. A language speaker to pick up on any of these different senses and extended further away from the core thus creating a new use for language.
9. How do radial categories work in affecting language use?
A radial category is also a frame and in some more complicated instances a single radial category will be composed of many different frames. We're ready aware that frames are basically another word for schemas and their basically encoded relationships between certain concepts. When a speaker goes to use language they basically open up the category for a certain concept is an also entails the opening up of frames. As mentioned above some concepts contained several different frames each of which is quite complicated in its structure. As we open up more concepts we select certain frames and proverbially lay them out. As we do this we then begin to form connections between certain elements in certain frames. This is called mapping and basically entails forming associations between elements in frames in order to create an utterance. In this process a speaker is creating meaning based on the connections to clean the elements which are mapped together. This is a creative process and is often intertwined with other certain cognitive abilities like metaphorical extension. We can therefore take a certain aspect from a radial category and extend it further away from the core in the process of creating language. We do this in connection with other senses of other categories which are mapped together. In this way novel utterances and new senses are created and if they are accepted and further use they will become regularized. That is, they will actually become encoded in the mental lexicon as elements associated together in physical sense. The become psychologically real.
It is however, important to remember that radial categories and mapping work together with the idea of conventionality. There are essentially two forces at work here which help regulate the creation and comprehension of language in the cognitive model. The first of these is the idea of systematized creativity. Because cognitive models link language and cognition in profound ways to have to be certain underlying similarities and how all people create language because we'll think implementing similar processes. This obvious from the start to think about comprehension. If I say to someone, "she eats like a bird." then the person I'm speaking to has to somehow be able to decode my message. They need to be able to understand my mapping based on my different radial category sets and what I have linked together. If this does not occur then language will break down and my mapping will simply have not worked. Undoubtedly this does occur within native speakers of the same language but for the most part even though we are constantly creating new and novel forms of language other speakers understand us and the point we're making. This would mean that there has to be some set of underlying system which we have maybe not uncommon but in a similar way. The system has to be similar some people can actually understand what other people say. At the same time there's another opposing force, that of conventionality. People can more easily understand what others say simply because they've heard it before. Certain constructions that are frequently used are encoded in the lexicon. They have become accepted conventions through use in this greatly aids in comprehension particularly the forms are based on extensions away from central core concepts.
10. Why is the concept of radial categories so important for cognitive linguistics?
Radial categories provide the very basis of linguistic forms. They are important because they not only allow us to extend concepts in new and different directions easily but they also provide a way for conventionalizing such extensions through the idea of associating frames. Without the specific types of categories language would not be as creative and actually appears to be. Thus, radial categories are the very basis of our creative ability in language.