Stephen van Vlack
Sookmyung Women`s University
Graduate School of TESOL
Introduction to Linguistics
Fall 2006
Week 9 - Questions for discussion
Intro. to Ling. pp. 226-276.
1. Briefly describe the serial-autonomous processing model.
The serial-autonomous processing model basically works in accordance with its apt name. It claims that language processing is only able to occur in a specific sequence, that is serially. Essentially, the claim is that there is a strict order in language processing. Each step is autonomous. None of these steps affect each other directly. One autonomous step in the process necessarily leads into another and in a specific direction. There is no backwash in this model. In order for precessing to occur in the serial-autonomous model, the entire procedure is divided into several special autonomous units or modules. Each module has its own specific job to do and that is all it does or can do. In effect each module functions differently (at a different level/with different sized units). The module does it job and this allows the collective information resultant from that one module to bleed down to the next step.
This is a strictly linear process in that each module needs to receive information from the previous module to do its job. If the correct information is received, it does its job and passes its own information on to the next module. Each module deals with only a small part of the whole information and is able to make a minimal contribution to the entire sequence of processing. In this way, the serial-autonomous model of language processing is seen as being phonologically driven.
The term phonologically driven is used because the phonemic breakdown, from front to back (remember the cohort model), of the word in question is the first necessary step in this model. Also, if for some reason the listener does not manage to catch any part of the phonemic make up of the word, then regular processing will stop and a back up extra-linguistic system will take over. The total word will be lost to the listener. Regular processing in the serial-autonomous processing model requires that all the phonemic information be available. If all the sounds are not available then the listener will have to use other mechanisms to guess what the word might be. Bear in mind the same model has had a profound effect on views of reading as well using graphemes in place of phonemes.
2. Briefly describe the parallel-interface processing model.
The parallel-interface model of language processing claims that language precessing is a necessarily integrative process. All possible or necessary parts/bit of information are interrelated and can be accessed at any time and with any amount of information. There is no proper order or strict directionality in language processing in this model. Processing is not strictly linear nor modular.
When exposed to a particular sequence of language data, then, the parallel-interface model claims that the listener is able to access all available information at the same time. Based on this, we can see that the parallel-interface model is context driven. This does not mean that the phonological composition of the word/lexeme is not important. Obviously it is, but the listener is able to use other means/information as well to figure out (process) what they are hearing. For example, last semester we had an exciting presentation on the McGurk effect which is often used to support this type of processing model in that visual information (particularly in relation to facial expressions and related vocal movements) seems to play a very important role in language processing
From a strictly theoretical point of view this model is less interesting because it very general. There are no real constraints imposed on the process, so the model really says little about how words are processed. It also introduces the potential problem of reliability. Since there are no real constraints we might suppose that people would make more mistakes than they do if people actually processed language according to the generalized processes suggested by this model. For these reasons, this model has been less successful in academic circles despite its seemingly obvious general correctness.
3. What does it mean when I say that a particular lexeme is primed. How does priming work?
The idea of priming is based on the way in which the brain works. Most of the cells in the brain are glial cells. The main purpose of these glial cells is to connect things in the brain. They function like glue to attach all the different parts of the brain. The main activity of the brain is created and regulated by neurons. Neurons compose only about one tenth of the total cells in the brain (This information comes from Carter. 1998. Mapping the Mind. University of California Press. This book provides an informative and rather easy general introduction to the physical as well as functional composition of the human brain.). Even thought they are comparatively few (10 billion), neurons are extremely efficient, each one being connected with up to ten thousand other neurons. Thinking about the numbers of neurons in the brain we soon realize that the number of connections in the brain is like grains of sand on the beach. Following this, we see that the brain is all about connections. Single neurons, in and of themselves, are rather useless. Neurons only achieve power in their combination with other neurons. We saw something like this in our study of how word meanings are stored in the brain. It might be more helpful to imagine the brain more like a liquid (chemical) entity as opposed to a solid entity. It is also from this perspective that we need to understand what priming means. Think of the brain as being like the ocean where all cells are interconnected through chemical bonds but in various degrees and strengths. Through these fluid connections the brain can be seen as a tightly bound, yet somewhat free organ.
When you send an electrical impulse to a certain part of the brain it is effecting much more than one neuron. This is due to the fact that neurons are so intertwined that it is only through the connections that any neuron can be understood (sounds familiar - lexemes) or used. So, if you fire (activate) one neuron, you are also affecting the neurons that it is connected to in various ways depending on the strengths of the connection as well as the affected neuron`s own activation level. It works exactly like dropping a pebble in a tranquil lake. You drop the pebble and it makes a splash and the splash sends waves out affecting an ever increasing circle of cells, but to an ever diminishing degree. In effect, all molecule in the lake will be affected but to highly variant degrees.
In reality, it makes little sense to talk about single neurons (I am doing so to make it initially easier for you to understand this). In reality all neurons are part of what are called neural networks. A neural network is a group of neurons tightly wired together (glial cells) which work together for a single purpose. The tricky bit is that the brain is multidimensional in that a single neuron might be part of several/many different neural networks and the different neurons in a single network will all simultaneously be part of several different networks. FUN!
Is important to remember here will be talked about class namely that physical proximity in the brain is not necessarily important and priming as a very important part of how the brain works effectively is not subject to proximity. The priming does not necessarily occur with things are physically adjacent to the object which is being fired. Priming occurs across chains of connection. We know that there are billions of different connective links in the brain. The strength of connection to the type the connections will determine exactly how the priming actually works and of course this works in relation to threshold levels as we will see below.
We already saw how there seems to be different levels of integration or connectedness between lexemes in the mental lexicon. Based on that we can see that priming will effect different lexical entries differently depending on the strength of their connections to the fired lexeme and their threshold level.
The threshold level is the amount of electricity it will take to fire a certain lexical entry. Some lexical entries have a very low threshold level, which means it take s a very small amount of electricity to fire them. The threshold level of a lexical entry is reduced based on the frequency of the word`s use and or the frequency of its priming. In this way it is easy to envision the connections in the brain as being like muscle in the body. Once a muscle is built it is there forever, but its size and efficiency will increase and decrease with use or disuse. In the same way, lexical entries (words for arguments sake), once they are there and have been connected to something else are always there, but the connections will whither away to almost nothing if the word is not accessed or primed. The more connections that are built between lexical entries and indeed concepts outside of the mental lexicon, the easier it will be to remember and use a word.
It is for this simple reason that direct translation does not work as an effective means of second or foreign language production . When we learn words as translation equivalents, then the new L2 lexical item will only be connected to its L1 translation equivalent and nothing else. Of course it will very rarely be fired or primed, so it will disappear quickly. Once more, when this lexeme is fired the only kind of information that we will be able to get about it, both the form and the lemma, will be information from the translation equivalent in the L-1.
Now, getting back on track, priming effects have been used to study a wide variety of phenomena in the field of psycholinguistics. Priming is interesting because we can use it to study not only how closely and generally how words are related, but also contextual effects. In the SA model, for example because context does not come into play until the very end of the whole process we would expect words that have more than one meaning (and most do) to prime all the possible meanings, at least initially, no matter what the context is. This would not be true for the PI model. Since contextual information is available at any time, then we would expect the number of possible meanings to be limited by the context and these meanings would, thus, not be primed. Sounds simple. The evidence, however, goes both ways and both models are supported at different times, meaning that both come into play. One of the reasons for this, and this is something they do not mention in the chapter, is that not all the meanings a lexical entry carries are equal. Some are more central than others. The example that they use in the book, BANK, is a good example of this. The meaning, financial institution, is much more central than side of a river. The different strengths of the meanings will have a strong effect on priming.
4. How are the lemma and form connected? Do they have to be separate?
The lemma and form are connected via lexical pointers. Exactly what these lexical pointers are is not at all clear. We know that the form part carries phonological information about the lexeme (both segmental and suprasegmental info). The lemma carries semantic information. This is not only information about the meaning, which we know is done through connections to other lexical entries, but also syntactic information. For example, for nouns it will include possible theta-roles and well as idiosyncratic information like if the noun is a non-count noun. For verbs the amount of information carried in the lemma is staggering. All syntactic information (that is not UG) is housed in the lemma of verb entries. We also know that lexical entries, the lexeme, is connected to conceptual information (in the for of imagens and logogens) which is housed outside the mental lexicon. This does not leave much room or need for lexical pointers. Based on this I envision lexical pointers as being nothing more than a bridge which connects the two. The bridge does not contain any content or real form. It is just a bridge which works, but can be blocked and stopped (glial cells).
It is quite easy to see that the lemma and the form are indeed separate, that they have to be. In most cases it remains true, however, that information is easy exchanged between them. It is still possible to use one but not the other, which proves their separateness. Look at the example below.
A. Chris was a good friend but suffered from acute logorrhea, which made it impossible for him to ever stop talking. Sometimes we just had to yell at him to shut up to get him to stop talking even for a minute.
In A, you can probably figure out that logorrhea means to speak excessively or compulsively, but you probably do not know how to pronounce it /logɘriɘ/. This shows that you can access and for a lemma without accessing the form. Likewise, think of the tip of the tongue state. When you are experiencing this, you have access to the meaning (lemma), but can`t find the right form. On the other hand, you can access the form, but not the meaning. Look at the words below.
B. shlime
spling
zinl
farthling
You can phonologically process these words and in the process will quickly figure out that shlime and zinl are not permissible in English while spling and farthling are, but you have no access the what these words might mean (the lemma).
This is probably as good a space as any to introduce the idea we talked about which is the idea of lexical convergence zones. The concept of lexical convergence zones is based on the idea of convergence zones (located in hippocampus region of the frontal lobe) as the basis of memory. This does not seem to be such a strange stretch when we think about the idea that the mental lexicon really is just a specific type of long-term memory which is is related to and composed of linguistic data as opposed to extra-linguistic data. The basic idea states that in the brain (and it probably argued that there might need to be two different convergence zones, one for reception (Broca's area) and one for production (Wernicke's area) there would be convergence zones which control the lexical information which is accessed. Basically these convergence zones would act as controllers to the lexical information. We need these kind of convergence zones because it is clear from our discussion about semantic meaning and lexical connections that each lexeme is connected to other lexeme's in any (possibly hundreds) of different ways, both linguistic and extra linguistic. It would seem there for necessary to we have sums of mechanism which can regulate and take control of this rather tricky endeavor.
5. Is there a lexical entry for every concept?
No! Concepts and lexical entries exist on different planes: that, is they are housed in different parts of the brain and work in different ways. Concepts are part of general cognition. They are cultural and are linked with the memory and experience. The words that are located in a conceptual field are not necessarily related semantically. Take the example of the word blend lesson and lecture, lection. These two words lesson and lecture and not synonyms, nor are they co-hyponyms. They simply exist in the same conceptual field.
This does not mean that there is no overlap between concepts and the arrangement of semantic information in the mental lexicon. There is overlap, but the overlap is not 100 percent. The semantic information in the lemma is arranged based on the relationship of entailment. Words also acquire part of their meaning through interaction with concepts, but it important to note and remember that concepts lie outside the realm of the mental lexicon. They are part of e-language.
6. Why do you think the Canadian children in the chart on page 259 had more trouble assimilating phonologically to their new dialectical surroundings?
The experiment on the Canadian children supports something that people talk about very often, which is that semantic change occurs in a fundamentally different way than structural change. The perceived stagnation (fixedness) of structural elements of language (phonology, morphology and syntax) after a certain point has given rise to the much debated Critical Period Hypothesis and its more modern version the Sensitive Period Hypothesis. Forgetting about that, let`s just look at the nature of semantics as opposed to purely structural elements. In a totally monolingual situation it has been claimed that the structural elements of language are basically finished developing by the age of 7. People, however, continue to expand and constantly rearrange their semantic networks throughout their life. Even when readjusting to a new dialect, this obviously taxes the structural system and it is not easy to change. We said earlier in this course that language is an encapsulation of the world. This occurs principally thorough the semantic system. As we encounter more things in the world or as the world changes the semantic system might change as well but it doesn`t have to. According to Clutter Theory, it requires more effort to make these changes as a person grows older and has more experience with the world and thus has more sophisticated semantic systems. The more we have the harder it is to incorporate more ideas and the change concepts.
7. What do borrowings and linguistic change tell us about the structure and working of the mental lexicon?
As mentioned above, structural and semantic elements seems to function in different ways, the main difference being that semantic change occurs much more easily and longer than structural change. One reason for this might be the way that they arrange their connections and their scope. Structural elements all have very limited scope. For example, phonology is only about sounds. These sounds form a very tight net which is highly restricted by the principles of UG and the parameter settings for the specific language. The same can be mentioned for morphology and syntax, but on different levels. We have seen that changes in these structural areas, probably because of the tightness of their systems, will result in a whole series of changes. (A very poignant example of this is the push and pull chains we saw in phonology). Semantics, due to its size, tremendous scope, and relative freedom from UG, doers not seem to follow the same restrictions that we see on other components. Dropping a word or a creating a whole slew of new words will not cause the entire system to alter itself. Changing word meanings is a normal thing, but changing syntactic systems is an extremely taxing, time consuming process. Structural changes may take hundreds of years to achieve.
One way of dealing with this is to think of the structural components of language as being essentially hard-wired while components like semantics and pragmatics are more soft wired. Way back in the beginning of this course we said that UG is, at least partly, a set of empty land lots along a series of preset roadways that are just waiting to be filled, well that is true and once the empty spots along the road are filled they can`t be undone except by some catastrophic event because the moving or killing of one lot profoundly effects the whole road.
Further notes
Section 14: Lexical processing and the mental lexicon
This week we are looking, for the most part, at two of the main models of processing in addition to trying to figure out the vicious conundrum of how exactly words are stored in the mental lexicon. The two processing models we are interested in are the serial-autonomous and the parallel-interactive models. The two are more similar than they appear in the book and differ themselves primarily in the area of ordering.
In the serial-autonomous (SA) model, processing occurs in a one-way strictly linear series of basically bottom-up operations. These operations occur in self-contained modules which do not affect each other. The exact order of the operations is not that important for us here, but it seems to make sense to say that word recognition begins with the sounds or sighs of the word (its phonetic or orthographic composition). The first of these operations must be finding the word in our mental lexicon according to its phonological or orthographic shape. Once we have done this, then we can move onto a morphological analysis of the lexeme. This analysis will only tell us about the morphological possibilities of the lexeme. For example, the lexeme PLANT can be a verb or a noun. Usually the only way to tell if such a lexeme is being used as noun or a verb is to get access to syntactic or semantic, as well as contextual information. In the SA model, however, none of this information is available at the same tine as the morphological information. We have to access everything separately. Also, once we have moved through one module in the process and gone on to another it is impossible to go back. So, for example, if you are not able to find a phonetic representation for a concept that you hear, you will not be able to process that sequence of sounds at all. Certainly, it is possible to understand what the speaker is saying, but you do so in a way not connected to normal processing.
This is not the case in the parallel-interactive (PI) processing model. In the PI model, while we go through the same operations as in the SA model, there is no set order. The different operations can occur simultaneously if they need to. So for example, you are talking to a friend out in the street and traffic noises and the like stop you from hearing a word or short sequence of words very well, then you will still be able to ultimately find the lexeme you are searching for by using morphological, syntactic, semantic, and/or contextual information.
Before we go on to analyse support for each on the models, let me just mention that there seems to be a few weakness in the SA model. The main one of these is that native speakers are able to manage to process information. This is similar to Chomsky`s `poverty of the stimulus` argument, which he uses to support his claim that children have some sort of LAD which helps them acquire the language around them. It might be a stretch, but maybe acquiring at one stage of development is nothing more than processing at another. In affect, we are all forever language learners in our L1 as well as all the other languages we have. Part of the role of processing might just be a learning role as well (this is expressed very well indeed in usage-based models of language and language acquisition [see M. Tomasello]) . The basic idea is this, as well process we learn. By being able to process new, unfamiliar lexemes, syntactic or morphological constructions etc., we are acquiring those forms hopefully making them available for later use. This is the time in between processing and learning and I feel we can`t separate them. By thinking in this way, only one of the discussed processing models in valid.
The problem with the SA model is that it has no way to account for language learning within processing. Learning and accounting for incomplete, impoverished input are one in the same system and I believe, happen so often that it has to be part of our normal processing. It is not an uncommon thing to be presented with a linguistic sample that has holes or gaps in it. In fact, it is the norm. Rarely, do we get perfectly uttered words and sentences that are nice and easy to process by moving through the phonetic shape first (bear in mind that we all have different surface representations of the supposedly same underlying phonemes). What happens when you travel to another region of the country where people speak a dialect that you are not very familiar with. Are you not able to understand them at all? No of course not. You might not be able to understand each word immediately, but you can understand what they mean and you are probably able to go back and fill in the meaning of different words a little not later. In the SA account this might be possible but only through inferencing. In short this would have to occur apart from normal processing. My argument is that this kind of understanding is a fundamental part of normal processing.
In the same vein, when we learn new lexemes do we do so inductively or deductively? Well usually, we get the word in context and figure out the meaning from there, this also supports the PI model. Since we are constantly learning new words and new bits of language this too has to be part of our processing. Our goal in doing linguistics is to try to find the simplest systems to account for what we observe in language and since the PI is simpler in that it can account for a wide variety of operations, it would seem to be the better model. This does not , unfortunately, mean that there is no support at all for the SA model.
Central to the idea of how the brain works is the idea of priming of firing. The brain works on electricity. Since lexemes in the brain are linked in many different ways (sound, meaning, syntactic groupings, etc.) we can assume that when we fire a lexeme, i.e,; when we trigger a word by shooting an electrical impulse into it, we are also priming the lexemes that are attached to it. If a lexeme has been primed, it is assumed that it is easier to access because it has recently had electricity hit it. By looking carefully at priming relations, we can try to determine which processing model seems to be working.
Lexemes, as was already stated, are connected in a systematic way to other lexemes that share features. These connections occur at a variety of levels. Many of these connections are semantic in nature. Some of these semantic connections involve dual meanings. The word Flasche in German has two meanings. The first is `bottle`, the other is `dorky, idiotic or foolish person`. Look at the sentence below;
(1) Hohl mir diese Flasche da. `Bring me that bottle/female dork over there.`
In this sentence, it is not clear which meaning (bottle or female dork) is intended and we would expect in both models both meanings to be primed. In the following sentence, however, the context makes the meaning clear.
(2) Hohl mir den Flasche da. `Bring me the bottle/male dork over there.`
The noun marking shows that the den Flasche in this sentence is a male entity, while the noun die Flasche `bottle`is feminine. We therefore know from the semantic information that the word Flasche can only mean `dorky idiot`. Gender in German is marked in the noun marker as well as endings sometimes. Die/diese is feminine while den is a masculine, accusative noun marker.
In sentence 2, the context tells us what the noun Flasche means. In the PI model, since we have access to this information, the bottle meaning of Flasche need not be fired. In the SA model, since we do not have access to syntactic information from the start, then both meanings will have to be fired equally. Researchers have found that people do seem to access both words in the case of lexemes that have two or more distinct meanings. This means, they argue, that there does not always seem to be access to other information at the same time.
There is just as much support for the PI model, however. We do not want to argue that two different systems are working. Remember, we have to assume that the system is as simple as possible. I think there is a way to account for these differences. Assume that in the PI model that processing prototypically tends to start with the phonetic shape of the word (based on salience) and only if there are problems encountered will it then shift into parallel processing mode. This might be a better way. There is also the severe problem of trying to argue behavior and processes conducted in an experimental situation is indicative of normal behavioral processing behavior. It might be very well possible that behaviors vary according to the situations.
This would also account for the what we assume to be the standard representation of lexemes in the brain. The main idea here is that phonetic/phonological information is kept separate from all other information. The form and lemma need to be distinguished because tests have shown that we are able to make decisions about a word appropriateness apart from meaning, syntax, or morphology. If we posit a processing model like the PI model, but one where phonological recognition is the first step to be followed by an integrated pattern of processing information, then this works well.
Section 16: Lexical variation and change
In this section, the book takes a quick look on lexical change and variation. Without a doubt, the most important way that English has changed has been through borrowing. But looking at this only gives us a part of the picture. Semantics is, as a system, similar to all the other areas we have looked at so far; phonetics, phonology, and morphology. It is a system with a certain number of variable configured in a certain way. This means that if something changes then this will force other things to change as well. (Remember the push and pull chains we saw in phonological change.) Things in semantics are basically the same. As new words come in through borrowing, as they have in English and Korean, then the entire system can change.
We saw in our brief look at semantics that words do not exist in a vacuum, they are linked to other words in a variety of complicated ways. Words are defined not only by their root meaning, but also by their relatives; what they are and what they are not. When many new words enter a language then the system is unstable and thins are changed so that the systematicity can be preserved.
It does not take a genius to figure out that English does not really seem much like a Germanic language. Well, at least it seems different from other Germanic languages. The grammar is quite Germanic, but the vocabulary of English is quite the opposite. Most of the words used in English today are of foreign, mostly Latinate/Greek origin just as most words in Korean (70% is a very conservative estimate) are of Chinese origin. The entrance of these words into the language caused huge upheaval of the already existing words. The large amount of broadening and narrowing discussed in the book occurred in large part due to the great influx of foreign words.
What you want to get out of this section, then is that semantic change and variation is systematic. Change does not occur just for change. There is always a catalyst, but often the cause of the change is quite distant from the effect.
CogLing, Unit 7: Language Change, pp. 137-146.
8. Is the distinction between count and mass nouns motivated or arbitrary?
The main driving force behind this chapter is to clearly show that the distinction traditionally made between mas and count nouns is not arbitrary as has generally been argued by structural/generative linguists but is actually motivated. The term `motivated` means that there is a reason for each of these distinctions. Since Saussure in the early 1900s, the forms that are used in language have been claimed to be arbitrary (with the possible exception of exciting onomatopoetic words like snarl, gnarl, smash, moo, miaou, crunch. This is the fundamental reasoning behind what became structural linguists, which championed the idea that language can only be studied as a scientific phenomenon if the relationship between linguistic forms on all levels and the world at large is arbitrary. This central belief in structural linguistics has since been developed into the modularity hypothesis, which claims that language is a separate module in the brain which is only minimally affected by other brain functions. According to the Modularity hypothesis the forms and the information related to language is unique and informationally encapsulated. All idiosyncratic information and/or behavior in language (in the generative model this basically means all information not determined or strictly controlled by UG) is going to have to housed in some part of the linguistic module - the mental lexicon. Thus, in the generative view the lexicon is the storehouse of all idiosyncratic linguistic information (forms and meanings) and grammatical behavior (that which does not adhere to principle or parameters of UG, or language-specific productive processes, such as /-z/ plural formation in English). It is also claimed that all this idiosyncratic information and behavior is arbitrary. Thus, there is no reason why a certain animal is referred to as porcupine in English and hoho`i in Papago or muungyawu in Hopi. Likewise, there is no reason, so the generativists claim, for why certain verbs are transitive (require an object) while others are ditransitive (require two objects), or why some even seem to require objects of a certain category type. For example verbs of cognition, such as think and wonder, seem to require objects (complements) that are entire clauses.
(3) John thought [that he would never escape the drudgery of his linguistics class].
(4) Bill thought [John was a total jerk].
(5) Mary wondered [if she really liked her pet porcupine or not].
(6) Bob wondered [whether if he would ever really make the grade].
The position in the generative model is that the fact that these take special kinds of complements has no reason. It is arbitrary and idiosyncratic. This is supported by the theory Thea not only is language a separate module from other components of the brain but the different parts of the language module are also modules unto themselves as well. Thus there is no supposed effect from semantics on syntax or morphology or any other component. Thus, in the generative module syntactic phenomenon are unaffected by other components. The syntactic phenomenon of mass and count nouns is no exception to this general argument. Count nouns and mass nouns are seen as being a purely syntactic phenomenon which is not affected by meaning. The distinction is arbitrary, and idiosyncratic and thus must be stipulated in a noun`s lexical entry. Well, this is how the generativists see it.
Cognitivists like Lee see this in a very different way and they argue that there are no clear modules to language. All linguistic behavior is motivated in some way shape of form from the cognitive system upon which language is based. Looking carefully we can see that this does indeed seem true for count and mass nouns but maybe not exactly in the way that Lee describes in this chapter. He posits that it is a matter of categorization. That there is a category for mass and a category for count based on certain cognitive properties related to real world features of the nouns. The basis of this is the concept of individuation. If an object is able top be broken down into discernable component parts and also has identifiable boundaries. In addition, to this we can ad a specific shape which cannot be altered. Entities that neatly fit into this category will be count. Those that do not will be mass. This is a nice simple idea but the fact remains that really all words appear to be able to be used in both mass and count shapes as determined by their meaning or what aspect of their own radial categories are being focused on. Really I don`t think that words are categorized as being count or mass. All words are potentially both count and mass as determined by the context and meaning which is culled out of their radial categories. Thus, the same basic theory that Lee is proposing is indeed true in that it does seem that the grammatical phenomenon of count/mass is indeed motivated by meaning and a cognitive system but it may work in a different way than he proposes.
9. How can this count/mass distinction be applied to grammatical systems in general.
The basic idea is that if the grammatical distinction between count and mas nouns is clearly motivated by properties of the words themselves then it is certainly possible that there are many other motivated systems in grammar. In fact the cognitive claim is that all linguistic phenomenon is indeed motivated. It is this idea of cognitively motivated linguistic phenomenon that is the main core of the cognitive model to language. Here after slugging through lots of rather obtuse explanations of concepts in the model we finally get a chance to see how it really works. We can see how a speaker might need to make choice between whether to use to word `pork` as a count noun or mass noun in real speech situations and the decision will basically be determined by the situation. It is, however, the cognitive system which makes all this possible.