According to Dirk Geeraerts, linguistic theories often tend to be fairly insular affairs, whereas Cognitive Linguistics seems to form an archipelago instead of an island. Rather than a demarcated area, it is an agglomeration of closely connected linguistic research that share the same view, but have not yet been gathered under one precise theory. When compared to other hypotheses Cognitive Linguistics is an open, flexile theory rather than a confined one. In addition, Cognitive Linguistics demonstrates its own concepts from the viewpoint of category structure. It exemplifies the fact that when defining a category one may want to convey the image of its principal members rather than provide an abstract definition, but it also stresses "that the abstract deï¬nition need not consist of a single set of defining characteristics that belong uniquely and distinctively to that category."( Geeraerts 2006: 2) As aforementioned Cognitive Linguistics is a rather flexible framework, it is not a single precise theory, but rather an array of overlapping approaches and although it has not yet been combined into a rigid theory there are many common features, shared perspectives, and forms of research that have been labeled as Cognitive Linguistics. There is, however, a difference between Cognitive Linguistics and the uncapitalized cognitive linguistics, which refers to approaches studying natural language as a mental phenomenon. Cognitive Lingustics is just one of many forms of cognitive linguistics, and as generative grammar they are both from of linguistic research within the field of cognitive science. A question arises, what makes Cognitive Linguistics distinguished from other cognitive sciences? One could say that there is one basic foundational principle and four tenets referring to the aforementioned. The key point states that language is all about meaning. As stated in the first issue of the Cognitive Linguistics journal (Geeraerts 2006: 3), the Cognitive Linguistics approach perceives language "as an instrument for organizing, processing, and conveying information." Given this fact, one may conclude that language is primarily semantic. Unlike generative grammar where the knowledge of the language is the primary focus, Cognitive Linguistics sees language itself as a form of knowledge, which should be analyzed with a focus on meaning. As mentioned before, besides having one foundational point Cognitive Linguistics has also got four specific characteristics that describe how this approach deals with meaning.
Linguistic meaning is perspectival
Geeraerts (2006: 4) describes meaning not as way to objectively depict the outside world, but as a way to construe that world, to incorporate a perspective into the world. In order to understand this characteristic, one needs to think of spatial perspectives that are present in linguistic expressions, where an objective situation can be construed linguistically in different ways. As an example let us think of a situation where a person is standing in an office,by a door. One could say that he or she is standing in front of the door or behind the door. Although these two statements seem to be contradictory, they both embody different perspectives. Moreover, as the language is full of such perspectivizations Cognitive Linguistics tries to examine them.
Linguistic meaning is dynamic and flexible
"Meanings change, and there is a good reason for that: meaning has to do with shaping our world, but we have to deal with a changing world."( Geeraerts 2006: 4) As one experiences the changes in an environment he or she has to accustom their semantic categories to the given transformations and circumstances. Given this fact, language cannot be perceived as an adamant system, but rather as an flexible one. As an example let us think of insects. A set of strict characteristics that would define all and only insects does not exist, however, there is a "flexible family resemblance structure that is able to deal with marginal cases." (Geeraerts 2006: 4)
Linguistic meaning is encyclopaedic and non-autonomous
When constructing a meaning with the use of the language one needs to take into account the fact that it is not a detached and autonomous part of the brain. Given this fact, one can assume that linguistic meaning is not separated from other forms of knowledge of the world. It can be described as encyclopaedic for it includes knowledge of the world and non-autonomous as it is incorporated with human cognitive capabilities. As beings equipped with a body our experience, which is reproduced in the language we use, is affected by our organic nature. Furthermore, human beings have a cultural and social identity, which may be uncovered by the language, as languages can contain the experiences of different groups of people. Let us think of insects once again. As aforementioned, language is suppose to be encyclopaedic, meaning that people will not only think of the general definition of an insect, but also about some specific species. Knowing this, one needs to assume that the experiences regarding insects in one culture will be different from those in another, which will have an impact on the knowledge that is affiliated with the category 'insects'.
Linguistic meaning is based on usage and experience
Linguistics meaning is said to be non-autonomously integrated with the rest of the experience one has, meaning that it is rooted in experience. One has to note that there are a lot of abstract structures in the language, for instance the pattern Subject - Verb - Direct Object - Indirect Object like in the sentence 'John sent David a message'. Such structures cannot always be directly observed. Knowing this, a question arises, how is the lexicon related to the syntax? Assuming that grammatical patterns as SVOO have an experiential basis in concrete, detectable strings of words, one has to notice that those strings of words do not exist in the abstract, but are a part of real conversations and utterances. "The experience of language is an experience of actual language use, not of words like you would ï¬nd them in a dictionary or sentence patterns like you would ï¬nd them in a grammar." (Geeraerts 2006: 6) Cognitive Linguistics as a usage-based approach takes into account both the nature of grammar as well as the experience of language, which together are the experience of authentic language use. Compared to other theories of the twentieth century such as generative grammar where the langue was more important than the parole, this seems to be a rather innovative approach.
In conclusion, Cognitive Linguistics is not a single precise language theory, but rather a group of approaches that overlie each other. Furthermore, the key point of the aforesaid approaches is that the most important part of language is meaning not abstract structures as exemplified by generative grammar.
Corpus Linguistics
Corpus linguistics originates back to late 19th century when lexicographers and dictionary makers would gather samples of language in order to explain the meaning of words. The new era of corpus linguistics began with the creation and evolution of computers, which led to the creation of modern-day corpora. The Brown corpus, the first ever corpus based on computer technology, incorporated approximately 1 million words. The corpora that are currently in use comprise of hundreds of millions of words. A question arises, what is corpus linguistics and what is a corpus? Bennett (2010) defines corpus linguistics as an approach to the study of language in use by means of corpora analysis, and corpus as a collection of texts containing samples of natural language, stored in electronic databases. As there are many types of research, approaches to corpora analysis as well as different types of corpora had to be designed.
The Corpus Approach
Defining whether corpus linguistics is a methodology or a theory seems rather difficult, however, when using corpora in order to study language methods as 'The Corpus Approach' (Biber, Conrad, Reppen: 1998), which consists of four main features, can be utilized. Firstly, this method is empirical for it analyses authentic language present in modern-day corpora, which gather samples from sources such as literature, newspapers, and TV shows. Secondly, it uses a large collection of texts for analysis, regardless of whether it is a written or spoken corpus. Thirdly, computers are used broadly not only to store corpora, but also to analyse them with appropriate programs. Without computers effective analysis would not be possible. Finally, 'The Corpus Approach' relies on quantitative and qualitative analytical methods. As said by Bennett (2010: 8), "we take the quantitative result generated from the corpus and then analyze them qualitatively to find significance." The use of such methods answers the questions concerned with language and its use.
Types of corpora
Knowing that, a corpus is large collection of naturally occurring texts stored in electronic databases a question arises, what are the types of corpora and what were the designed for? According to Kennedy (1998), depending on the purpose of research, several types of corpora are available for analysis ranging from general to specific, from written to spoken and bilingual and multilingual.
General corpora
General corpora, such as the British National Corpus or the American National corpus, usually include a very large database of words, containing samples from various types of language. The research for which they have been created is undetermined that is why it may range from the study vocabulary to the study grammar. Furthermore, generalized corpora are said to be balanced as they incorporate texts from different domains of use, often used as a basis for comparative research.
Specialized corpora
Corpora like the Michigan Corpus of Academic Spoken English were designed for a specific type of research, containing texts of a type needed by a given study. The aim of specialized corpora is not to depict the language as a whole, but to show particular parts of it. The database is much smaller when compared to general corpora as the focus of analysis is narrower. Moreover, the limited number of texts may be an advantage in specialised areas such as academic English where the corpus needs only to represent this language variety.
Spoken vs. written corpora
Even though spoken language is the most common type of language use, the majority of research performed with the use of corpora is based on written corpora. This is due to the fact that in order to create a spoken corpus greater resources have to be devoted. Designing such a corpus is much more time consuming as requires the transcription of spoken language, involving complex phonetic and prosodic features. Written corpora are easier to assemble as they do not require considering characteristics of spoken language as hesitations, fillers or ungrammatical constructs and also whether orthographic representation will be adequate or whether annotation on diverse levels should be supported.
Bilingual and multilingual corpora: parallel and comparable
Originally, bilingual and multilingual corpora were created for the purpose of mechanical translation, which was the primary objective of computational linguistics. Nowadays, they are not only used for translation, but also for contrastive studies. When compared to monolingual corpora, the two types of bilingual and multilingual corpora, parallel and comparable, offer many more possibilities as it easier for linguists to gain new information about the features of a given language. A question arises, what is characteristic for the two aforementioned types of bi- and multilingual corpora? Parallel corpora, which are mostly used for translation studies, are designed to incorporate texts and their translations into one or more languages, however, the pairs of translated texts may sometimes be translations of a third text as in computer manuals. Comparable corpora, frequently used in contrastive studies, consist of texts from various languages, but are not direct translations. The given samples are not connected in the same way as in a parallel corpus, but were selected because they have a number of common features such as the type of the text or topic.
Summing up, corpus linguistics as an approach to language study provides the necessary tools to investigate the ways languages are used by means of diverse methods as well as a number of different corpora.
Corpora in Cognitive Linguistics
According to Grondelaers, Geeraerts and Speelman (2007: 149), "the methodological position of Cognitive Linguistics vis-à-vis corpus research (and the use of empirical methods at large, for that matter) seems to be characterized by a certain amount of reluctance." Although corpora have been used in Cognitive Linguistics from its creation, there exists, a distinction between the early European and early American studies in this field. The research conducted in Europe was rather corpus-based whereas American research was primarily introspective. The fact that Cognitive Linguistics is often regarded as a non-objectivist approach to the study of language whereas corpus-based research tries to be as objective as possible may be one the reasons for such a situation. Nowadays, however, there is a tendency to perceive Cognitive Linguistics as a usage-based approach, meaning that the study of actual language and its usage is crucial. Furthermore, Cognitive Linguistics highlights the fact that the knowledge of the world human beings posses is not just a passive reflection of world but rather an active construal. The linguistic knowledge we have may differ from culture to culture that is why broader empirical basis is needed, which can be provide by corpora. Adopting a methodology as the use of corpora in the field of Cognitive Linguistics may result in a number of consequences. First of all, research conducted by linguists may become more synergetic as developing new tools for corpora study as well as creating new corpora requires a lot of resource not likely to come from individuals, but rather organized groups. Second of all, the data coming from corpora may become a shared basis against which hypotheses can be tested, making them more comparable than they are now. Finally, studying the language is likely to become slower as perfecting a hypothesis will require the analysis of large amounts of data, which is time consuming.
In conclusion, although the view that Cognitive Linguistics is a usage based approach is becoming more and more popular, the use of such methodologies as the use of corpora is still infrequent, however, many scholars are turning towards this direction.
Models Used in Cognitive Linguistics and Corpus Linguistics
Annotating resources in Corpus Linguistics
When discussing the future of corpora, the already present trend to use mark-up languages must be mentioned. The early corpora were prepared as 'plain texts', and attempts were made after a while to keep a record of some aspects of the document formatting which could help in interpretation - headlines, bold face, etc. Texts rapidly became difficult to analyse because of the lack of any agreed conventions for digitizing them. In the 1980s, SGML (Standard Generalized Mark-up Language) became widely used for transferring texts from one host to another, and in 1986 it was accepted in the description of the norm ISO 8879. SGML established the convention of using diamond brackets to separate text from tags. Tag became the universal word for whatever went inside the diamond brackets, as shown below (from Burnard 2006: 517, tags marked in bold by myself):
<conversation>
<utterance speaker='Mutt'>My dog's got no nose.</utterance>
<utterance speaker='Jeff'>How does it smell?</utterance>
<utterance speaker='Mutt'>Terrible.</utterance>
</conversation>
SGML gives no guidance on what the content of the tags should be, but prescribes only the form. It is therefore a flexible method for making notes of almost any kind on a text, not only for restoring lost formatting information. Once corpus providers had accepted SGML for document management and encoding, they were pleased to find other uses for the conventions and it is highly likely that SGML tags will develop in the future.
Nowadays, the rules regarding markup languages are predominantly set by the World-Wide Web Consortium (W3C, http://www.w3.org), as the languages interpreted by web browsers are based either on SGML or XML (eXtensible Markup Language) - a successor to SGML. It appears to be stricter in terms of the usage of tags, but also provides additional features, such as the possibility to combine an opening and closing tag within a single pair of diamond brackets (e.g. the line-break '<br/>' equals '<br></br>'). This is useful if no textual content is allowed between the opening and closing tag - in building language corpora, it can be used to indicate 'invisible' elements of language, such as null morphemes.
It is widely felt that the linguistic study of corpora is aided if information about parts of speech are available, and many corpora now have each successive word followed by a tag indicating its part of speech. Other kinds of analysis, semantic, pragmatic, stylistic, etc. have already been developed, and spoken texts are subject to the addition of tags concerning rhythm, tone, intonation, pausing, etc.
Image Schemata, Conceptual Metaphors and Embodiment
Our imagination plays an essential role in the way we organise and use language. As will be mentioned in the following sections, people prefer memorising concrete objects, instead of abstract ones. Another tendency is that we develop universal patterns (some kind of basic framework), to which new experiences can be adjusted. These 'mentally visual' patterns are referred to as image schemata. The term has been used as early as in 1987 by both Lakoff and Johnson. Fig. 1 presents a containment schema, one of many pictorial representations presented in Johnson (1987), and one which is frequently provided as a simple example of how the mind 'sees' a certain selection of objects, concrete (tangible) or abstract.
Fig. 1. Containment schema (source: Johnson 1987: 23)
The above illustration can represent a great variety of experiences. As the back cover of Johnson (1987) explains: "meaning, understanding and rationality arise from and are conditioned by the patterns of our bodily experience. (...) Imagination links cognitive and bodily structures, showing that such basic concepts as balance, scale, force, and cycles emerge from our physical experiences and can be metaphorically extended to express abstract meaning and rational connections."
The schema shown in Fig. 1 presents a trajector (marked by 'X') and the landmark represented by the large circle surrounding the trajector. In the simplest terms, the relation between the two is that the trajector is IN (or INSIDE) the landmark, and can be exemplified by sentences involving concrete entitities, such as "The book is in the box", as well as more abstract objects, as in "The price includes VAT" or "I was possessed by anger. As Johnson (1987: 21) describes, we are aware of our bodies as three-dimensional containers, we experience physical containment in our surroundings, while moving in and out of rooms, clothes, vehicles, etc. This particular set of examples shows that containment is a typical image schemata - even if the objects appearing in an utterance are unknown to a person participating in a communication process, there are ways (e.g. the use of prepositions) to comprehend the relation between two or more entities.
Moreover, there are multiple consequences of the in-out orientation delimited by the aforementioned example of image schema. First of all, that particular type of experience might involve protection from, or resistance to, external forces (just as glasses in a dedicated case are protected from a harmful impact). Secondly, containment restricts forces inside the container (e.g. walls in a room limit a person's movement). Due to this limitation of forces, the contained object can e.g. remain in a fixed location delimited by the boundaries of its 'box'. In turn, this fixture enables/disables the visibility of the trajector from a certain perspective.
Eventually, Johnson (1987: 22) points to the fact that particular mental schemata follow some sort of logic. For example, if object A contains object B, and object B contains object C, then object A also contains object C. Thus, if the sentences "The book is in the box" and "The box is in the room" are true, then it is also true that "The book is in the room". It may sound simple, but our minds use such techniques subconsciously, on a regular basis.
Metaphor was originally defined by Aristotle as "giving the thing a name belonging to something else, the transference being (...) on the grounds of analogy". In the traditional views, metaphor is defined as a non-standard meaning used to obtain a poetic effect. As Lakoff and Johnson (1980: ix) observed, most traditional philosophical views believed that metaphor had little, if any, role in understanding world and people. The two linguists discovered solid evidence showing that metaphor is present in everyday language and thought.
That is because a large number of metaphors can become conventionalised, such as: "love is a journey", which can be seen as an instance of more general cross-domain mappings, briefly introduced in the sections above. There are also some other instantiations of a more general mapping between long-term activities and progress along a path, e.g. 'Career is a journey', and even 'Life is a journey'. The latter is a part of a very abstract mapping scheme known as the event structure metaphor (see Lakoff and Johnson 1999). These mappings respect and conform to all the cognitive instruments and frameworks mentioned in the course of this chapter.
For example, by agreeing with the fact that life is a journey, we use the notions of the source and target domain - as a result, we can refer to 'progress in life', we may speak of what is "ahead of us" and "leave behind" unpleasant experiences in life. Moreover, we can refer to image schema and notice that life perfectly fits the source-path-goal schema defined by Johnson (1987: 28, 113-117). Eventually, it is also possible to evaluate the 'standard' features of life according to our mental representation of what someone's life should be like and compare it to actual instances of people's lives. It seems almost natural that we are treating the concept as if it were a real journey.
Lakoff and Johnson believe that the source of metaphor is connected with so-called 'embodied human understanding' (Lakoff and Johnson 1999: 16-45). They claim that a metaphor is not an intellectual abstraction and restricted to poetic use. Both meaning and metaphor derive from people's everyday bodily experience. The source of metaphor can be described as a variety of physical experiences from which we extract image schemas, dynamic patterns of people's perceptual interactions (our observations) and motor programs (our actions).
It is an important fact that the understanding coming from a metaphor may be shared, which means that all members of a particular culture can experience similar feelings and have similar associations. The image schemas that construct the understanding constitute a large part of what humans mean by the form itself in their experience. The same idea is also found in an earlier publication entitled Women, Fire, and Dangerous Things by George Lakoff (1987). The most important thesis in the book states that the rules of reason emerge from the bodily experiences. Moreover, imagination transforms the general schemas, defined by the experience, into forms of reason.
As already mentioned, the cognitive storage (our memory) is based on individual experience. That is why, sources of metaphors are found in categories that are already rooted in human memory - these are already existing categories of objects (the ones we already know). Metaphors work as 'cognitive instruments'. For example, when one says 'Time is money', s/he does not use this metaphor linguistically, but considers time, which is the target category, through the structure of source category (money). As time in English language and culture is considered a precious concept in relation to money, the link between the two categories can be also found in this example and many other metaphorical expressions, such as: wasting time, running out of time or spending time.
These expressions reflect human mental perception of the concepts. According to the definition given by Ungerer and Schmid, the process of applying features of one category to a new one is called "mapping of the structure of a source model onto a target model" (Ungerer and Schmid 1996: 120).
Another observation is that the target is often represented by an abstract concept, while the source, is a concrete (tangible) one. It shows again that people make use of concrete words to conceptualise abstract events. Here, to illustrate an abstract concept of time; the word money is used only because it refers to a more tangible object, something that people can see or touch. Thus, conceptualisation depends on a number of factors, such as: our experience, people we encounter, everyday objects, actions, and events, but most of all, it is the result of our way of thinking.
Idealized Cognitive Models and Categorisation Models
Idealized Cognitive Model (ICM) is a term coined by George Lakoff (1987). It refers to a device that is used by people to create a subjective representation of the world in their minds. Lakoff's models represent our knowledge about various categories of objects. This method of remembering objects and experiences is the same for all people, but individual 'memories' of certain categories of objects (or clusters of such categories) are different for every human being. However, people from the same cultural background can share significant similarities in perception.
According to Taylor (2006: 571), ICMs capture the fact that knowledge about a particular domain may be idealised and may not fit some specific situations. For example, when one tries to imagine a chair, we see a piece of furniture, with four legs that is used to sit on. Nevertheless, we all see a different object in our minds. That is because the vision is based upon our memory of "chair". If somebody remembers two similar categories, like "chair" and "stool", he may consider attributing a certain object to either of the categories, basing on his previous associations - that is the reason why people do not always agree on naming specific things. It is all a matter of previous experience and the resulting memorisation.
Many semanticists insist on the need to make a distinction between the meaning from the dictionary and the encyclopedia, i.e. between what one knows from knowledge of a language and what one knows from his experience of the world. Cognitive semantics does not approve of such distinction. A person's linguistic knowledge agrees with the person's total world knowledge - this is why it is so easy to forget words, which are meaningless to us.
An individual word provides access to a small segment of encyclopedic knowledge, but there are no limits to which such a segment of knowledge can be extended using additional words. The encyclopedic nature of linguistic semantics is found in the terms, such as profile, base, domain, and Idealized Cognitive Model (ICM). The terms 'profile' and 'base' were coined by Langacker (1987). By saying something, we make a reference to our memory of a certain object. This is often illustrated on the example of the word "hypotenuse". The word designates a straight line, but only that one which constitutes a specific element of a right-angled triangle.
Although hypotenuse does not designate the triangle, the notion of a triangle is essential for the understanding of the word. The profile of the object is simply a straight line, but it is not enough to define the word precisely. What we need is a base, i.e. the triangle. In fact, it is axiomatic, in cognitive semantics, that all expressions achieve their meaning through profiling against the relevant background knowledge.
Langacker's Theory of Domains is complementary to a somewhat earlier framework developed by Fillmore under the name of Frame Semantics. Both theories underline the fact that meaning is essentially encyclopaedic and stress the fact that words and concepts to which they refer cannot be properly understood without the existence of a larger, hierarchical knowledge structure (Evans and Green 2006: 230).
Returning to the example from the last section, it is evident that the base (triangle) presupposes broader knowledge configurations, namely, those pertaining to planar geometry, which themselves are based in notions of space and shape. These broader knowledge configurations are referred to as domains. Some domains may be basic, which means that they cannot be reduced to other domains. Examples of such domains include time, space, colour, temperature or weight. A knowledge structure, no matter how complex, can also function as a domain, e.g., the rules of a game, a scientific theory, kinship networks, gender stereotypes, educational, political, and legal systems. Furthermore, domains are created by beliefs about life, nature, causation, the supernatural, etc. (Taylor 2006: 570).
Most concepts are characterised using a set of more than one domain. Uncle, for example, uses a male human being against the base of a (portion of a) kinship network (family relationships), specifically, that part of the network that relates the uncle to his nephews/nieces. It is more complicated, however. The meaning of kinship depends on other terms, such as gender, procreation, marriage, inheritance and others. Moreover, uncle profiles a human being, which is understood against multiple domains typical of life forms, and specifically to three-dimensional bodies and their various parts, with their features of weight, extension, shape, and so on (Taylor 2006: 570).
Domains serve the processes of categorisation and conceptualisation, but the example presented above shows that items can be embedded within a complex network of dependencies. The theory of domains binds concepts existing in our minds as ICMs into larger structures governed by relationships, such as the one between the base and the profile.
The next approach to the issue of categorisation was proposed by Lakoff (1987: chapter 6), as he attempted to define words as radial categories. In his book, the term is introduced using the example provided by Lakoff's student Claudia Brugman whose MA thesis was related to the polysemy of the word 'over'. The preposition has multiple senses and evades simple definitions, that is why the dictionary entry for the word usually occupies much space and requires a lot of examples.
What is most important in tackling the problem of different meanings of the same lexeme is the fact that some of them can be considered more prototypical (central) or less prototypical (peripheral). For example, Brugman (1981) states that "over" is more commonly associated with its meaning, which is synonymous to "above" rather than as a determiner of control (as in "take control over something").
However, despite the differences in frequency, Lakoff claims that all senses of a word are interconnected - he rejects the idea that each meaning occupies a separate "mental space" in the mind. His view promotes the study of polysemy, which is a phenomenon that can be explained by the extension of meaning occurring due to various cognitive mechanisms.
According to Feldman (2006: 99), radial categories are very complex and particularly interesting. He provides a definition based on an example of the concept of 'mother'. The typical cases of mother are marked by properties, such as birth, nurturance, genetics, marriage and culture. All of these features apply in the central case, but there are many situations (e.g. with 'stepmother') when only some of the typical roles can be applied. Feldman notices that in some cases only one feature is present, as it is with a woman who donates genetic material to a child, but does not give birth or raise the child. The woman is a type of mother who only satisfies the 'genetic' feature.
The example is interesting enough to be described in more detail to properly understand the radial nature of some categories. Understanding the set of criteria is essential to categorise a specific case as more or less central. At times, the meaning of constituent words can be somewhat misleading. For instance, the concept of a 'working mother' is defined in relation to nurturance. It is insufficient if an employed woman gives birth to a child, but later places the baby for adoption. Although some criteria for being a mother are met and the person is professionally active, she cannot be considered a working mother. The woman must actually be raising the child.
Basing on this example, radial categories rely on the fulfilment of certain predefined conditions. The most central cases are those which comply with the whole structure of related properties. On the other hand, the lower number of criteria is met, the less central a particular instance is. As Feldman (2006: 99) notices, sometimes only a single property has to be met for a case to be included in the category. However, the example with the working mother indicates that there are conditions, which create a sort of border for radial attribution - in logical terms they could be termed necessary conditions.
Langacker's concept of a network model used in the process of categorisation can be treated as an alternative to Lakoff's radial models presented above. Moreover, it joins many of the aforementioned ideas together, including image schema, ICMs and radial categories. It also provides additional support for the cognitive account of polysemy (described in the next section).
Langacker's framework involves a network model that represents the structure of categories. In this model, members of a category are viewed as nodes in a tree. The links between nodes in a network indicate relationships resulting from categorisation which connects the symbolic (linguistic) units stored in the grammatical inventory. Langacker's network model is described in brief by Evans and Green (2006: 545-546). Three sorts of relationships are presented in the figure below.
Fig. 2. Langacker's Network Model
Firstly, there is a relationship joining an extension to a prototype. A prototype is Langacker's terms resembles Lakoff's ICM. The extension shares selected (not all) attributes of its prototype, hence it is categorised as an 'instance' (example) of that category. Just as Lakoff argued: there is an idealised vision of an object, and there are 'real' objects, which fit this vision, but also possess features that are not considered in the ideal prototype.
Secondly, there is a type of categorising relationship between schema and instance. A schema reflects the nature of the model and structures the related units as a category within the network; new items can be compared against such categories. It can be said that a schema contains a definition for the model type. According to Langacker, the network can grow 'upwards' via extended schematisation, 'outwards' via extension of the model and 'downwards' as more detailed instances are added (Evans and Green 2006: 546).
All elements are intertwined: the mental definition of an object, its imaginary prototype and actual instances which are more or less similar to the prototype (or in Lakoffian terms, more or less central). Langacker claims the network model characterises not only polysemous elements, but also other kinds of linguistic categories including items related to sound, meaning and grammar (Evans and Green 2006: 546).
Conceptual Metaphors and Language Corpora
In recent years, computerised corpora have been used to study linguistic metaphors from a range of perspectives. Some writers have used corpora to compare and contrast the use of metaphors in different genres, notably Charteris-Black (2004). Cross-linguistic analyses have been carried out into the frequencies and meanings of the different metaphors that are used to discuss the same topic across different languages, for instance in work by Boers and Demecheleer (1997). Unlike these types of work, the corpus studies reported here do not focus on specific genres or topics, but attempt to look at some metaphorical patterns found in a general corpus of English. Rather than considering specific topics, meanings or ideological orientations, this work examines detailed linguistic patterns and contrasts them with the patterns found in literal uses of the same words. These patterns are considered against the backdrop of Conceptual Metaphor Theory. It will be argued that the patterns found in naturally-occurring linguistic metaphors are a type of evidence that has been somewhat neglected to date, but can usefully inform theoretical development.
The literal and metaphorical senses of words are from the same part of speech: literal nouns are used metaphorically as nouns, and so on. However, once concordance data is studied in any amount of detail, another pattern emerges (Deignan 2006: 109). There are many words that have pairs of meanings, apparently related to each other by metaphor, that are not the same part of speech. Squirrel is one such word. Used with its literal meaning, it is a noun, but with its metaphorical meaning it is rarely or never used as a noun; instead it is well-established as a verbal metaphor, in corpus citations such as:
(1) as consumers squirrel away huge sums for the downpayment on a home.
The verb to squirrel is very rarely or never used to refer to the behaviour of squirrels or similar animals, and thus there is no overlap of form between literal and metaphorical meanings.
According to the instructions provided in Meyer (2004: 30), before the texts to be included in a corpus are collected, annotated, and analysed, it is important to plan the construction of the corpus carefully: what size it will be, what types of texts will be included in it, and what population will be sampled to supply the texts that will comprise the corpus. Ultimately, decisions concerning the composition of a corpus will be determined by the planned uses of the corpus. If, for instance, the corpus is be used primarily for grammatical analysis (e.g. the analysis of relative clauses or the structure of noun phrases), it can consist simply of text excerpts rather than complete texts. On the other hand, if the corpus is intended to permit the study of discourse features, then it will have to contain complete texts.
