Cognitive Science: An Introduction/Linguistics Methods
- 1 Introduction to Linguistic Methodologies
- 1.1 Linguistics Uses a Top-Down Approach
- 1.2 Why Study Linguistics in Cognitive Science?
- 1.3 Data
- 1.4 Methods
- 1.5 Structural Analysis
- 1.6 Functional Analysis
- 1.7 Neurological Methods
- 1.8 Deficit Studies
- 1.9 Psycholinguistics
- 1.10 Computational models of language
- 1.11 Phonology
- 1.12 Output data
- 1.13 Ways of Avoiding Threats to Internal Validity
- 1.14 References
Introduction to Linguistic Methodologies
Linguistics Uses a Top-Down Approach
There are two ways in which to study a cognitive phenomenon: top-down and bottom-up. Top-down analysis starts from behavioural and introspective data to give an abstract characterization of cognitive processes. The bottom-up approach starts from the analysis of entities that are smaller in scope and build up to an explanation of the phenomenon. This includes starting with low-level information systems (either in computers or neurons) which to give us information on the basis of cognition. Thus, Linguistics, being a top down-study, creates abstract theories of mind which are then tested by cognitive neuroscientists, cognitive computer scientists, and psycholinguists to see if they may be plausible for cognitive models and theories of mind (MacWhinney,1999, 402-403).
Linguists often find "puzzles" in language. For example, we can say "give the muffin to the mouse," and we can say "give the mouse the muffin." However, though we can say "They drove the car to Chicago" and "I gave heather a headache," you cannot say "They drove Chicago the car" or "I gave a headache to Heather." Why? We can't tell just by introspecting--that is, just thinking about how you do it. Linguistics must come up with a theory of why we treat these sentences differently.
In this case, muffins and headaches differ in motion (nothing moves when you give someone a headache) and in terms of causation (giving someone headache means you did something to cause a headache). So the linguist might speculate that we have categories in our heads for these kinds of transactions that result in differing grammatical structures.
Why Study Linguistics in Cognitive Science?
Language use is a cognitive phenomenon that is incredibly complex, and is one of the things that distinguishes humans from other animals. Our language is much more complex than any other known species' forms of communication. Many believe that language plays a large role in human cognition. We learn words at rapid rates, learning about 45,000 words around the time the average person graduates high school. Human brains are specialized for learning language. Our use of and ability to learn language are thus important cognitive phenomena that require study.
There are several legitimate sources of linguistic evidence: 1. The grammatical intuitions of a linguist who is a native speaker of the language in question 2. Data collected in experimental settings from native speakers of the language in question (informants or experimental participants) 3. Transcriptions of actual spoken language (text)
The Three Claims of Text as Evidence.
Linguists have three stances on the acceptability of written text (transcribed from speech) should be used as data.
The Strong Claim: Texts are the only evidence in linguistics.
The Weaker Claim: Texts are the best evidence in linguistics; (better, in any case, than native speaker's intuitions)
The Weakest Claim: Texts are very good evidence in linguistics (Stainton, 1994, 1-18).
Many Cognitive scientists base models of grammar upon Chomsky's Theory of generative grammar, which explicates that grammar is generative (can create innumerable new sentences) and that the best judgment is a native speakers' as to which sentences are grammatical (Chomsky, 1957). This, along with some spoken languages having no text basis/written language give insight into what most Cognitive Scientists claim. Thus, many linguists dismiss the Strong Claim, are wary of the Weaker Claim, while most find Weakest Claim to be acceptable (Stainton, 1994) .
--It is to be noted that the proper term for one participating in a Linguistic test is an 'informant' : one who is a native speaker ; an authority figure on grammaticality. This term is specific to linguistics. The term participant will be used as well, as a general term for a more person participating in any type of experiment. (This will be done to not over use the term informant.)
The goal of structural analysis is to study the underlying structures and rules of a language. Language is said to have specific rules, which Noam Chomsky defined as a Universal Grammar (UG) (Chomsky, 1957). This being that all languages had universal rules to them. UG states that children learn structure dependent rules. This means that there is an abstract grammar structure for all languages based on hierarchical grouping of words, based on their role in language. All words are grouped into parts of speech: nouns, pronouns, verbs, adverbs, adjectives, conjunctions, prepositions and interjections, from then they are hierarchically organized by rules into sentences which form our grammar, or the universal grammar of languages. These analysis of structure in phonology, morphology, and semantics give an abstract structural representation of language which give insight into language formation, acquisition and use, all of which are key aspects of cognition (Frank 460). A large amount of structural analysis is done with texts, mathematics, and grammaticality judgments (see below).
The goal of functional analysis is “to determine the affects of function, being that forms of natural languages are created, governed, constrained, acquired, and used in the service of communicative functions, on language” (MacWhinney 402-403).
Some aspects where functional analysis are key:
Cue and form reliability: this means learning cues and forms for specific words (topicaztion). Wherein, if the cue markers were taken out our switch the sentence would be incomprehensible to the listener. Cue markers are governed by function and can vary from a simple allophones to word order in a sentence (MacWhinney 408). Such as in English, word-order usually distinguish what category the word is in.
Lexical influences on syntax: From the sentence 'Child's Stool is Great for Use in Garden' one is able to interpret it child's feces being great for compost in the garden, or a small stool, typically used by children, is great for kneeling onto when gardening (Pinker 94-95). Two largely different sentence meanings. Yet, based on the structure one interpreted it as, means the difference between harvesting child feces for fertilizer and buying a cheap stool to help your sore back. In order to find which one is grammatical and judge it based on a criteria of how likely it is to be understood a certain way is a type of functional analysis.
Complex grammar paradigms: This is the study of why certain words of language are categorized in certain ways (e.g. Why certain words are masculine or feminine in some languages.) Functional analysts look for cues and forms to predict categories for certain words (MacWhinney 410).
Most of the data used in functional analyzing is from studies of language typology of surveys of lexical patterning (MacWhinney 411).
Grammaticality tests are given to informants to judge whether something in their idiolect is grammatical. This is the main source of empirical evidence for functional analysts searching to analyze complex grammatical paradigms along with structural analysts to further their studies of grammar.
Neurological Methods attempt to localize brain areas with roles in cognition. They are able to correlate damage, experimental changes, abstract cognitive processes, and language to parts of the brain.
Deficit Studies are studies that test and naturally observe those with neurological deficits (e.g. lesions, tumours, trauma, mutations) on language centres in the brain. Famous deficit studies include Broca's Area and Wernicke's area (Firth 380-385).
Transcranial Magnetic Stimulation (TMS)
TMS is a technique which apply magnetic forces to certain parts of the brain. This makes a target area of the brain unable to operate properly causing almost temporary controlled lesions.* Which can be studied as if they were deficit studies.
Transcranial Electric Stimulation (TES)
TES is the opposite of TMS. It applies electrical forces to certain parts of the brain. This gives the target area a large amount of stimulation. When non-localized parts of the brain are stimulated one is able to see its affects on cognition.
Magnetic Resonance Imaging (MRI) and functional Magnetic Resonance Imaging (fMRI)
MRI and fMRI record oxygen levels in the brain. Solutions are added into the brain and then the oxygen is able to be measured. Change in oxygen levels mean change in brain stimulation (Buckner, and Peterson 413-415). The difference between a MRI and fMRI is that the fMRI the participant is doing a task while the measuring is being done. The informant may be asked to read, write, or think about something. A MRI is just a patient sitting still not doing any task.
Positron Emission Tomography (PET)
PET tests indicates increases in blood flow within the brain. A slightly radioactive solution is added to the participants blood and the radioactivity is measured overtime while the blood flows throughout the brain (Buckner, and Peterson 413-415).
An EEG is a way of mapping electrical communication of neurons in the brain. The machine itself is a hairnet like hat with many nodes on them which measure the spread of electrical activation through the skull. Seeing the spread of electrical communication gives us insight into how the information is being processed on an electrochemical level.
The area of psycholinguistics is the study of how humans psychological make sense of, and process a sentence. In the beginning of psycholinguistics, which includes work from Miller, Mckean, and Slobin, informants were tested on their ability to process sentence, the experimental aspect was the transformations in a sentence's derivation the reader had to make (Abrahamsen, Bechtel, and Graham 41-42). What was found was the more transformations the longer time it took for participants to read and understand the sentence.
This was refuted and proven empirically wrong by modern psycholinguistics, including Fodor, Garrett, and Bever, which stated that the relationship held between linguistic performance and competence is more abstract than originally thought (Abrahamsen, Bechtel, and Graham 41-42). This brought upon a paradigm shift in psycholinguistics, straying away from transformations of sentences and into the study how language is internally represented and processed on a psychological level. Psycholinguistics caused computational models for linguistics to emerge, attempting to use computers to represent the internal processes and information being used and transformed in deep structure.
Computational models of language
Computational models of language are used to study (a) the study of computational models of the structure and function of language, its use, and it acquisition; and (b) the design, development, and implementation of a wide range of system such as speech recognition, language understanding and natural language generation (Joshi 162).
Parsers in computational linguistics refers to computer programs that attempt to parse a language based on certain coded rules (Joshi 163) . This is an attempt to try and re-create the ways in which humans may represent, and manipulate linguistic data, along with how we form rules and grammars. Parsers give us large amounts insight and constraints on our abstract theories of languages and mind. The goal of parsers is to parse only grammatical observations and grade them on a specific criteria. Just as we humans would do with natural language. We only understand somewhat grammatical sentences and if we don't fully understand it we are able to grade it against other more/less grammatical sentences along with sentences that don't make sense at all.
Statistic data of language from computational a decently new field of study in computational models for language(Joshi 164). This uses corpus data (see below) in order to statistically find regularities in language (e.g. Two words being found beside each other 25% of the time either one of the words is present). This leads to large statistical data about language and from that one is able to form language rules. This along with Behaghel's first law, words that belong together mentally are placed close together syntactically gives insight into mental storage of words (MacWhinney 404).
The study of the physical aspect of speech (i.e. The sounds made to create speech, more specifically phonemes and allophones). This branch of linguistics has the most empirical evidence for their claims. Phonology does not have a large impact on cognition. Yet, Phonological studies have given good insight to a “critical period” and Chomsky's language acquisition device(Abrahamsen, Becthel, and Graham 41-42 ). Some speakers of one language can differ between two allophones that speakers of another language could not distinguish, even though they both exist in the both their languages. This means we may acquire only some allophones to which we can hear, because the allophones we do not learn may not be an important aspect for distinguishing terms. Therefore, some of our cognition for classifying terms rely on Phonology.
Almost all output data within Linguistics are published in journals, books, or electronic articles.
Ways of Avoiding Threats to Internal Validity
There are many ways to assure, in a Linguistic study involving informants, that you receive reliable and powerful data: Make informants comfortable. Have breaks every so often and allow them to relax. Those being questioned may make mistakes or change their mind. When this occurs you may want to rephrase the question or ask it in a different way. Rephrasing difficult or complex questions increases your power in a study. Making sure you were able to draw conclusions where connections actually existed. Informants have a bias towards 'yes'. So, adding a 'not sure' option will reduce the unreliability of your testings. Having yourself as an informant is not typically done and is not recommended. You probably hold a bias towards a favoured outcome. This will result in unreliable data. Using corpus DATA, or a body of data, that many linguistic researchers have used before to test their hypothesis. This is largely used in computational models, when attempting to test the performance of their programs versus others in an objective manner. Always re-check data, even if the data may disprove your hypothesis. Be thorough in your investigation and be as fair as possible with competing data.
Abrhamsen, Adele, William Becthel, and George Graham. "The Life of Cognitive Science." A Companion to Cognitive Science. Paperback ed. 2. Malden, MA, USA: Blackwell Publishing Ltd, 1999. Print.
Buckner, Randy L, and Steven E. Peterson. "Neuroimaging." A Companion to Cognitive Science. Paperback ed. 1. Malden, MA, USA: Blackwell Publishing Ltd, 1999. Print.
Chomsky, Noam. Syntactic Structures. 1. 1. The Hague: Mouton, 1957. Print.
Firth, Christopher D. "Deficits and Pathologies." A Companion to Cognitive Science. Paperback ed. 1. Malden, MA, USA: Blackwell Publishing Ltd, 1999. Print.
Joshi, Aravind. "Computational Linguistics." The MIT Encyclopedia of Cognitive Sciences. 1. 1. Cambridge, MA: MIT Press, 1999. Print.
MacWhinney, Brian. "Functional Analysis." A Companion to Cognitive Science. Paperback ed. 2. Malden, MA, USA: Blackwell Publishing Ltd, 1999. Print.
Pinker, Steven. The Language Instinct. Modern Classes ed. 3. New York: Harper Perennial, 2007. Print.
Stainton, Robert J. "Texts as Evidence in Linguistics." Carleton University. 1.1 (1994): 1-18. Print.