Data Collection & Analysis Procedures

LANGUAGE ACQUISITION

Data Collection & Analysis Procedures

Abdureim I. Abdurashytov

ELT-517

29 May 2013

Nicosia

In general, speech perception studies are interested in speech discrimination of children, especially infants, and the ways in which these abilities may aid language learning. Recent advances in technology, especially digital recording and computers, have aided researchers in isolating, reproducing, and combining sounds for research. Infants can even be tested while still in the womb for their responses to speech sounds in isolation and in connected speech. Responses may consist of moving or kicking. With older children and adults, speech perception is often tested with more specific responses, such as pointing. One new approach is called online or real-time research in which responses are paired with brain-imaging techniques, such as magnetic resonance imaging (MRI) to identify areas of the brain where perception occurs.

Language comprehension studies are interested in understanding. Subjects usually respond

to structured procedures by looking, pointing, acting out, or following directions in response to

a spoken or written stimulus. Of necessity, this type of research requires a standardized, structured experimental design to ensure that all subjects have the same input.

Expressive language studies can take a number of formats from very structured and experimental to more open-ended and observational. The primary difference is the degree of control the experimenter has over the context. So, I’ll be primarily writing about expressive language studies in the following sections.

As mentioned, expressive language-development data are usually collected in two ways:

spontaneous conversational sampling or natural observation and structured testing or experimental manipulation. Each method raises issues of appropriateness for the language feature being studied. Either one alone may be insufficient to describe a child’s linguistic competence, that is, what he or she knows about language. Data yielded in one context may not appear in another. For example, in a studyof pronouns in which the author participated, children produced a wider variety in conversation and produced more advanced forms in more formal testing (Owen, 1985). Other researchers have also found that formal elicitation tasks, such as testing procedures, produce more advanced child language than conversational sampling. Ideally, the linguist would employ both informal and formal or structured approaches, using the structured procedures to obtain more in-depth information on the data collected by the more broad-based naturalistic or informal procedures.

Some researchers prefer testing or experimental manipulation in order to control for some

of the variables inherent in more naturalistic collection. Within a test or experimental procedure,

various linguistic elements may be elicited using verbal and nonverbal stimuli in a structured

presentation. Such control of the context, however, may result in rather narrow sampling.

Formal procedures enable researchers to gather data that may not be readily available using conversational or observational techniques. Forexample, it is difficult to assess children’s

comprehension or their metalinguistic skills without direct testing. Some hypotheses cannot

be tested directly, however, so researchers must test indirectly or observe some features of

language development.

Language and experimental factors must be manipulated with caution. One aspect of language can affect others, even though the researcher does not intend for this to happen. For example, among both children and adults, new information introduced into a conversation is consistently more phonologically accurate than older, shared information. Thus, pragmatics influences phonology.

Likewise, experimental factors can have unintended consequences. For example, a researcher may highlight an item in a picture in an attempt to ensure a child’s accurate comment. Unfortunately, although the accuracy of the message is not increased when one item is marked, the amount of redundancy or inclusion of irrelevant information does increase.

In addition, testing and experimental tasks do not necessarily reflect a child’s performance in everyday use. For example, in an experimental task, a child may rely on on-the-spot problem-solving techniques rather than on his or her own everyday operating theories and hypotheses. In addition, noncompliance with testing or experimental procedures may not mean noncomprehension or lack of knowledge. Especially with preschoolers, incorrect responding may indicate a lack of attention or interest.

Language processing is not a single unitary operation as is often assumed in test construction, but consists of component operations, such as lexical or vocabulary access, syntactic decoding, and discourse processing, that are engaged at different times and with varying degrees depending on the linguistic task. So-called offline test tasks, as in fill-ins, or providing the missing word, measure only the end points of several linguistic processes.

During offline testing, components of the overall process are overlooked. For example, the process of guessing the missing word may be the reverse of what happens in conversation. Conscious guessing is too slow in conversation. Rather than context aiding in predicting the

next word or phrase of the speaker, contextual information seems to provide a check that correct items have been uttered. Although such offline language collection techniques may tell us what children know, they tell us little about how children process or access language.

In contrast, online tasks attempt to measure operations at various points during processing and describe individual and integrative components. For example, at what point in the cue “Mary has a blue dress and a red dress; she has two_____”does the child access the word dresses? We might be able to determine this information by the online technique of asking a child to paraphrase or answer yes/no questions after only limited information is presented. For example, if we say “Mary has a blue dress and a red dress,” a child may access dresses based on and or red or dress. Online techniques would be interested in discovering at which point this occurred. Techniques can be much more elaborate than this simple example suggests. Although still in their infancy, online techniques are beginning to provide valuable linguistic-processing data.

In short, testing and experimental data may be very accurate but very limited.The results must be examined within the context of the specific tasks designed to elicit certain behaviors. A better measure is the consistency of use of a language feature across various tasks.

Jerome Bruner, renowned child development specialist, began his career studying language in very controlled situations, analyzing discrete bits of language. The model was confining, and the language data felt artificial. He then began studying children at home, videotaping open-ended interactions with their families. As a result, his later data hada more authentic quality to it. Naturalistic studies, such as language samples, may yield very different data than experimental manipulations.

Child language data may also be obtained from the CHILDES system of database transcripts. The system includes programs for computer analysis, methods of linguistic coding, and systems for linking these transcripts to digitized audio and video.

Any given naturalistic situation may be insufficient for eliciting a child’s systematic knowledge of language. Nor is there certainty that a given test situation will represent a child’s naturally occurring communication. Thus, it is best to have data from a combination of collection procedures. In either case, the linguist is sampling the child’s performance. The child’s linguistic competence—what he or she knows about language—must be inferred from this performance.

The researcher must be concerned about two samples: the sample or group of children from

whom data are collected and the sample of language data from each child. In both samples, the

researcher must be concerned with size and variability. Too small a sample will restrict the conclusions that can be drawn about all children, and too large a sample may be unwieldy. The two

samples, subjects and language, also interact, one influencing the other.

The number of children or subjects should be large enough to allow for individual differences and to enable group conclusions to be drawn. The overall design of the study will influence the number of subjects considered adequate. For example, it may be appropriate to follow a few children for a period of time, called a longitudinal study, but inappropriate to administer a one-time-only test to the same limited number of children. Other considerations will also influence the number of children studied. In a longitudinal study, for example, as many as 30% of the children may be lost because of family mobility, illness, or unwillingness to continue over a four or five year period. It might be better, therefore, to adopt an overlapping longitudinal design with two different age samples, each being observed for half the length of time that would have been needed in a longitudinal study.

The sample of children should accurately reflect the diversity of the larger population from which they were drawn. In other words, the children sampled should represent all socioeconomic, racial and ethnic, and dialectal variations found in the total population, and in the same proportions found there. Other variables that may be important include size of family, birth order, presence of one or both parents in the home, presence of natural parents in the home, and amount of schooling. Some variables, such as socioeconomic status, may be difficult to determine, although parental education and employment seem to be important contributing factors. Mixed-race children may force the researcher to make decisions about racial self-identity that are not appropriate. Other variables, such as birth order, may be more important than more traditional variables, such as gender.

Characteristics of the tester, experimenter, or conversational partner are also important. In general, preschool children will perform better with a familiar adult. There is also some

indication that children of color may perform better with adults with the same identity.

Some children found in the general population may be excluded when the study attempts to determine typical development. These may include children with known handicaps; bilingual children; twins, triplets, and other multiple births; and children in institutional care or full-time

nursery school. Children may also be excluded who are likely to move during the course of the

study or whose parents were deemed uncooperative or unreliable. For example, children with parents in the military are likely to move frequently, possibly prior to the completion of a longitudinal study. With each exclusion, the “normal” group becomes more restricted and, thus, less representative.

Any sample should fulfill the twin requirements of naturalness and representativeness. Even

testing should attempt to use familiar situations with a child in an attempt to meet these two

requirements. A conversational sample will be more natural if the participants are free to move

about and are uninhibited by the process of sample collection. A representative sample should

include as manyof the child’s everyday experiences as possible. Unfortunately, little is known

about the range and frequency of children’s activities. To make it clear we will look at the example.

Wells (1985) sampled randomly throughout the day for short periods.

Each day of collection, Wells collected 24 randomly scheduled samples of 90 seconds’ duration each. Samples were scheduled so that four occurred within each of six equal time periods throughout the day. Eighteen of the 24 samples, totaling 27 minutes of recording, were needed for analysis. This allowed a possible 25% of the samples to be blank as a result of having been recorded while the child was beyond the range of the microphone. Two samples from each of the six time blocks were randomly chosen for transcription. After these had been transcribed, the process continued randomly with the remaining six samples until 120 intelligible utterances had been amassed. The remaining utterances were not transcribed for analysis. This procedure was followed once every three months for two years for each child.

As we can see, it is not always easy to obtain natural and representative language data. At least three potential factors may be problems. One problem is the observer paradox. Stated briefly, the absence of an observer may result in uninterpretable data, but the presence of an observer may influence the language obtained, so that it lacks spontaneity and naturalness.

The presence of an observer can also affect the type of sample collected. The behavior of the child and the conversational partner may be influenced by the presence of an other person.

For this reason, Wells (1985) collecteds amples on a tape recorder, with no observer present. The recorder was programmed to begin taping at randomly assigned times throughout the day.

The absence of an observer may also complicate the process of determining the exact context of the language sample. At the end of each recording day, parents might be asked to identify contexts by the activity and participants present, although the reliability of such recalled information is doubtful. In addition, the immediate nonlinguistic context of each utterance cannot be reconstructed from audiotape alone. Digital audio and video recording may address this concern.

A second problem is a child’s physical and emotional state at the time the information is collected. Usually, a child’s caregiver is asked to comment on the typicalness of the child’s performance.

A third problem relates to the context in which the sample is collected. Quantitative values such as mean or average length of utterances (MLU) or the number of utterances within a given time, or the number of root words vary widely across different communication situations and partners. For example, a play situation between a mother and child elicits more language than one in which a child plays alone. Productivity, or the amount of language, may be even more affected by a child’s conversational partners than by different situations.

Questions relative to collection of the language sample must of necessity concern the presence

or absence of a researcher and the actual recording method. Wells (1985) attempted to mini-mize observer influences by having the child wear a microphone that transmitted to a tape recorder preprogrammed to record at frequent but irregular intervals throughout the day.

Of course, there are problems with this process, such as the compactness and sensitivity of the

microphone transmitter. In contrast,Brown (1973) used two researchers in the setting, while

data were recorded on a tape recorder. This concern is somewhat addressed by the compactness of digital recording devices.

Several collection techniques exist, such as diary accounts, checklists, and parental reports,

as well as direct and digitally recorded observation. The first three are alternatives to researcher

observation and have been used effectively in the study of early semantic and morphologic growth. Such methods enable researchers to collect from more children because they are less time consuming and have been pronounced reliable and valid while remaining highly representative.

Electronic means of collection seem essential for microanalysis. Videorecording, while more intrusive, is better than audio alone, because it enables the researcher to observe the nonlinguistic elements of the situation in addition to the linguistic elements. Althoug huseful in some collecting, written transcription within the collection setting is the least desirable method for microanalysis. First, it is easy to miss short utterances. Second,it is nearly impossible to transcribe the language of both the child and the conversational partner because of the large number of utterances with in a short period of time. Third, transcription within the conversational setting does not enable the researcher to return to achild’s speech for missed or misinterpreted utterances.

The language sample should be transcribed from the recording as soon after it is collected as possible. Caregivers familiar with a child’s language should be consulted to determine if the sample is typical of the child’s performance.

Because transcription offers many opportunities for error, studies should ensure intra-transcriber reliability. This is not always easy to accomplish. Several factors contribute to transcription errors, including the type of speech sampled, the intelligibility of the child, the number of transcribers, the level of transcription comparison, andt he experience of the transcriber. In general,the more defined the speech sampled, the better the intelligibility; the greater the number of transcribers, the larger the unit of comparison; and the more experienced the transcriber, the better the chance of having an accurate transcript.

Actual analysis may be ticklish, especially when trying to determine the bases for that analysis.

For example, MLU is still the most common quantitative measure of language growth, although its value is questionable. In general, quantitative measures, such as numerical scores and MLU, are inadequate for describing language development in detail. Other quantitative values might include total number of words, number of words per clause, or clauses per sentence. Such values collapse data to a single figure. Breadth of behavior might be obtained by the number of different forms used by a child, such as number of different words and number of unique syntactic types.

In contrast, qualitative research uses a variety of methods within natural situations or contexts to describe and interpret human communication. Given the interwoven character of communication and social interaction, it seems logical to study the two together. As a result, language is studied as a social tool used within the complex relationship of context and communication. Thus, qualitative research is holistic and emphasizes communication’s synergistic nature.

By their nature, qualitative research methods change the units being studied. A single word or utterance cannot be analyzed as a separate entity but must be examined in the context of surrounding utterances, topics, or conversation or between partners.

It is also difficult to determine when a child or group of children actually knows or has mastered a language feature. The criteria for establishing that a child knows a word or a feature have not been preestablished. For example, with word knowledge, the researcher must have clear evidence that a child comprehends the word. In contrast, production criteria would probably be based on spontaneous use and consistent semantic intent. With young children, a researcher would also note consistent phonetic form and semantic intent, with decisions of knowledge not necessarily based on whether the form and meaning are related to an adult word.

Usually, mastery can be based on children using a feature in 90% of the obligatory locations or on 90% of the children using the feature consistently, but these percentages vary with individual researchers. Some researchers consider the average age for acquisition to be that point at which 50% of children use a language feature consistently. Of course, such measures are complicated by the complex nature of most language features and the extended period of time often needed for mastery. For example, forms such as correct use of be may take several years from first appearance to full, mature use.

Within the last half century, linguists have proposed several theories of language acquisition. Over that time, many linguists did not adhere strictly to one theoretical construct but preferred to position themselves somewhere between. This apparent fence straddling reflects the complexity of language and language acquisition.

Complex topics such as language and language development require a great amount of

study and research. If the data that result from such research are to be of value beyond the children from whom they were collected, researchers must consider a great variety of questions relative to the language features studied, the children selected, the amount of data, and the collection and analysis procedures. Describing child language development accurately is a difficult and time consuming job.

References

Lightbown, P. M., & N. Spada. (2006). How languages are learned (3rd ed.). China: Oxford University Press.

Robert, E. & Owens, Jr. (2011). Language development: an introduction (8^th ed.). Boston, MA: Pearson

Wells, G. (1985). Language development in the pre-schoolyears. New York: Cambridge University

Press