The Effect of Adjective Order on Response Latency in Audiovisual Delayed Match-to-Sample

More Writings

Jules Ochoa, Katherine Drummond, Kyle Roundtree, & Evan Schleifer-Katz

University of North Texas

The study of language has been an ongoing process to explain complex, everyday behavior in a scientific way. While several scientific communities have sought explanations for various parts of language, some have taken interest in the use of adjectives. Adjective orders, in particular, have varied theories about their selection and use to the present day, with disagreements occurring about what the proper placement is for entire categories of adjectives (“Adjectives: Order,” n.d.; “Order of Adjectives,” n.d.; Voa, 2019). Even within similar disciplines, little agreement has been found to systematically explain the mechanism behind word order, even in shortened phrases. Several fields have offered their own ideas of how verbal communities choose the word orders that exist within them, including linguistics and behavior analysis.

Linguists have discussed the possibility of cyclical scanning procedures (Martin, 1970), pragmatic communication (Danks & Schwenk, 1972), and even innate ability (Chomsky, 1959). Cyclical scanning procedures involve placing the most easily observed and directly modified adjective closest to the noun (Martin, 1970). Any adjectives added in front of the adjective directly modifying the noun modifies the entire phrases that follow, including the following adjectives. Pragmatic communication offers a different interpretation, where word order is intentionally switched by the speaker to aid in the listener’s discrimination between objects (Danks & Schwenk, 1972). When combined with a possible change in intonation, the changed word order is used as a functional tool to aid the listener in making a faster, correct selection as the differentiating adjective category is moved to the beginning of the phrase regardless of whether it is supposed to be there or not, according to conventional rules. Chomsky (1959) offered yet another explanation by stating that language and its accompanying rules may be acquired when a person has reached an age of maturation where they are innately able to understand such concepts. He differed from the previous literature in offering the idea that it was not so much the behavior of a speaker on a listener that mattered, but having the biological and genetic capability to understand the language one is exposed to (Chomsky, 1959). While all authors came with a linguistic point of view, their approaches varied widely.

There were some agreements to be found in the underpinnings of the linguistic theories despite these differences in the overall concept. Similar patterns were noted by various authors in adjective placement and in the subjectivity of some categories of adjective classes. Several mentioned that the more absolute and intrinsic an adjective was, two descriptors that tended to highly correlate, the closer it tended to be placed to the noun (Belke, 2006; Danks & Schwenk, 1972; Martin, 1970). On the other hand, adjectives that were more abstract tended to be placed further from the noun. Martin (1970), as well as Danks & Schwenk (1972), acknowledged that there was a conventional ordering of adjectives within the English language, although neither stated a source for which they used to determine what the proper order was. As there can be a varying amount of adjective categories depending on the source used as a reference, and no definitive guide to refer to on the matter, this makes the question of what is normal versus abnormal a hard one to answer. None of the listed sources, either linguistic, behavior analytic, or otherwise, could give an authoritative and final answer on the matter. Behavior analytic literature, like the others, has chosen to focus on other areas of verbal behavior for areas of research and leave the more subjective categories for a later time and field.

Behavior analysis has attempted over the years to decode verbal behavior and understand its function for verbal communities. B.F. Skinner (1953) recognized the complexity in identifying what part of spoken verbal behavior controls the behavior of the listener, as verbal behavior often shares common elements. In contradiction to Chomsky, he stated that verbal behavior is shaped by the community in which a speaker is placed and has their spoken behavior reinforced (Skinner, 1953). Without the feedback and consequences from the community, the speaker would learn a very different pattern of verbal behavior. While not addressing adjectives or word order directly, Skinner’s argument lends to the present discussion by offering some insight into how a specific word pattern may be learned by one group as “right” while another group may consider the same word pattern “wrong.” A third group may even find both word patterns acceptable depending on the circumstance in which either word order is used.

Due to the extensive interest taken in word placement, past research indicates that there may be an effect of word placement on listener behavior (Belke, 2006; Dank & Schwenk, 1972; Martin, 1970). The outcomes of the studies conducted provided conflicting results, suggesting further research needs to be done to provide more insight into how word order may affect the response of a listener. Adjectives are features of language that are frequently used to a varying degree in everyday language and might be a less noticeable change to the participants than changing noun or verb placement within a sentence. As adjective order is already possibly subjective, it is an interesting area to examine whether the order of adjectives affects responding or if listeners are responding to another feature not intentionally controlled. The present study compared response latency in selecting a picture after being presented with adjective descriptions in conventional and unconventional orders, according to standard language practices in the English language. This was determined by consulting various resources for teaching English grammar (“Adjectives: Order,” n.d.; “Order of Adjectives,” n.d.; Voa, 2019).

Method

Participants

Four individuals, aged 26 to 28, participated in the experiment. Two (one male, one female) were enrolled in the graduate program in a behavior analysis department in Denton, Texas, one (female) was enrolled in the same department’s doctoral program, and the fourth (male) was admitted to the program for attendance later in the year. All four subjects were fluent in English, and one of the four was also fluent in Arabic. Participants were recruited via a convenience sample of availability on a weekday afternoon and acquaintance with the student experimenters. All individuals agreed to participate.

Materials

Sessions were conducted on a Windows 7 Dell-PC computer in a 2 m x 3 m room in a university building in Denton, Texas. The room was empty besides a table, chair, computer tower, display, keyboard, mouse, window, and headphones. The investigation was run using an experimenter-designed program on the Paradigm Experiment for Desktop software, which appeared on the display for participants. The software used built-in trial and stimulus selection randomizers, and automatically detected, collected, and transduced data from participant responding to Microsoft Excel files.

Experimental phases (Pre-test, Testing, Order discrimination post-assessment, Post-test survey) were displayed sequentially on a dark green background (RGB: 142, 188, 139) with other stimuli varying as programmed by experimenters. Instructions preceding and following experimental phases were presented in text boxes on a gray background (RGB: 119, 136, 152). Equidistant, horizontally-distributed gray squares (RGB: 166, 166, 166) of the same size (250 px x 250 px; left, center, and right position) were displayed during match-to-sample delays (i.e., during auditory stimulus presentation). Sample stimuli, described below, were presented in the same horizontally-distributed, equidistant location (i.e., left, center, or right). Two separate task performance contingent feedback screens were used, each with a preset text box against a distinct background: simple positive feedback (“You got it!”) over a light-green background (RGB: 144, 238, 144), or simple corrective feedback (“That’s not it.”) over a light-red background (RGB: 206, 92, 92).

Primary visual experimental stimuli were images of common objects (e.g., pencils, hats, cars) prominently isolated against neutral, contrasting backgrounds (e.g., white, gray). Stimuli for primary testing were selected in 10 sets of three images (30 total), each set containing the same object type with multiple varying features (see Table 5).  Pre-testing stimuli were a separate set of fruit: orange, apple, and grapefruit. All images were .JPG files equally sized at 250 px x 250 px.

Primary auditory stimuli were 60 spoken phrases recorded by the first author on a Zoom H2n Handy Recorder microphone (see Table 5 for phrases). Adjective adjustment in scrambled auditory samples was counterbalanced through re-positioning adjectives from their typical first, second, third, or fourth slots (e.g., A1-laced, A2-inflated, A3-colored, A4-leather ball) into scrambled permutations with 75% or greater difference from the original order (e.g., A4-leather, A2-inflated, A1-laced, A3-colored ball). Efforts were made to hold two adjectives constant across stimulus sets of the same type (e.g, “big”, “thick” for all books), while two were left to vary (e.g., old, leather, brown, animated, etc.); eight of ten stimulus sets successfully met this program design. Pre-testing auditory stimuli were preset recording files of spoken words (orange, apple, and grapefruit) that come with the Paradigm software. All auditory stimuli were .WAV files between 2.0 s and 3.1 s duration.

Measures

Dependent variables in the testing phase were response latency (ms) and percent correct performance per set. Response latency was defined as the duration between the presentation of the sample array (simultaneous with the termination of the auditory stimulus) and the participant’s subsequent selection of any sample stimulus (whether correct or incorrect). Percent correct performance was defined by the total number of correct responses in N or F-set divided by the 30 total trials and multiplied by 100 (i.e., (#-correct/total)*100). Secondary measures included response latency (ms) and percent correct performance during the pre-test, and percent correct performance during the order discrimination post-assessment.

Independent Variable

The primary variable manipulated by the experimenters was the order of adjectives spoken within the auditory stimulus during the testing phase. Half of the stimuli presented were spoken with adjective order typical of the English language (N-set), whereas the other half were spoken in a systematically scrambled order (F-set). Efforts were made by experimenters to hold all other experimental variables constant, counterbalanced, and/or randomized (e.g., trial-order, total number of trials for each type of order, number of trials for each type of order for a specific stimulus, position of correct sample in array, number of adjectives held constant across stimuli sets, within-F-set adjective-scrambling order, etc.) in order to isolate the effects of adjective order on dependent variables measured.

Experimental Design

            Following pre-testing to familiarize participants with the apparatus and programmed delayed match-to-sample procedures to criterion-based levels of success, the testing phase used a multielement experimental design to compare the effects of typical versus scrambled adjective order on participant’s response latencies and correct responding. The multielement experimental design employed rapid and randomized alternation between conditions across 60 trials. The rapid alternation between conditions aimed to minimize potential threats to internal validity, such as sequence or testing effects from presenting several trials in a row of one condition prior to introducing the other condition (Barlow & Hayes, 1979). In a multielement design, robust experimental control is demonstrated by the differentiation between test and control conditions.

Procedure

Entrance and instructions. Prior to entering the room and agreeing to participate, participants were told the experiment would take approximately 10 minutes. After participants entered the room and sat down, the experimenter exited and closed the door, and participants interacted with the program by using the keyboard and mouse. Prior to participant interaction, a procedural overview and instructions on how to interact with the software were delivered by the program as follows: “Welcome to the Experiment! Please put on your headphones and turn up the volume. There will be four sections. You will use the mouse for each. Press Space for Section 1.” Participants pressed Space to begin.

Pre-test. Participants were familiarized with the apparatus and procedural stimuli through simple, consecutive audiovisual delayed match-to-sample trials. First, three (non-clickable) gray squares simultaneously appeared, equidistant and horizontally distributed (i.e., left, center, and right squares), while a brief spoken auditory stimulus played (“orange”, “apple”, or “grapefruit”). Following termination of the auditory stimulus, the set of three clickable visual stimuli (orange, apple, and grapefruit) appeared in the same size and location as the gray squares. Correct or incorrect feedback via a feedback screen was provided following participant responding and was followed by an inter-trial interval of 500ms. Trial order and stimulus positions (i.e., left vs. center vs. right) were randomized and counterbalanced across trials. A mastery criterion of nine consecutive correct responses was established by the experimenters to ensure participant readiness for the subsequent testing phase.

Testing. Minimal instructions were provided prior to testing: “You have completed Section 1. Press Space for Section 2.” Participants pressed Space to begin. Audiovisual delayed match-to-sample trial design and within-trial order of stimulus presentation were identical to pre-testing: three gray squares and audio, the appearance of three visual samples with one correct match, feedback, inter-trial interval. All participants were exposed to 60 consecutive trials, each of which began with an auditory stimulus of four adjectives, in typical (N-set) or scrambled (F-set) order, followed by a noun. The set of three visual samples corresponding with the preceding audio’s noun (e.g., three books, three pencils) appeared following the audio (see Table 5 for list). Each of the 30 visual samples was designated as the correct match exactly twice throughout the testing trial-block: once with N-set audio and once with F-set audio. Visual sample array positions were pre-counterbalanced, and trial order (i.e., types, samples, N vs. F-set, etc.) was randomized for all 60 trials. Visual feedback was provided following responding as in pre-testing, but correct and incorrect responding did not affect programmed conditions; each auditory stimulus was presented once, and the phase ended following 60 trials (i.e., no mastery criterion).

Order discrimination post-assessment. Minimal instructions were provided prior to the 30-trial assessment: “You have completed Section 2. You will not receive feedback on the next sections. Press Space for Section 3.” Participants pressed Space to begin and were immediately presented with three vertically-distributed, horizontally-centered text boxes. The top text box, which remained unchanged and unclickable throughout the assessment, stayed positioned slightly towards the top of the screen with the question “Which sentence looks correct?” in 27pt font. The second and third text boxes were clickable, shorter in length, used 14.25pt font, and were positioned slightly towards the bottom of the screen, one above the other. Each trial textually presented an N-set phrase (correct) in one clickable box, and the corresponding F-set phrase (incorrect) in the other box (see Table 5). The position of N-set and F-set text (middle or bottom clickable text box) was pre-counterbalanced for each textual stimulus set, and the trial order was randomized across all 30 trials. Responses were followed by a brief inter-trial interval of 250ms. No feedback was provided following correct or incorrect responses. The phase ended following 30 trials, one for each textual stimulus set.

Post-test survey. Minimal instructions were provided prior to the three-question survey: “You have completed Section 3. The last section is a quick survey. Press Space for Section 4.” Participants pressed Space to begin. The first item of the survey asked, “Is English your first/native language?”. The second item asked, “Do you regularly speak any other languages besides English?” Subjects either clicked a ‘Yes’ or ‘No’ button for both items to advance to the next item. The third item prompted: “If you regularly speak any other languages, please list below:”. A textbox was provided below for open-ended responses using the keyboard, with a “Press Enter to Submit” button below the textbox. After participants pressed Enter, the final instruction screen stated “You’re all done! Thanks for participating! :)”.

Results

A table and figure were created for each participant. Each table displays the average, minimum, maximum, and standard deviation for the latency to respond in milliseconds to the presentation of the visual stimuli for the Pre-test, N-set, and F-set of the indicated participant. All values in the tables were rounded to the nearest tens place. Exact data values were used in the figures.

During the pre-test, all participants answered match-to-sample audiovisual discriminations with criteria being 9 correct in a row in order to move onto the presentation of the randomized N-set and F-set trials. Each participant correctly answered 9 match-to-sample audiovisual discriminations without any extra trials. In the testing phase, all of the latencies to respond were closely grouped within and across participants (Fig. 1-4). The average response latency for participants G, E, and A were all slower in response to the N-set as compared to the F-set. Participant R was the only one that showed slower responding to the F-set (Table 1-4). When measuring latency to respond the first time the stimuli were presented, all participants responded with longer latencies when the N-set version of the stimuli was presented first (Table 5).

Each Figure represents latency to respond in milliseconds to the presentation of the visual stimuli on the ordinate and represents the trial number on the abscissa. N-set data points are presented in closed circles and F-set data points are presented in open squares. The markers filled with red indicate trials the participant errored on by selecting the wrong stimuli described by the audio sample. Participants G, A, and R all had 3 errors while E had 2 errors. Across participants, only two errors ever occurred on the same stimulus and there were no patterns in regard to the latency with which participants took to respond to errored trials. All participants show undifferentiated latency to respond between the N-set and F-set (Fig. 1-4). During the order discrimination post-assessment, all participants scored between 63-74% correct in identifying which written sentence had the correct adjective order (data not shown).

Discussion

Based on the results of this study there is no conclusive effect on latency to respond with “correct” or “incorrect” placement of adjectives. No participant responded in the way expected - differentiated responding to N-set and F-set with latency to respond being higher for F-set at the beginning of the test phase. This may imply that there are variables within the present study that may contribute to a longer N-set latency. This can be due to a number of contributing factors including the number of adjectives used and the placement of adjectives in the respective sentences. Based on the order discrimination post-assessment at the end of the procedure, all participants scored over sixty percent correct on presumed correct sentence structure, which suggests there may be agreement on proper adjective placement within the participants in this study. Another potential factor as to why there might have been undifferentiated responding could be the multiple exposures to the randomized correct and incorrect sentence placements. However, there is uncertainty as to why the participants may have produced these scores with this information alone. Since the presentation of sentences during the testing phase were brief audio recordings, there was limited exposure to sentence structures compared to the questionnaire which may play a role in correct and incorrect responding.

Additional contributing factors to the overall results may have been the number of adjectives used in the sentences presented. To test this theory, the subtraction of one or more adjectives may allow for clearer discrimination of the preferred adjective placement in a population's verbal behavior. In addition, the modification of stimuli presented after correct/incorrect statements could be changed where all stimuli in a series share a discernable adjective such as color or shape. Limitations within the present study may include the type of measurement used, the selection of adjectives and visual stimuli used to represent them, and the number of adjectives used during the presentation.

Conclusions and Future Research

There are variations of the current study that could resolve the previously mentioned limitations. This includes setting a limited hold (i.e., capping time) for responses, systematically selecting and placing adjectives, and limiting the number of adjectives included in the presentations. Setting a limited hold for responses would introduce more errors in the accuracy of response for participants, but it would allow for closer analysis of whether the order of adjectives presented did improve accuracy given a restricted time to respond, or whether the order has no effect. Restricting the adjective categories to size, shape, age, and material along with restricting the number of adjectives used may more closely mimic how adjectives are used in real scenarios as discriminations between objects. These restrictions to more commonly used types and numbers of adjectives may make it easier to capture an effect of disrupting the placement of adjectives from a conventional to an unconventional order if there truly is one. Being more systematic in the placement of specific categories of adjectives, such as moving material to the front of the adjective phrase and holding all others constant, would be another avenue to consider to see if there is an effect of word order on responding. Danks and Schwenk (1972) offered that intonation may contribute to the selection of adjective order depending on if one or both adjectives were required to make a discrimination between two objects. This suggests that which word order a participant may respond faster to in a given condition may fall under a complex learning history due to conditional discriminations the experimenter may not be aware of, and there may be further areas of research yet to be explored.

 Figures & Tables

References 

Adjectives: Order. (n.d.). Retrieved March 20, 2019, from https://dictionary.cambridge.org/us/grammar/british-grammar/about-adjectives-and-adverbs/adjectives-order

Barlow, D. H., & Hayes, S. C. (1979). Alternating treatments design: One strategy for comparing the effects of two treatments in a single subject. Journal of applied behavior analysis, 12(2), 199-210.

Belke, E. (2006). Visual determinants of preferred adjective order. Visual Cognition,14(3), 261-294. doi:10.1080/13506280500260484

Chomsky, N. (1959). Reviews: Verbal Behavior by B.F. Skinner. Language,35(1), 26-58. doi:10.2307/411334

Danks, J. H., & Schwenk, M. A. (1972). Prenominal adjective order and communication context. Journal of Verbal Learning and Verbal Behavior,11(2), 183-187. doi:10.1016/s0022-5371(72)80075-6

Martin, J. (1970). Adjective order and juncture. Journal of Verbal Learning and Verbal Behavior,9(4), 379-383. doi:10.1016/s0022-5371(70)80076-7

Order of Adjectives. (n.d.). Retrieved March 20, 2019, from https://www.perfect-english-grammar.com/order-of-adjectives.html

Skinner, B. F. (1953). Science and human behavior. New York: Free Press.

Voa. (2019, February 07). What Is the Word Order of Adjectives? Retrieved from https://learningenglish.voanews.com/a/what-is-the-word-order-of-adjectives-/4775294.html