Retrieval-Based Word Learning in Developmental Language Disorder

Overview

Children with developmental language disorder (DLD; also referred to as specific language impairment) experience a significant deficit in language ability that is longstanding and harmful to the children's academic, social, and eventual economic wellbeing. Word learning is one of the principal weaknesses in these children. This project focuses on the word learning abilities of four- and five-year-old children with DLD. The goal of the project is to build on our previous work to determine whether, as we have found thus far, special benefits accrue when these children must frequently recall newly introduced words during the course of learning. In this first of a series of studies, we seek to increase the children's absolute levels of learning while maintaining the advantage that repeated retrieval holds over comparison methods of learning.

Although much of the DLD literature has emphasized morphosyntax, those children with morphosyntactic weaknesses also have major deficits in vocabulary (McGregor et al., 2012), and sophisticated statistical approaches applied to large databases find no basis for treating lexical and morphosyntactic skill as separate dimensions (Tomblin et al., 2014). Longitudinal studies show that individuals with DLD fall further behind their peers in vocabulary ability from preschool to 21 years of age (Rice \& Hoffman, 2015). Vocabulary scores for adults with a history of DLD are significantly lower than those of other adults, with most scores falling within a clinical range (Tomblin et al., 1992). Many studies have shown that the vocabularies of children with DLD have less breadth and depth than those of their same-age peers (Kail \& Leonard, 1986; Krzemien et al., 2020; McGregor et al., 2002; McGregor, Oleson et al., 2013; McGregor \& Waxman, 1998). To examine the dynamic aspects of word learning in these children, investigators have often presented the children with novel words to be learned, carefully controlling for the frequency and type of input provided. These studies have shown that children with DLD require more encounters with a word before learning takes place, with weaknesses evident in both semantic and phonological encoding (Alt, 2011; Alt \& Plante, 2006; Alt et al., 2004; Gray, 2003, 2004; Gray et al., 2014; Leonard et al., 1982; McGregor, Licandro et al., 2013; Rice et al., 1994). Novel word learning weaknesses have been documented for nouns, verbs, and adjectives (Kan \& Windsor, 2010; Oetting et al., 1995; Skipp et al., 2002; Rice et al., 1990; Windfuhr et al., 2002). In our previous work, we added retrieval practice to the standard novel word learning protocol. In previous studies of word learning by children with DLD, retrieval was used only as an assessment tool - to determine when or if a child had met a particular criterion in learning a set of new words. Based on our understanding of the growing literature on retrieval effects in the field of cognitive psychology, we introduced retrieval practice as a word learning facilitator, an application that has both theoretical and clinical implications. We tested the hypothesis that the act of retrieving new words significantly assists the learning of these words by children with DLD by markedly increasing long-term retention relative to levels seen through other types of word learning, even when the total exposure to the words is the same or less. Based on very encouraging findings, we continue our research with the goal of moving the use of retrieval closer toward becoming a procedure that can be incorporated into real-word clinical and educational settings. The rationale behind our work is grounded in findings from the memory literature that, following an initial study period, repeated practice in retrieving the material results in greater long-term retention than additional study of that material (Karpicke, 2012; Karpicke \& Blunt, 2011; Karpicke \& Roediger, 2007b, 2008; Roediger \& Karpicke, 2006; see meta-analysis in Rowland, 2014). The benefits are dramatic; in these studies, long-term retention is often 50% to 150% greater in the repeated retrieval condition. This effect has been seen for a wide range of material, from simple paired associates (e.g., knee-cloud) to fictional stories. The benefits of repeated retrieval can be further enhanced when there is spacing between retrieval attempts. Spacing can be defined in terms of time but also in terms of intervening material. Ideal spacing is one which is "effortful" but short enough to prevent forgetting. Repeated retrieval that occurs immediately after study can be helpful but does not yield the long-term recall seen for repeated spaced retrieval (Karpicke \& Roediger, 2007a). Our application of repeated spaced retrieval to the word learning of children with DLD has produced a rather clear set of results. The general method has been the same across experiments. Children with DLD age 4;0 to 5;11 and their same-age peers with typical language development (TD) learned sets of novel words (e.g., /nɛp/, /zogi/, /kudɪp/) in two sessions held on consecutive days and were then tested immediately after the learning period and again one week later. A within-participant design was used where each child learned half the novel words in a repeated spaced retrieval (RSR) condition and the other half in a comparison learning condition. Both conditions included "study" trials, where the child saw the picture and heard both the word form (e.g., /nɛp/) and, for studies employing words representing nouns, the word's "meaning" (e.g., A /nɛp/ likes rain). For conditions that included retrieval trials, the child saw the picture and was asked, for example, "What's this called? What do we call this?" and then "And what does this one like? What does it like?" For the RSR conditions in these studies, spacing was created by having two or three intervening words (depending on the experiment) between a word's previous study trial and its retrieval trial. Testing after the learning phase involved word form recall, meaning recall, and recognition (selecting the correct picture from an array, e.g., "Where's the nep?"). Our main findings were that: (1) children learned more novel words representing nouns (names of exotic plants and animals) in a RSR condition than in a repeated study condition that provided the same amount of exposure to the words; (2) children learned more novel nouns in a RSR condition than in an immediate retrieval condition that provided not only the same amount of exposure but also the same number of retrieval opportunities; (3) RSR produced better learning even when exposure was reduced relative to the comparison condition; (4) RSR was similarly advantageous when children learned novel words referring to adjectives that required the children to generalize to new objects showing the same attribute; (5) RSR advantages were seen not only in behavioral tasks but also at the neural level as captured through event-related potentials (ERPs); (6) the word learning gap between children with DLD and their TD age mates was smaller for words in the RSR condition than for words in the comparison conditions; and (7) the recall of the children with DLD from immediate testing to one-week testing was very stable and indistinguishable from that of their TD peers. Despite these encouraging findings, much remains to be learned about RSR and its effects on the learning of children with DLD and how this learning compares to that of children who are developing language on schedule. One important question is the absolute level of learning achieved through RSR. In our studies to date, in the more optimal condition of RSR, children with DLD have shown high accuracy levels on meaning recall (e.g., recalling that a /pobɪk/ likes birds) and recognition (e.g., pointing to the picture of the /pobɪk/ out of an array of four pictures). However, for word form recall (recalling the phonetic form of the word /pobɪk/), the children with DLD have recalled only about 55% of the novel words on average - this despite recalling twice as many word forms as in the comparison conditions. If the children were successful in recalling the word immediately after the learning period, they also recalled the word one week later. The main obstacle appeared to be initial word form encoding (Haebig, Leonard et al. 2019). In the current study, we seek to improve word form recall by the children through modifications that may improve encoding in particular. The principal means by which we hope to improve encoding and, as a result, subsequent recall, is to create a more gradual spacing of retrieval as the learning period proceeds. In our most frequently used "standard" procedure (Leonard, Karpicke, et al., 2019), the early trials involved the children being asked to recall the word form and meaning immediately after the study trial for the same word. For subsequent recall trials, three other words intervened from the last time the child heard the word and when the child was asked to recall it. Subsequent retrieval trials involved the same degree of spacing. In the current study, we will compare this standard procedure with an expanding retrieval schedule (e.g., Karpicke \& Roediger, 2007a; Landauer \& Bjork, 1978). As in the standard procedure, the first retrieval trial is an immediate retrieval trial. Then, the spacing between study and subsequent retrieval of the word is initially narrow -- one intervening word - and expands more gradually to three intervening words. The words in the two learning conditions - standard retrieval and expanding retrieval - will be presented in separate sets. The sets will be counterbalanced in order. Each set will involve two learning sessions held on consecutive days with a recall test administered five minutes after the second learning session, followed by another recall test one week later. The second set will begin the next week, following the same scheme as the first set. Four novel words will be learned in each set. The words will match those used in our previous studies on novel nouns (e.g., /nɛp/, referring to an unusual animal; /bog/ referring to an unusual plant). Colored photographs appearing on a laptop screen will be the visual referents for these words. The novel nouns serve as the word forms; we will also associate each word with something the referent "likes" (although what it likes will not be shown on the photograph. This will serve as the "meaning" of the word. During each study trial, the novel word will be mentioned three times and its meaning will be mentioned once, as in: "This is a /nɛp/. It's a /nɛp/. A /nɛp/ likes rain". For conditions that include retrieval trials, the child will see the picture and will be asked, "What's this called? What do we call this?" and then "And what does this one like? What does it like?" Audio stimuli will be pre-recorded and presented along with the visual stimuli (scanned photographs) on the laptop. The experimenter will manually move the program from one trial to the next to ensure that the child is attending to each trial. The novel words will consist of monosyllabic (consonant-vowel-consonant) shapes, thus conforming to the novel words used in our initial studies with the standard procedure. The novel words assigned to the two conditions will be matched according to phonotactic probability and neighborhood density (based on Storkel \& Hoover, 2010). The two learning sessions within each set will be identical. For the set constituting the standard condition, each novel word will be presented initially in a study trial, followed by an immediate retrieval trial, followed by another study trial. We refer to such retrieval trials as "0" retrieval trials, reflecting the fact that there are zero words intervening between a word's study trial and its retrieval trial. After each novel word has had a 0 trial, all subsequent retrieval trials will occur after three other words had intervened since the last time the word had appeared in a study trial. Such trials will be followed by a study trial for the same word. We refer to these retrieval trials as "3" retrieval trials, again, reflecting the number of intervening words used. The retrieval schedule for each day for the words in the standard condition will be 0-3-3-3-3. However, after the second 3 trial of the four words, we will insert one study trial for each word. This study trial is intended as a "refresher" under the assumption that some words will not yet be recalled at that point. The insertion of these study trials do not change the 3 spacing between a word's retrieval trial and its previous study trial. The next day will likewise have the 0-3-3-3-3 schedule with a study trial after the second 3 trial. For the set constituting the expanding retrieval condition, each word will first occur in a study trial, followed by an immediate retrieval trial, and then another study trial. The next two retrieval trials will occur after one other novel word had intervened since the word's most recent study trial. These retrieval trials will be followed by a study trial. We refer to these retrieval trials as "1" retrieval trials, as there was one intervening word. The two remaining retrieval trials for words in this condition will be 3 retrieval trials, with each retrieval trial followed by a study trial for the same word. The schedule for each day, then, will be 0-1-1-3-3. After the second 1 retrieval trial for the four words in the set, a study trial will be presented for each word as a "refresher," as was done for the words in the standard condition. With this design, the two learning conditions will provide the same number of exposures to the word forms and meanings, and the same number of retrieval opportunities throughout the two-day learning period. The two conditions will differ only in the retrieval schedule - 0-3-3-3-3 versus 0-1-1-3-3. For each set, after the child has completed the learning session of the second day, a five-minute break will be given. Then a recall test of each word form and meaning will be administered. Two items for each word form and meaning will be used, with the second item for each word presented only after all four words had been tested once. Prompts used for the recall items will be identical to those used in the retrieval trials. One week later, the recall test will be re-administered. This test will be followed by a recognition test. The photo of each novel word referent will be shown along with the referents of the other three referents (in an array of four photos) and the child will be asked, "Show me the e.g., /nɛp/." More details are provided in the Outcome Measures section. We hypothesize that the use of 1 retrieval trials prior to proceeding the 3 retrieval trials will increase the children's ability to encode the novel word forms and meanings. Accordingly, we expect that recall and recognition scores will be higher for words in the expanding 0-1-1-3-3 condition than in the standard 0-3-3-3-3 condition. This is a rather stringent test given that our previous work has shown that the standard condition was more successful than comparison conditions that did not involve spacing. Nevertheless, it is necessary to attempt to improve on the standard condition given that it did not enable children to learn more than approximately half of the words employed. We expect that the expanding condition will lead to better results than the standard condition for both the children with DLD and the children with TD. In our earlier work, group differences were often seen favoring the TD children for the standard condition but the difference was smaller than the group differences seen for the comparison conditions. The current study involves two conditions that each involve spaced retrieval, with only the particular retrieval schedule distinguishing the two. We are hopeful that the expanding condition will show narrower differences between the two groups of children than the standard condition.