Semantic Structure in Vocabulary Knowledge Interacts With Lexical andSentence Processing in Infancy

Arielle BorovskyFlorida State University

Erica M. EllisCalifornia State University, Los Angeles

Julia L. EvansUniversity of Texas at Dallas

Jeffrey L. ElmanUniversity of California, San Diego

Although the size of a child’s vocabulary associates with language-processing skills, little is understoodregarding how this relation emerges. This investigation asks whether and how the structure of vocabularyknowledge affects language processing in English-learning 24-month-old children (N = 32; 18 F, 14 M). Paren-tal vocabulary report was used to calculate semantic density in several early-acquired semantic categories.Performance on two language-processing tasks (lexical recognition and sentence processing) was compared asa function of semantic density. In both tasks, real-time comprehension was facilitated for higher density items,whereas lower density items experienced more interference. The findings indicate that language-processingskills develop heterogeneously and are influenced by the semantic network surrounding a known word.

There is wide variability in vocabulary growth acrossinfancy, and these early differences can have impor-tant consequences for later cognitive and academicoutcomes (Cunningham & Stanovich, 1997; Hart &Risley, 1995; Ramey & Ramey, 2004). Therefore, thefactors that mediate the growth of early vocabularyskills are a subject of intense study. We explore novelconnections between two factors—processing skilland conceptual development—which have beenindependently studied in their role in vocabularygrowth in infancy. Specifically, we ask whether andhow these two factors may be inter related by mea-suring the real-time recognition of words that vary insemantic density in 2-year-old children. Such a con-nection can extend theoretical and practical insightinto the processes that underlie early vocabularydevelopment and on the specific ways in whichvocabulary development can support overall lan-guage processing. There are strong theoretical,

empirical, and computational reasons to suspect thatlanguage-processing skills and conceptual develop-ment could be directly connected and dynamicallyinteractive across early language development. Webegin by reviewing how these conceptual develop-ment and language-processing skills are each inde-pendently related to vocabulary growth.

Explaining Vocabulary Growth: ConceptualDevelopment and Processing Skill

Gopnik and Meltzoff’s (1987) “specificity hypothe-sis” was an early proposal that linked conceptualdevelopment and vocabulary growth. This idea pos-ited a causal link between the onset of the vocabularyspurt and the understanding that objects can besemantically grouped into basic level categories. Theproposal indicated that shifting a child’s conceptualorganization could promote vocabulary growth byallowing the child to recognize semantic linksbetween novel and known objects. This hypothesiswas supported by numerous studies that linked suc-cess in exhaustive two-group sorting tasks with theonset of the vocabulary spurt in infancy (Gopnik &Choi, 1990; Gopnik, Choi, & Baumberger, 1996;Gopnik & Meltzoff, 1992; Mervis & Bertrand, 1995;Poulin-Dubois, Graham, & Sippola, 1995; but cf.

We are deeply grateful to families who participated in thisproject. We also wish to thank Kim Sweeney, Virginia Prestwoodand research assistants at the Center for Research in Languageand the Language and Cognitive Development Laboratory fortheir assistance in collecting and coding the data. Katie Von Hol-zen and Susan Bobb kindly shared several R scripts that weadapted for our analyses. This work was supported by grants toArielle Borovsky: F32 DC10106, R03 DC013638, Erica Ellis: F31DC012254 Julia Evans: R01 DC005650 and Jeffrey Elman: R01HD053136.

Correspondence concerning this article should be addressed toArielle Borovsky, Department of Psychology, Florida StateUniversity, 1107 W. Call Street, Tallahassee, FL 32306. Electronicmail may be sent to borovsky@psy.fsu.edu.

© 2016 The AuthorsChild Development © 2016 Society for Research in Child Development, Inc.All rights reserved. 0009-3920/2016/xxxx-xxxxDOI: 10.1111/cdev.12554

Child Development, xxxx 2016, Volume 00, Number 0, Pages 1–16

Gershkoff-Stowe, Thal, Smith, & Namy, 1997 andSchafer & Plunkett, 1998, for alternative viewpoints).Computational simulations subsequently supportedthis hypothesis by demonstrating that lexical acquisi-tion could be facilitated in cases where semanticcategory knowledge is well organized (Borovsky &Elman, 2006). Other work has indicated that infantstend to learn words in a “clustered” fashion, that is,by preferentially acquiring words that are similar inmeaning and sound to other known items, ratherthan learning unrelated words (Altvater-Mackensen& Mani, 2013; Hills, Maouene, Maouene, Sheya, &Smith, 2009; Steyvers & Tenenbaum, 2005). Impor-tantly, coherence in the semantic microstructure ofinfants’ early vocabularies is associated with vocabu-lary growth and word learning strategy usage (Beck-age, Smith, & Hills, 2011; Yurovsky, Bion, Smith, &Fernald, 2012). In sum, these findings suggest thatconceptual structure, as manifested in organizationof categorical links among words, develops inconcert with the growing lexicon.

Linguistic-processing skills and vocabularygrowth are also correlated in childhood. There ismuch work that finds concurrent associations withvocabulary and language-processing skills like rapidauditory sound processing (Benasich & Tallal, 2002),size of phonological working memory (Gathercole &Baddeley, 1989), and speed of lexical recognition(Fernald, Perfors, & Marchman, 2006). Early-proces-sing skills also predict later vocabulary outcomes.For instance, speed of lexical recognition at18 months predicts subsequent vocabulary size at24 months in typically developing and late-talkinginfants (Fernald & Marchman, 2012). Furthermore,lexical processing at 25 months predicts cognitiveand linguistic outcomes at age 8 (Marchman & Fer-nald, 2008). This early relation between vocabularyand linguistic processing scales up to more complexlanguage tasks, such as in anticipatory processing insimple sentences (Borovsky, Elman, & Fernald, 2012;Mani & Huettig, 2012). In sum, the early relationbetween vocabulary and linguistic processing is avital component in long-term development of lan-guage skills: Children who are faster to interpretauditory and linguistic information in speech alsomore quickly recognize and acquire novel lexicalitems when they are spoken, though the directional-ity of this relation is not yet clear.

Are Language-Processing Skills and SemanticDevelopment Inter-Related?

There has been little work that explores whetherand how real-time language comprehension and

conceptual development may be directly intercon-nected in infancy. Establishing this potential connec-tion is the major aim of the current research. Ifconceptual development and processing skills areindeed linked, then we also expect the microstructureof vocabulary to associate with differences in linguis-tic processing within the same individual. Thishypothesis leads to a testable prediction that the timecourse of recognition for a particular item shouldvary according to semantic network that surroundsthe item. More specifically, we would expect facili-tated processing for lexical items that have a rela-tively denser “semantic network” than those withfewer semantic neighbors. We test this prediction inthe current study by asking whether and howsemantic density directly influences processing skillsduring two language-processing tasks: lexical recog-nition and sentence processing (Altmann & Kamide,1999; Fernald, Pinto, Swingley, Weinberg, & McRo-berts, 1998).

The two selected language-processing tasks havebeen associated with vocabulary size in infants,such that higher overall vocabulary skill leads tofacilitated performance, as measured by speed andaccuracy in looking toward a target item (Fernaldet al., 2006; Mani & Huettig, 2012). Yet, it is notknown whether processing skill is tied directly tothe semantic network that surrounds a lexical itemor is simply driven by overall vocabulary size, assuggested by prior work. We describe assessmentof semantic density and the experimental tasks inthe following sections.

Defining Semantic Density

Although there are a number of reasonablepotential metrics of semantic density that have beendefined for the average adult lexicon (e.g., semanticconnectivity, semantic set size, number of features;Mirman & Magnuson, 2008), the tremendous differ-ences in early lexical knowledge between individualchildren present challenges for developmentalresearch. However, because vocabulary in infancyis small, it is possible to take a comprehensivesnapshot of the child’s productive vocabulary tomeasure semantic density within an individual child.

In the current study, we measure overall vocabu-lary with a common infant vocabulary instrument,the MacArthur-Bates Communicative DevelopmentInventories (MBCDI; Fenson, 2007). The items inthe Words and Sentences form of the MBCDI wereselected to reflect many words that typically appearin infant productive vocabularies (Nelson, 1973).Moreover, nouns in this inventory are organized

2 Borovsky, Ellis, Evans, and Elman

according to several early-acquired semanticdomains, such as animals, clothes, body parts, andvehicles. These aspects of the MBCDI, therefore,have the potential to yield meaningful measures ofsemantic density within early-acquired categories.For the current study, we operationalize semanticdensity for each lexical item as the proportion ofwords that child says within its correspondingsemantic category. Concretely, the semantic densityof dog would be defined as the proportion of allpossible animal words on the MBCDI that the childsays. We use items from six early-acquired semanticdomains including animals, body parts, clothes,drinks, fruits, and vehicles.

Using Eye Tracking to Measure Processing Skill inInfants

Measurements of gaze in response to spokenlanguage are often used as an index of language-pro-cessing skill in infancy (e.g., Fernald, Zangl, Portillo,& Marchman, 2008). This technique, variously ter-med the “looking while listening” or “visual worldparadigm,” measures changes in gaze in relation tothe unfolding of speech in real time (Huettig, Rom-mers, & Meyer, 2011). For example, Fernald et al.(1998) presented children with arrays containing adistractor and target image (e.g., car and shoe) whilesimultaneously labeling one image (Look at the shoe!)and found that the speed and accuracy to view thecorrect target item improved from 15 to 24 months.

Semantic Density and Processing Speed: Facilitation orInterference?

Prior findings (reviewed earlier) reveal that largervocabulary size boosts lexical and sentence recogni-tion (Fernald et al., 2006; Mani & Huettig, 2012). Byextension, these findings support the possibility thatsemantic density should facilitate lexical recogni-tion. Contrasting evidence from the phonologicaldensity literature indicates that dense networks mayinstead promote lexical interference. In this case,increasing semantic density may result in enhancedcompetition from semantically similar items in thechild’s lexicon, resulting in slower lexical recogni-tion in denser semantic networks. Such a patternwould be analogous to inhibitory effects on wordrecognition in young children and adults in wordsoccurring in dense phonological neighborhoods(Garlock, Walley, & Metsala, 2001; Mani & Plunkett,2011; Vitevitch, Luce, Pisoni, & Auer, 1999).

Although there is no direct evidence that youngchildren experience enhanced semantic interference

in dense networks, recognition of semantic andphonological links between words does change con-siderably over the first 2 years of life. Notably,semantic priming effects have been inconsistentlyobserved at 18 months and reliably reported by24 months in a variety of methods (Arias-Trejo &Plunkett, 2010; R€am€a, Sirri, & Serres, 2013; Stylesand Plunkett, 2009; Torkildsen, Syversen, Simonsen,Moen, & Lindgren, 2007; Willits, Wojcik, Seiden-berg, & Saffran, 2013). In contrast, although phono-logical priming appears to be present at 18 months(Mani & Plunkett, 2010), this pattern shifts fromfacilitation to interference by age 2 (Mani & Plun-kett, 2011). This shift may be due to the changingnature of vocabulary density and structure. Mayorand Plunkett (2014a) carried out simulations thatsuggest that inhibitory (phonological) connectionsbetween words increases as vocabulary size (anddensity) increases. It is not clear if the semanticstructure of the early lexicon similarly influencesmechanisms of lexical activation and recognition.

Therefore, we test 24-month-old children on twolanguage-processing tasks that are sensitive tovocabulary effects: (a) a lexical-recognition task and(b) a simple sentence-processing task consisting of asemantically informative verb followed by a relatedthematic object. As described earlier, children’sindividual knowledge with six early-acquireddomains is also assessed through parental report onthe words and sentences form of the MBCDI. Inboth studies, we posit that if semantic density facili-tates lexical and sentence processing, then wewould expect to see relatively more robust recogni-tion of higher (vs. lower) density target items.Alternatively, if semantic density instead createsgreater lexical competition, then we should expectto find the opposite pattern, with boosted recogni-tion for low- (vs. high-) density items.

Method

Participants

Thirty-two 24-month-old English-learning chil-dren (18 F, 14 M) from the metropolitan region sur-rounding a large city of southern California wererecruited via advertisements in the community. Allparticipants had previously participated in a sepa-rate experimental task and cognitive screeningusing the Bayley Scales of Infant Development, 3rded., Cognitive Subscale (BSID-III; Bayley, 2006) at18 months of age. All infants received a standardBSID-III score of 85 or above (M = 106.3, SD = 11.9,range = 85–130), meaning that no child fell more

Semantic Structure and Language Processing 3

than 1 SD below the mean in cognitive ability. Datacollection took place between November 2012 andSeptember 2013.

The demographic characteristics of our samplewere relatively diverse, with 39% reporting mem-bership in an ethnic minority group. Our familieswere highly educated. All mothers had completedat least high school, and 82% had graduated fromcollege. Additionally, 89% of children were livingin two-parent households. Infants were learningEnglish (hearing fewer than 20 hours per week ofanother language at home), and parents reportedthat their children had normal vision and hearingand no other concerns about their child’s languageor cognitive development. All participants had nor-mal birth histories and no recent or chronic earinfections.

Stimuli

Selection of Category Domains and Items

The MBCDI checklist is organized into severalcategory sections that correspond to early-acquiredvocabulary items. We consulted the Cross-Linguis-tic Lexical Norms database (CLEX; Dale & Fenson,1996) to select six category domains with wide rep-resentation in 24-month-old vocabularies, includinganimals, clothing, vehicles, body parts, and two subcat-egories from food, fruits and drinks. More informa-tion about the categories can be found in Table 1.

We used the CLEX database (http://www.cdi-clex.org) to select two highly known words in eachcategory (i.e., produced by at least 70% of Ameri-can 2-year-olds), with one word in each categoryassigned to the lexical-recognition task, and theother to the sentence-processing task. The items in

the lexical-recognition task were: animals: bird; fruit:banana; clothing: diaper; vehicles: airplane; drinks:juice; body parts: nose. The items for the sentence-processing task were: animals: dog; fruit: apple;clothing: shoe; vehicles: car; drink: milk; body parts:teeth.

Visual Stimuli

We selected a colorful, photorealistic 400 9 400pixel image for each experimental item centered ona white background. Several additional imageswere used either to direct the child’s attentiontoward the screen during the study (e.g., a smallflower, or happy face) or to maintain infant interest(e.g., a large image of the Sesame Street character,Elmo).

Auditory Stimuli

Speech stimuli for all items were recorded in aninfant-directed voice by a female native speaker ofCalifornia American English (E.E.) on a monochannel at 44,100 kHz sampling rate. To preciselycontrol the onset and offset of the auditory stimuli,the duration of all linguistic stimuli in both taskswas calculated in Praat (Boersma & Weenink,2012).

In the lexical-recognition task, the auditory itemswere the spoken words that were selected for thestudy. The lengths of these words were automati-cally adjusted in Praat to a mean length of1,020 ms and intensity of 70 dB. In the sentence-processing task, the linguistic stimulus consisted ofa simple sentence beginning with a semanticallyselective verb, followed by a noun from one of theselected semantic domains. The sentences in each

Table 1Information Regarding Category Assignment Including Total Number of Items in Each Category on the MBCDI, Distribution of Category Assign-ment to High- and Low-Density Domains Across Infants, and the Mean Proportion, Range, and Median Number of Words Produced in EachDomain

Total MBCDI items

High Low Words produced

% of children N % of children N Mean proportion (SD) Range (median)

Animals 43 .687 22 .313 10 .745 (.225) 17–43 (33.5)Body parts 27 .687 22 .313 10 .795 (.240) 2–27 (24.9)Clothing 28 .094 3 .906 29 .623 (.221) 5–26 (18.0)Drinks 7 .281 9 .719 23 .664 (.175) 3–7 (6.2)Fruits 7 .656 21 .343 11 .769 (.220) 3–7 (6.0)Vehicles 14 .593 19 .406 13 .786 (.148) 5–14 (11.0)

Note. MBCDI = MacArthur-Bates Communicative Development Inventories.

4 Borovsky, Ellis, Evans, and Elman

domain were: animals: pet the doggy; body parts:brush the teeth; clothing: wear the shoe; drink: drinkthe milk; fruit: eat the apple; vehicles: drive the car.As with the lexical-recognition task, all sentenceswere normalized for duration and intensity suchthat the duration of each word in the sentenceswas identical across items. The total sentence dura-tions were 1,449 ms, with verb duration of 563 ms,article duration of 189 ms, and noun duration of697 ms.

The speaker also recorded several encouragingitems and tag phrases that were intended to main-tain the infants’ attention during the study. Forexample, each lexical item was followed by a tagphrase such as “That’s cool!” or “Do you like it?”and the lexical item was preceded by the speakersaying, “Look!” Other encouraging phrases werepresented in between experimental trials along withother colorful images. All items were normalized to70 dB intensity.

Procedure

Vocabulary Assessment

Parents were mailed a MBCDI words and sen-tences form approximately 1 week before their lab-oratory visit and returned the completed checklistduring their appointment. In addition to theMBCDI, parents completed an additional compre-hension checklist for all items that appeared in thestudy on the day of their laboratory visit. Thisquestionnaire asked parents to mark whether theirchild understood the words to be used in thestudy, including all nouns and verbs in both tasks.Parents rated comprehension for each item on ascale of 1 = child does not understand the word to 4 =child definitely understands the word. Because wewere only interested in toddlers’ responses toknown words, we removed experimental responsesto any items where their parents marked a “1” or“2” to the target item.

Measurement of Lexical Recognition and SentenceProcessing

Infants were seated either in a car seat or ontheir caregiver’s lap while they viewed images on a17-in. LCD monitor positioned by a flexible armmount. The eye tracker camera was mounteddirectly underneath the monitor and was positionedbetween 580 and 620 mm from the infant’s fore-head. Auditory stimuli were presented via aspeaker placed behind the monitor. During the

experiment, caregivers listened to music via head-phones and were instructed not to name theimages.

Before the experimental trials began, the positionof the eye tracker was adjusted and the camera wasfocused while the infant viewed a short video. Thetracker was then calibrated with a 5-point routineusing an animated looming bull’s eye image pairedwith a whistling sound. Toddlers generally viewedthese calibration images without any explicitinstruction, but the experimenter pointed to thescreen when necessary to help direct the child’sgaze toward the appropriate images.

After the tracker was calibrated, the experimen-tal trials began. Each trial began with the presenta-tion of central colorful 30 9 30 pixel image. Oncethe toddler fixated toward this image, it disap-peared and the target and distractor image werepresented on the left and right side of the screenfor 2,000 ms in silence. The silent preview periodserved to familiarize the child with the objectimages and their locations in advance of the targetlabel. Next, a 100 9 100-pixel image (e.g., a smil-ing sun, a star) appeared at the center of thescreen as the sound “Look” was played. The cen-tral image automatically remained on the screenuntil the child fixated within its bounds for100 ms. Note, our stimulus arrangement placedthe edges of central stimulus and target/distractorimages approximately 1.3° visual angle apart, andthe center of the stimulus within 5° of visualangle. This visual angle value is within the 5°–10°of useful visual field that is commonly notedwithin visual search studies that include multipleobjects (reviewed in Irwin, 2013). Our stimulusarrangement therefore ensured that each objectremained within a useful field of view (i.e., requir-ing minimal working memory commitment forobject location), despite drawing momentary atten-tion to the center stimulus. We implemented thisgaze-contingent procedure to ensure that the infantwas attentive to the screen before proceeding withthe spoken stimulus. Once the infant fixated to thecenter image, the auditory stimulus was spoken(e.g., “Car!” or “Pet the doggy!”) followed by anencouraging tag phrase (e.g., “That’s cool!”) andthe central image disappeared. The target and dis-tractor images remained on the screen for 4,000 mspost label onset (see Figure 1, for an illustration ofthe task).

Each image appeared four times across thecourse of the study, as a yoked pair with anotherobject. The yoked object pairs in the lexical-recogni-tion task were banana/juice, diaper/nose, and

Semantic Structure and Language Processing 5

bird/airplane. In the sentence-processing task,yoked pairs were presented across the same cate-gories with the following items: apple/milk, dog/car, and shoe/teeth. Each object was twice labeledas the target and twice unlabeled as the distractorimage (and the order of target and distractor pre-sentation varied across versions), and each imageappeared on the left and right pseudorandomlywith equal frequency across these four trials. Thisyielded 12 trials in each task (24 total trials) thatwere interspersed and distributed equally acrossthree blocks. Filler trials appeared between eachblock with images containing complex scenes andexciting characters (e.g., Elmo), along with encour-aging statements such as “Wow, you’re doinggreat!” Within any one version, the target and dis-tractor images were equally likely to appear on theleft and right side across all categories. The entireprocedure lasted approximately 5–10 min.

Recording of Eye Movement Data

Moment-by-moment visual behavior wasrecorded from image onset to offset at 500 Hz byan SR Research EyeLink 1000 (SR Research:Ottawa, Ontario, Canada) eye tracker and binnedinto 50 ms intervals for offline analysis. Eye move-ments were automatically classified into fixations,saccades, and blinks according to default trackersettings and areas of interest were defined as the

two 400 9 400 pixel locations of the target and dis-tractor images.

Approach to Analysis

Assignment of High- and Low-Category Domains

Because our primary interest was to measurehow semantic density influences the recognition ofthat item, independent of the effects of vocabularysize, we implemented a metric of relative (ratherthan absolute) semantic density with respect to eachchild’s vocabulary. For each category, we calculatedeach child’s semantic density by dividing the num-ber of words that the child says in the category bythe total number of words surveyed in each cate-gory domain. Then, for each child, the categorieswith the three highest proportions were assigned tothe high-density condition, and the lowest propor-tion categories were assigned to the low-density set.We use a median-split assignment for high densityand low density rather than a continuous propor-tion measure to control for overall vocabulary sizeacross participants. In three participants, the third-and fourth-ranked category proportion was equiva-lent, so we assigned both categories to the high-density condition. This procedure therefore yieldeda unique arrangement of experimental items thatwas classified as high density and low density foreach infant. The distribution of category

Silent Preview [2000 ms]

Look! + Gaze-Contingent Center Stimulus [Variable Timing]

Language Onset (Exp 1:Shoe! That’sCool!, Exp 2: Pet the Doggy!) [4000ms]

Figure 1. Illustration of experimental trial in Experiments 1 and 2. In each task, the trial began with a preview period, where the targetand distractor images initially appeared alone on the left and right side of the monitor. Next, a small, colorful center stimulus (e.g., asmiling sun) appeared, and an auditory stimulus was concurrently presented (Look!). Once the infant fixated on the center stimulus for100 ms, the center image automatically disappeared, and the label for the target image was spoken (e.g., Shoe! That’s cool!/Drive thecar!). Due to licensing restrictions, similar, but not identical, images of depicted objects appeared in the study. All illustrated imagescourtesy of http://dreamstime.com.

6 Borovsky, Ellis, Evans, and Elman

assignments to high- and low-density conditions isindicated in Table 1.

Analysis of Eye Movement Data

In both tasks, we explored how looks toward thelabeled target item vary as a function of domainknowledge over larger and smaller windows ofanalysis. For the larger windows of analysis, we ini-tially calculated a measure of accuracy that is com-monly used in infant eye-tracking research, whichis defined as the fixations to the target divided bythe sum of fixations toward the target and distrac-tor images. With this metric, accuracy ranges from0 to 1, with values > 0.5 indicating a preference tolook toward the target (vs. distractor) over the spec-ified time region. In the lexical-recognition task, thisaccuracy metric was averaged over a broad timewindow from 300 to 1,800 ms post label onset tocorrespond with time windows used in other stud-ies of infant lexical recognition (Fernald et al.,2008). In the sentence-processing trials, there weretwo potentially informative events of interest acrossthe linguistic stimulus (the spoken verb and thesentence final object). Thus, we defined two timewindows of interest for statistical analysis, follow-ing precedent from prior research using a similarlystructured task (Mani & Huettig, 2012). The firsttime window, the anticipatory window, wasdefined as the portion of the sentence that occurredbefore the onset of the sentence final noun, span-ning from 300 post verb onset to the 750 ms timebin (noun onset was 752 ms). The second time win-dow, the noun window, spanned from 300 ms postonset of the noun to 1,800 ms post noun onset (i.e.,from the time bins spanning 1,050 to 2,550 ms postsentence onset).

Next, we carried out an finer grained analysis ofreal-time comprehension in both tasks using smaller50 ms time bins to address whether and when therewere differences in looks to the target (vs. distrac-tor) as a function of semantic density. We adopteda dependent measure, log-gaze proportion ratio,that is centered around zero and varies betweenpositive and negative infinity. With this measure,larger scores indicate a relative advantage to viewthe target (vs. distractor). Log-gaze ratios have beenadopted in recent years in the eye-tracking litera-ture as a way to overcome violations of statisticalassumptions of linear independence (becauseincreases in fixations to one item necessarilydecrease looks to the alternate image) and homo-geneity of variance (because simple proportionratios vary between 0 and 1; see Arai, van Gompel,

& Scheepers, 2007; Borovsky, Sweeney, Elman, &Fernald, 2014; Knoeferle & Kreysa, 2012, for a simi-lar approach). In each time bin, we calculated thelog-gaze proportion ratio for the target versus dis-tractor as log(P[target]/P[distractor]). Because logratios are undefined for zero values, we replacedevery instance of a zero value in the numerator ordenominator with a value of 0.01.

Using this log-gaze metric across 50 ms timebins, we addressed our primary questions regard-ing differences in domain density in lexical-recogni-tion and sentence-processing trials using astatistical approach that seeks to detect reliable dif-ferences across time between conditions while con-trolling for multiple comparisons (Type I error). Weimplemented a nonparametric cluster-based permu-tation analysis that has been most commonlyapplied to FMRI (functional magnetic resonanceimaging) and ERP (event-related potentials) timeseries analyses (Groppe, Urbach, & Kutas, 2011a;Maris & Oostenveld, 2007) and has become morecommon in eye-tracking data analysis (Barr, Jack-son, & Phillips, 2014; Von Holzen & Mani, 2012).This test calculates a cluster t statistic that sumsacross temporally adjacent point-wise t values thatexceed a predefined threshold. This cluster t statis-tic is then compared to a null-hypothesis distribu-tion of cluster t values that are generated via aMonte Carlo permutation approach (outlined inBarr et al., 2014, appendix). Following standards setby Groppe, Urbach, and Kutas (2011b), we used2000 random permutations to generate a distribu-tion of the null hypothesis with sufficient precisionto control family-wise error rate to < 0.05.

We used this cluster approach in both tasks toaddress whether differences in the time course oflinguistic processing vary by domain knowledge bydirectly comparing high- versus low-density log-gaze proportion ratios across time (using a thresh-old of p < .05, two sided for point-wise compar-isons). We additionally asked when fixations to thetarget exceed those of the distractor by identifyingtime clusters where log-gaze proportions exceed 0(using a threshold of p < .05, one sided) for high-and low-density domain items, separately.

Because we are primarily interested in howresponses to known words are influenced by thestructure of the child’s vocabulary, we onlyincluded trials in analysis where the parentreported that their child understood the targetword in both tasks, as well as the verb in the sen-tence-recognition task. Using this criterion, weexcluded two trials (of 384, or 0.52% of total trials)in the lexical-recognition task and another eight

Semantic Structure and Language Processing 7

trials (of 384, or 2.1% of total trials) in the sentence-processing task. Next, we removed trials whereinfants focused on the target and distractor for< 20% of the 300–1,800 ms analysis time windowfor lexical-recognition trials, or < 20% of the 300–2,550 analysis windows for sentence-processing tri-als. We adopted this threshold following precedentset in other eye-tracking studies with young chil-dren, such a Nordmeyer and Frank (2014), whoremoved trials with over 30% of samples missing,and Quam and Swingley (2014), who excluded tri-als where children fail to view the pictures for 17%of the time window (for 300 ms out of a 1,650 msanalysis window).

This 20% criterion led us to remove 36 of theremaining 382 trials in the lexical-recognition task,or 9.4%, and an additional 27 trials of the remain-ing 376 trials in the sentence-processing task (7.2%of remaining trials). In total, 38 trials were excludedfrom analysis, or 9.9%, of the lexical-recognitiontask data set, leaving 346 remaining trials in thefinal analysis. In the sentence-processing data set,8.9% of total trials were removed, leaving 350 trialsover which subsequent analyses were performed.

Results

Lexical-Recognition Task

Figure 2A illustrates the time course of fixationproportions toward the target and distractor images

for high- and low-density items in the lexical-recog-nition task. As expected, there was a rapid rise infixations post label onset and fixations toward thetarget appeared to exceed those of the distractorwithin 800–1,000 ms post label onset in both high-and low-density conditions before the offset of thelabel at 1,020 ms. The time course plot also illus-trates differences between the real-time recognitionof high- and low-density items. Infants appear tobe faster to fixate uniquely toward the target itemin the low- versus high-density condition. This dif-ference appears to be partially driven by a rise infixations toward the distractor item relative the tar-get in the low-density condition that does not existin the high-density condition. This pattern is echoedin the log-gaze ratio plots in Figure 2B, which indi-cate an early advantage in distractor versus targetfixations (negative scores) for low- but not high-density items.

We initially carried out an analysis of target accu-racy from 300 to 1,800 ms across high- and low-den-sity items (Figure 3). Target accuracy exceeded 0.5in both conditions, thigh(31) = 7.5, phigh < .0001,dhigh = 2.69; tlow(31) = 7.09, plow < .0001, dlow = 2.54,indicating that participants successfully interpretedthe label in this time window in both conditions.Further analysis also found no significant differencesas a function of semantic density (Mhigh = 0.68,SDhigh = 0.13; Mlow = 0.65, SDlow = 0.12),t(31) = 0.96, p = .34, d = 0.34. These comparisonsindicate that listeners identified the appropriate

A B

Figure 2. (A) Plot of fixation proportions to target and distractor images in high- and low-density domains in 50 ms time bins startedat the onset of the target label in Experiment 1. (B) Log-gaze plots indicate relative advantage of fixations to the target (positive values)or distractor (negative values) in high- and low-domain items across 50 ms time bins, starting at the onset of the spoken label for itemsin Experiment 1. Error bars represent � 1 SE of the mean.

8 Borovsky, Ellis, Evans, and Elman

target item in both high- and low-density conditions,but there were no differences due to density in over-all accuracy of gaze toward the target item over thisrelatively broad time window (Figure 3).

The next analyses addressed whether and whenthere were finer grained differences across the timecourse of lexical recognition as a function ofdomain density. We first asked when looks to thetarget exceeded those to the distractor (indicatingrecognition of the label) with a cluster-based per-mutation analysis that identified the time binswhen log-gaze proportion ratios were significantlypositive for high- and low-density items. This anal-ysis indicated a target preference from 650 to1,800 ms for high-density items (cluster t = 130.8,Monte Carlo p = .002) and from 850 to 1,800 ms forlow-density items (cluster t = 124.2, Monte Carlop = .001). Together, these results indicate that par-ticipants rapidly interpreted the spoken label, bydirecting their fixations consistently toward the tar-get item before the label was completely spoken (la-bel offset was at 1,020 ms) and continuing to do sountil the end of the analysis window. Additionally,this analysis suggests that the identification of thetarget item was approximately 200 ms faster in thehigh- versus low-density domain.

We next directly compared log-gaze proportionsin 50 ms time bins across high- and low-densitydomains using a cluster-based permutation analy-sis, described earlier. This direct comparison ofdomain effects indicated a difference between

domains from 300 to 650 ms post label onset (clus-ter t statistic = 18.32, Monte Carlo p = .027). Thisdifference, as seen in Figure 2B, was driven by rela-tively greater log-gaze ratios for the high- (vs. low-)density items and supports the findings in the firstcluster analysis that participants were relatively fas-ter to settle on the appropriate target item in thehigh-density condition.

The Role of Distractor Semantic Density

As one reviewer helpfully noted, the semanticdensity of the distractor object may exert an inde-pendent influence on target recognition. We there-fore carried out an exploratory analysis ofdistractor density on target recognition in four mainconditions: (a) high-density target–high-density dis-tractor (high T–high D), (b) high-density target–low-density distractor (high T–low D), (c) low-den-sity target–high-density distractor (low T–high D),and (d) low-density target–low-density distractor(low T–low D). We note that our experimental trialswere not balanced with respect to distractor. There-fore, participants did not contribute trials equallyacross all conditions in this analysis. Total trials ineach condition were high T–low T: 75, high T–lowD: 100, low T–high T: 99, and low T–low D: 72. Weconducted a (2 9 2) repeated measures analysis ofvariance (ANOVA) with target density (high andlow density) and distractor density (high and lowdensity) as factors. There were no main effects oftarget density, F(1, 70.29) = 0.89, p = .35, and dis-tractor density, F(1, 70.29) = 0.07, p = .78, and therewas no significant interaction effect, F(1,88.9) = 1.73, p = .19. Follow-up Tukey analyses didnot reveal significant differences among any of theindividual comparisons.

Sentence-Processing Task

The time course of fixations toward target anddistractor images for high- and low-density items inthe sentence-processing task is illustrated inFigure 4A. As expected, we saw a rapid rise in fixa-tions toward the target object before it was men-tioned, consistent with prior literature on predictionthat suggests that semantically selective verbs facili-tate (anticipatory) prediction of thematic objects(Altmann & Kamide, 1999; Mani & Huettig, 2012).

We initially carried out an analysis of targetaccuracy across two broad time windows: (a) acrossthe anticipatory period (300–750 ms post sentenceonset), and (b) during the noun time window(1,050–2,550 ms post sentence onset), illustrated in

Figure 3. Proportion of fixations toward the target and distractorimages from 300 to 1,800 ms post label onset in high- and low-domain categories in Experiment 1. Error bars represent � 1 SEof the mean. Shapiro–Wilk goodness-of-fit test statistics indicatedthat all means did not violate normality assumptions (ps > .11).

Semantic Structure and Language Processing 9

Figure 5. In the anticipatory window, accuracy forthe high-density items (Mhigh = 0.54, SDhigh = 0.20)exceeded that of the low-density items (Mlow = 0.42,SDlow = 0.24), t(31) = 1.99, p = .056, d = 0.71,though neither value significantly exceeded 0.5,thigh(31) = 1.13, phigh = .13, dhigh = 0.41; tlow(31) =�1.98, plow = .97, dlow = �0.71. This comparisonindicates that there was a difference as a functionof domain density in recognizing the coordinatingverb, although there is not strong evidence for pre-diction of the unspoken noun in this time window.In the noun window, accuracy did not differ as afunction of domain density (Mhigh = 0.76, SDhigh =0.13; Mlow = 0.73, SDlow = 0.13), t(31) = 0.75,p = .46, d = 0.27, and accuracy exceeded 0.5 forboth high-density, t(31) = 11.01, p < .0001, d = 3.95,and low-density items, t(31) = 10.38, p < .0001,d = 3.73. These comparisons indicate that listenersviewed the target item in response to the nounlabel in high- and low-density conditions, but therewere no differences due to density in overall accu-racy of gaze toward the target item over this broadtime window (Figure 5).

We next carried out a fine-grained comparison oflog-gaze fixation ratios for high- and low-densityitems across the time course to determine whetherthere is a difference in the timing of fixationsaccording to domain knowledge. In the anticipatorywindow, a cluster-based permutation analysis didnot identify any periods of time that significantlyexceeded 0 in the high- and low-density conditions.However, there is a significant density difference

from 300 to 650 ms post verb onset (cluster t statis-tic = 17.04, Monte Carlo p = .032). This differencewas driven by relatively larger log-gaze ratios (i.e.,a preference to view the target) for high- versuslow-density items in the anticipatory window. Inthe noun window, log-gaze ratios significantlyexceeded 0 across the entire time window for bothhigh-density (cluster t = 215.5, Monte Carlop = .002) and low-density (cluster t = 187.3, MonteCarlo p = .002) items. There were no significant

A B

Figure 4. (A) Plot of fixation proportions to target and distractor images in high- and low-density domains in 50 ms time bins startingat the onset of the verb and continuing through the article (art) and noun in Experiment 2. (B) Log-gaze plots indicate relative advan-tage of fixations to the target (positive values) or distractor (negative values) in high- and low-domain items across 50 ms time bins,starting at the onset of the verb for items in Experiment 2. Error bars represent � 1 SE of the mean.

Figure 5. Proportion of fixations toward the target and distractorimages during the verb window, and the target window in high-and low-density domain categories in Experiment 2. Error barsrepresent � 1 SE of the mean. Shapiro–Wilk goodness-of-fit teststatistics indicated that all means did not violate normalityassumptions (ps > .15).

10 Borovsky, Ellis, Evans, and Elman

cluster differences across density in the noun win-dow (cluster t statistic = 7.03, Monte Carlo p = .18).

Together, the findings across both windows indi-cate that participants recognized the target once itwas labeled, in both high- and low-density condi-tions. Importantly, differences due to densityemerged during the anticipatory window before thetarget was labeled. Because log-gaze ratios did notsignificantly exceed 0 during this period, this analy-sis indicates that density effects in the anticipatorywindow were not driven by differences in viewingthe distractor image. Children were more likely toview the distractor for low-density items. This pat-tern suggests that density effects in the anticipatorywindow were driven by greater interference fromthe distractor item in low-density domains.

The Role of Distractor Semantic Density

As in the lexical-recognition task, we explored theeffect of distractor density on the recognition of thetarget word across high- and low-density target anddistractor conditions with a 2 9 2 ANOVA with tar-get density (high and low) and distractor density(high and low) as factors in both the anticipatorytime window and the noun window. As stated ear-lier, trials were not balanced across conditions, suchthat the total number of trials in each condition washigh T–high D: 77, high T–low D: 99, low T–high D:102, and low T–low D: 72. In the anticipatorywindow there was a main effect of target density, F(1, 95) = 7.89, p = .006, with greater accuracy forhigh-density targets. There was also an effect of dis-tractor density, F(1, 95) = 6.63, p = .012, with greaterlooking toward the target in the presence of a high-(vs. low-) density distractor. There were no signifi-cant main or interaction effects in the noun window(all ps > .2). This analysis indicates that, during theanticipatory window, 2-year-olds showed greaterlooking toward the target when it was from a high-density domain, as well as when the distractor itemwas also from a high-density domain. Although thispattern might reflect greater referential certainty andanticipatory processing in conditions where both thetarget and distractor images are from categories thathave relatively greater semantic density, this effectdid not carry over into time period when the targetobject was labeled.

Discussion

Whereas previous studies had found that vocabu-lary growth was associated with linguistic-

processing skill and semantic development individ-ually, little was known regarding the relationbetween processing and semantic development. Wetherefore asked whether real-time lexical recogni-tion in infancy varied according to the structure ofthe 24-month-olds’ semantic knowledge using awithin-subjects experimental design.

We outlined two possible outcomes. One hypoth-esis (the interference hypothesis) predicted thatincreased semantic density would boost semanticcompetition from related items and result in lessrobust lexical recognition for high- (vs. low-) den-sity items. This pattern would be analogous to find-ings where lexical recognition is slowed for itemswith higher phonological neighborhood density(Garlock et al., 2001; Vitevitch et al., 1999). Anotherplausible but contrasting potential outcome (thefacilitation hypothesis) was that greater semanticdensity within a category might boost lexical recog-nition, as having knowledge of many semanticallyrelated items could increase activation for theintended lexical items. This outcome is predicted,for example, by distributional and semantic featuremodels of lexical activation, where activation forany item can be facilitated by activation of similarmeanings that share many features or contexts(Cree, McRae, & McNorgan, 1999; Landauer &Dumais, 1997; Plaut & Booth, 2000).

Our findings are most consistent with the latterpotential outcome. In both experiments, fixations tohigher density targets exceeded lower density targetfixations, as predicted by the lexical facilitationhypothesis. However, this finding was partially dri-ven by an early preference for the distractor item inlow-knowledge conditions. This early interferencepattern suggests that low-knowledge items experi-enced semantic interference from distractor items,whereas high-knowledge items did not. This pat-tern suggests that recognition of the high-densityitems may not be purely driven by boosted lexicalactivation for high-knowledge items insomuch as itis simultaneously driven by reduced semantic inter-ference.

We should also note that this distractor prefer-ence existed despite the fact that our counterbalanc-ing scheme ensured that all items were equallylikely to serve as targets and distractors, therebyruling out a simple explanation of this effect due tovisual saliency or preference for some items overothers. Similarly, although we used the same cate-gories across tasks, we selected different (highly fre-quent) items that are commonly known to 18- to24-month-old children in each task (Dale & Fenson,1996). Thus, this distractor preference for low-

Semantic Structure and Language Processing 11

density conditions remained consistent across dif-ferent items. We also individually verified knowl-edge of the experimental items (nouns and verbs)via parental report. Thus, the density effects areunlikely to be driven by a failure to understand theexperimental items. In sum, it appears lexical andsentential processing varies as a function of seman-tic knowledge, such that items that have a densernetwork of semantically related items experiencegreater simultaneous lexical facilitation and reducedinterference than items that have sparser semanticcategory neighborhoods.

Perhaps one of our more intriguing findings isthat semantic density influenced performance in asentence-processing task. In this case, we found nostrong evidence for linguistic prediction per se norwas there an effect of density as the target imagewas labeled. Instead, density effects emergedduring the anticipatory window before the targetnoun was even mentioned. Additionally, the dis-tractor image also contributed to this effect. We dis-cuss each of these points individually below.

With respect to the lack of strong support foranticipatory processing, several points about ourdesign are relevant here. First, we do not include asemantically neutral verb condition unlike otherparadigms that explore prediction with semanticallyselective verbs. The main reason for this decision isthat our design is not necessarily about predictionper se but rather an extension of how semantic den-sity may influence the linguistic processing beyonda single named object. Second, the timing of oursentence materials was faster than that of similarprevious studies. For example, the stimuli in Maniand Huettig (2012) had at least 1,500 ms from theonset of the sentential verb and disambiguatingnoun, whereas the same window in the currentstudy was 752 ms in duration. It seems plausiblethat these factors may explain the lack of stronganticipatory effects in our task. Future work isneeded to address how speech rate contributes topredictive processing in young children.

Another perplexing finding from our sentencetask was that there was not an effect of density asthe sentence final object was spoken. This patterndiffers from the lexical-recognition task, whererecognition of a spoken noun (which was not pre-ceded by a semantically constraining verb) variedas a function of semantic density. Instead, densityeffects appeared before noun onset, as the verb wasspoken during the anticipatory window. It seemslikely that semantic density effects resolved beforethe onset of the noun, suggesting that 2-year-oldshad already begun to (pre)activate coordinating

semantic information associated with the targetnoun while the verb was spoken.

Next, with respect to our findings of densityeffects during the anticipatory window, we foundthat density differences arose as the semanticallyselective verb was spoken—preceding the onset ofthe target label. This difference was driven by agreater tendency for participants to view the dis-tractor item during the anticipatory window forlower versus higher density items. This finding sug-gests that lower density items experienced greaterlexical interference from competing distractors thandid high-density items but only during the anticipa-tory window. This finding mirrors our results fromExperiment 1, where lower density target itemsexperienced relatively greater distractor interferenceat early points in linguistic processing as well.

At the same time, we also note an effect of thedistractor image density on target recognition dur-ing the anticipatory window. In this case, targetrecognition was facilitated when the distractorimage also belonged to a higher density domain.We interpret these findings with caution, becausethe experiment was not initially designed to care-fully control for distractor effects across all partici-pants. Our results nevertheless are suggestive thattarget recognition is facilitated by greater semanticdensity, and this effect extends from the semanticneighborhood surrounding both the target and dis-tractor items. In this case, it appears that targetrecognition is boosted in conditions where there isgreater referential certainty about the distractorobject so that it may more easily be discounted as apotential lexical referent.

Taken together, the sentence-processing taskfindings indicate that semantic domain knowledgeat the noun level has a general effect on languageprocessing that extends beyond the individual lexi-cal item. In other words, verb recognition variesaccording to the density of near semantic neighborsusing similar lexical activation mechanisms thatexist for nouns.

We next comment more generally on the lexicalinterference effect that emerged for low-densityitems across both experimental tasks. This interfer-ence effect for low-density items was relativelyunexpected given our knowledge of lexical interfer-ence from prior literature in young children. Whatare the mechanisms that might account for this pat-tern? One possibility is that items from lower den-sity categories may, in general, show relatively lessrobust lexical activation than items in higher den-sity domains. This difference may stem from some-what weaker or less tightly connected semantic

12 Borovsky, Ellis, Evans, and Elman

representations for lower density items, which, inturn, would create increased interference from unre-lated distractor items present in the array. Such anexplanation seems consistent with prior studies thatfail to find consistent semantic priming effects in18-month-old children (e.g., Arias-Trejo & Plunkett,2010; Styles and Plunkett, 2009). Younger children’srelatively smaller lexicons are likely to be less densecompared to older 24-month-old children. Clearly,it will be important to understand how the dynam-ics of lexico-semantic processing evolve in tandemwith the growing lexicon.

In sum, our findings also add to growing evidencethat early infant lexicons are organized semantically,much like adult lexicons. A number of prior studiesnow indicate that infants as young as 18–24 monthsrecognize lexical similarities between known andnovel words according to taxonomic, perceptual,thematic, and structural relations (Mani & Borovsky,in press). There is also some evidence that the recog-nition of these relations may change according to theoverall vocabulary knowledge of the individual(R€am€a et al., 2013). Our findings additionallydemonstrate that speech processing skills are tiednot only to the overall knowledge of the child butvary according to individual knowledge andexperience with items in different domains.

Most importantly, these results suggest that psy-cholinguistic mechanisms of lexical activation andrecognition that might be assumed to operatehomogeneously across our entire lexicon insteaddevelop idiosyncratically and heterogeneouslyaccording to the child’s knowledge and experience.More generally, our research supports a growingbody of evidence that early experiences can have atremendous impact on categorical and lexical learn-ing. For example, early home experiences (e.g.,experience with pets) can influence learning andattention to a category (e.g., cats) as early as4 months of age (Hurley & Oakes, 2015; Kovack-Lesh, McMurray, & Oakes, 2014). Mayor andPlunkett (2014b) have also recently described a U-shaped pattern in early variability in word knowl-edge between children, with expressive vocabularyincreasing in variability between 15 and 24 monthsof age, before becoming more coherent betweenchildren. Together, these findings indicate that earlyexperiences may have a particularly substantialimpact on early learning and attentional mecha-nisms, primarily in the first 2 years, and that thereis a need to explore whether and how early differ-ences in knowledge and experience may lead tolasting impacts in the developmental trajectory andoutcomes on early lexical and cognitive skills.

These findings also connect with a growing liter-ature that explores how the current organization ofthe lexicon influences vocabulary growth acrossinfancy. Rather than learning a random cohort ofwords, there is increasing evidence that childrenlearn words in a “clustered” fashion, via a tendencyto add new words to their vocabulary that arerelated to already known items (Beckage et al.,2011; Hills et al., 2009; Steyvers & Tenenbaum,2005). This pattern of vocabulary growth, termed“preferential attachment” by network researchers(Barab�asi & Albert, 1999), and initially connected tosemantic vocabulary growth by Steyvers andTenenbaum (2005), may facilitate word learning byproviding a basis by which learners generalize priorknowledge to novel words that share some seman-tic overlap (Borovsky & Elman, 2006). The structureof knowledge may also affect the basic strategiesthat children use to acquire words (Colunga &Sims, 2011; Yurovsky et al., 2012). Although themeasures that we use are not identical to the previ-ously mentioned network modeling analyses, ourmetrics of semantic density within categories docapture elements of semantic microstructure ofearly vocabulary. Importantly, this work suggeststhat increasing semantic density may simultane-ously reduce potential interference from lexicalcompetitors and boost activation for intended lexi-cal items. These processes extend to novel wordlearning, where it is clear that 2-year-old childrenencode and recognize semantic relationshipsbetween novel items that share similar perceptualfeatures and that appear in similar sentence con-texts (Wojcik & Saffran, 2013; Wojcik & Saffran,2015). Similarly, lexical processing is facilitated fornovel items within denser semantic domains (Bor-ovsky, Ellis, Evans, & Elman, in press). Additionalwork is needed to delineate how mechanisms oflexical competition and facilitation interact duringknown word recognition and novel word learning.

We also note some limitations in the currentstudy. First, this study explores the effect of seman-tic density on processing at a single age. It will beimportant to extend this work to address whetherand how domain knowledge influences perfor-mance as vocabulary size and structure changesacross development. Another issue arises from thefact that some categories were, on average, morelikely to fall in a high- or low-density domain thanothers (e.g., clothes). Future research would need toexplore whether these overarching differences maydrive density effects due to the salience or fre-quency in the input for these particular categories.A further limitation is that our measure of semantic

Semantic Structure and Language Processing 13

density relies on a measure of expressive vocabu-lary and not receptive knowledge. Recent worksuggests that productive vocabulary may have atighter relationship with processing skills than thoseof receptive vocabulary in 24-month-olds (Mani &Huettig, 2012); however, additional work is neededto explore whether this link also extends to seman-tic density as well.

Conclusions

Vocabulary growth is connected to the develop-ment of a plethora of linguistic and nonlinguisticskills. Ultimately, a full understanding of languageacquisition will rest not only in identifying the rele-vant mediators of vocabulary growth but also incharacterizing the interactions between these fac-tors. Our studies provide initial evidence that thesemantic organization of items within the child’slexicon leads to important changes in how wordmeanings are activated and recognized. Our find-ings additionally reveal some initial clues as to howthese processes may scale to complex multiwordlanguage-processing tasks. Our results additionallysuggest that “language-processing skill” is not apurely endogenous or unitary construct as it varieswithin an individual according to their experienceand knowledge. These findings indicate a hopefulpossibility that organizing vocabulary training toitems within semantic domains may yield benefitsfor processing skills and vocabulary growth. Weplan to explore whether targeted vocabularyinstruction within semantic domains could improveprocessing skills and, by extension, language andacademic outcomes more generally.

References

Altmann, G. T. M., & Kamide, Y. (1999). Incrementalinterpretation at verbs: Restricting the domain of subse-quent reference. Cognition, 73, 247–264. doi:10.1016/S0010-0277(99)00059-1

Altvater-Mackensen, N., & Mani, N. (2013). Word-formfamiliarity bootstraps infant speech segmentation.Developmental Science, 16, 980–990. doi:10.1111/desc.12071

Arai, M., van Gompel, R. P. G., & Scheepers, C. (2007).Priming ditransitive structures in comprehension. Cog-nitive Psychology, 54, 218–250. doi:10.1016/j.cog-psych.2006.07.001

Arias-Trejo, N., & Plunkett, K. (2010). The effects of per-ceptual similarity and category membership on earlyword-referent identification. Journal of ExperimentalChild Psychology, 105, 63–80.

Barab�asi, A.-L., & Albert, R. (1999). Emergence of scalingin random networks. Science, 286, 509–512. doi:10.1126/science.286.5439.509

Barr, D. J., Jackson, L., & Phillips, I. (2014). Using a voiceto put a name to a face: The psycholinguistics of propername comprehension. Journal of Experimental Psychology:General, 143, 404–413. doi:10.1037/a0031813

Bayley, N. (2006). Bayley scales of infant and toddler develop-ment, 3rd ed. San Antonio, TX: Harcourt Assessment Inc.

Beckage, N., Smith, L., & Hills, T. (2011). Small worldsand semantic network growth in typical and late talk-ers. PLoS ONE, 6, e19348. doi:10.1371/jour-nal.pone.0019348

Benasich, A. A., & Tallal, P. (2002). Infant discriminationof rapid auditory cues predicts later language impair-ment. Behavioural Brain Research, 136(1), 31–49.doi:10.1016/S0166-4328(02)00098-0

Boersma, P., & Weenink, D. (2012). Praat: doing phoneticsby computer (Version 5.3.22). Retrieved from http://www.praat.org/

Borovsky, A., Ellis, E. M., Evans, J. L., & Elman, J. L. (inpress). Lexical leverage: Category knowledge boostsnovel word recognition in two-year-olds. DevelopmentalScience. Advance online publication. doi: 10.1111/desc.12343

Borovsky, A., & Elman, J. (2006). Language input andsemantic categories: A relation between cognition andearly word learning. Journal of Child Language, 33, 759–790. doi:10.1017/S0305000906007574

Borovsky, A., Elman, J. L., & Fernald, A. (2012). Knowinga lot for one’s age: Vocabulary and not age is associ-ated with incremental sentence interpretation in chil-dren and adults. Journal of Experimental ChildPsychology, 112, 417–436. doi:10.1016/j.jecp.2012.01.005

Borovsky, A., Sweeney, K., Elman, J. L., & Fernald, A.(2014). Real-time interpretation of novel events acrosschildhood. Journal of Memory & Language, 73, 1–14. doi:doi:10.1016/j.jml.2014.02.001

Colunga, E., & Sims, C. E. (2011). Early talkers and latetalkers know nouns that license different word learningbiases. Proceedings of the 33rd Annual Conference ofthe Cognitive Science Society. Austin, TX: CognitiveScience Society.

Cree, G. S., McRae, K., & McNorgan, C. (1999). An attrac-tor model of lexical conceptual processing: Simulatingsemantic priming. Cognitive Science, 23, 371–414.doi:10.1016/S0364-0213(99)00005-1

Cunningham, A. E., & Stanovich, K. E. (1997). Early read-ing acquisition and its relation to reading experienceand ability 10 years later. Developmental Psychology, 33,934–945. doi:10.1037/0012-1649.33.6.934

Dale, P. S., & Fenson, L. (1996). Lexical developmentnorms for young children. Behavior Research Methods,Instruments & Computers, 28, 125–127. doi: 10.3758/BF03203646

Fenson, L., Marchman, V. A., Thal, D. J., Dale, P. S., &Reznick, J. S. (2007). MacArthur-Bates Communicative

14 Borovsky, Ellis, Evans, and Elman

Development Inventories: User’s guide and technical man-ual. Baltimore, MD: Paul H. Brookes.

Fernald, A., & Marchman, V. A. (2012). Individual differ-ences in lexical processing at 18 months predict vocab-ulary growth in typically developing and late-talkingtoddlers. Child Development, 83, 203–222. doi:10.1111/j.1467-8624.2011.01692.x

Fernald, A., Perfors, A., & Marchman, V. A. (2006). Pick-ing up speed in understanding: Speech processing effi-ciency and vocabulary growth across the 2nd year.Developmental Psychology, 42, 98–116. doi:10.1037/0012-1649.42.1.98

Fernald, A., Pinto, J. P., Swingley, D., Weinberg, A., &McRoberts, G. W. (1998). Rapid gains in speed of ver-bal processing by infants in the 2nd year. PsychologicalScience, doi:10.1111/1467-9280.00044

Fernald, A., Zangl, R., Portillo, A. L., & Marchman, V.(2008). Looking while listening: Using eye movementsto monitor spoken language comprehension by infantsand young children. In I. Sekerina, E. Fernandez, & H.Clahsen (Eds.), Language processing in children (pp. 113–132). Amsterdam, The Netherlands: Benjamins.

Garlock, V. M., Walley, A. C., & Metsala, J. L. (2001).Age-of-acquisition, word frequency, and neighborhooddensity effects on spoken word recognition by childrenand adults. Journal of Memory and Language, 45, 468–492. doi:10.1006/jmla.2000.2784

Gathercole, S. E., & Baddeley, A. D. (1989). Evaluation ofthe role of phonological STM in the development ofvocabulary in children: A longitudinal study. Journal ofMemory and Language, 28, 200–213. doi:10.1016/0749-596X(89)90044-2

Gershkoff-Stowe, L., Thal, D. J., Smith, L. B., & Namy, L.L. (1997). Categorization and its developmental relationto early language. Child Development, 68, 843–859.doi:10.1111/j.1467-8624.1997.tb01966.x

Gopnik, A., & Choi, S. (1990). Do linguistic differenceslead to cognitive differences? A cross-linguistic study ofsemantic and cognitive development. First Language, 10,199–215. doi:10.1177/014272379001003002

Gopnik, A., Choi, S., & Baumberger, T. (1996). Cross-lin-guistic differences in early semantic and cognitivedevelopment. Cognitive Development, 11, 197–225.doi:10.1016/S0885-2014(96)90003-9

Gopnik, A., & Meltzoff, A. N. (1987). The development ofcategorization in the second year and its relation toother cognitive and linguistic developments. ChildDevelopment, 58, 1523–1531. doi:10.2307/1130692

Gopnik, A., & Meltzoff, A. N. (1992). Categorization andnaming: Basic-level sorting in eighteen-month-olds andits relation to language. Child Development, 63, 1091–1103. doi:10.1111/j.1467-8624.1992.tb01682.x

Groppe, D. M., Urbach, T. P., & Kutas, M. (2011a). Massunivariate analysis of event-related brain potentials/fields I: A critical tutorial review. Psychophysiology, 48,1711–1725. doi:10.1111/j.1469-8986.2011.01273.x

Groppe, D. M., Urbach, T. P., & Kutas, M. (2011b). Massunivariate analysis of event-related brain potentials/

fields II: Simulation studies. Psychophysiology, 48, 1726–1737. doi:10.1111/j.1469-8986.2011.01272.x

Hart, B., & Risley, T. (1995). Meaningful differences ineveryday parenting and intellectual development in youngAmerican children. Baltimore, MD: Brookes.

Hills, T. T., Maouene, M., Maouene, J., Sheya, A., &Smith, L. (2009). Longitudinal analysis of early seman-tic networks preferential attachment or preferentialacquisition? Psychological Science, 20, 729–739.doi:10.1111/j.1467-9280.2009.02365.x

Huettig, F., Rommers, J., & Meyer, A. S. (2011). Using thevisual world paradigm to study language processing:A review and critical evaluation. Acta Psychologica, 137,151–171. doi:10.1016/j.actpsy.2010.11.003

Hurley, K. B., & Oakes, L. M. (2015). Experience and dis-tribution of attention: Pet exposure and infants’ scan-ning of animal images. Journal of Cognition andDevelopment, 16(1), 11–30.

Irwin, D. E. (2013). Fixation location and fixation durationas indices of cognitive processing. In F. Ferreira, & J.M. Henderson (Eds.), The interface of language, vision,and action: Eye movements and the visual world (pp. 105–134). New York, NY: Psychology Press.

Knoeferle, P., & Kreysa, H. (2012). Can speaker gazemodulate syntactic structuring and thematic roleassignment during spoken sentence comprehension?.Frontiers in Psychology, 3, 144–158. doi:10.3389/fpsyg.2012.00538

Kovack-Lesh, K. A., McMurray, B., & Oakes, L. M.(2014). Four-month-old infants’ visual investigation ofcats and dogs: Relations with pet experience and atten-tional strategy. Developmental Psychology, 50, 402.doi:10.1037/a0033195

Landauer, T. K., & Dumais, S. T. (1997). A solution toPlato’s problem: The latent semantic analysis theory ofacquisition, induction, and representation of knowl-edge. Psychological Review, 104, 211–240. doi:10.1037/0033-295X.104.2.211

Mani, N., & Borovsky, A. (in press). Building a lexicalnetwork. In G. Westermann & N. Mani (Eds.), Earlyword learning. Psychology Press.

Mani, N., & Huettig, F. (2012). Prediction during lan-guage processing is a piece of cake—But only forskilled producers. Journal of Experimental Psychology:Human Perception and Performance, 38, 843–847.doi:10.1037/a0029284

Mani, N., & Plunkett, K. (2010). In the infant’s mind’sear: Evidence for implicit naming in 18-month-olds.Psychological Science, 21, 908–913. doi:10.1177/0956797610373371

Mani, N., & Plunkett, K. (2011). Phonological primingand cohort effects in toddlers. Cognition, 121, 196–206.doi:10.1016/j.cognition.2011.06.013

Marchman, V. A., & Fernald, A. (2008). Speed of wordrecognition and vocabulary knowledge in infancy pre-dict cognitive and language outcomes in later child-hood. Developmental Science, 11, F9–F16. doi:10.1111/j.1467-7687.2008.00671.x

Semantic Structure and Language Processing 15

Maris, E., & Oostenveld, R. (2007). Nonparametric statisti-cal testing of EEG- and MEG-data. Journal of Neuro-science Methods, 164, 177–190. doi:10.1016/j.jneumeth.2007.03.024

Mayor, J., & Plunkett, K. (2014a). Infant word recognition:Insights from TRACE simulations. Journal of Memoryand Language, 71, 89–123. doi:10.1016/j.jml.2013.09.009

Mayor, J., & Plunkett, K. (2014b). Shared understandingand idiosyncratic expression in early vocabularies.Developmental Science, 17, 412–423. doi:10.1111/desc.12130

Mervis, C. B., & Bertrand, J. (1995). Early lexical acquisi-tion and the vocabulary spurt: A response to Goldfield& Reznick. Journal of Child Language, 22, 461–468.doi:10.1017/S0305000900009880

Mirman, D., & Magnuson, J. S. (2008). Attractor dynamicsand semantic neighborhood density: processing is slo-wed by near neighbors and speeded by distant neigh-bors. Journal of Experimental Psychology: Learning,Memory, and Cognition, 34(1), 65.

Nelson, K. (1973). Structure and strategy in learning totalk. Monographs of the Society for Research in ChildDevelopment, 38(1–2, Serial No. 149), 136.

Nordmeyer, A. E., & Frank, M. C. (2014). The role of con-text in young children’s comprehension of negation.Journal of Memory and Language, 77, 25–39.

Plaut, D. C., & Booth, J. R. (2000). Individual and devel-opmental differences in semantic priming: Empiricaland computational support for a single-mechanismaccount of lexical processing. Psychological Review, 107,786–823. doi:10.1037/0033-295X.107.4.786

Poulin-Dubois, D., Graham, S., & Sippola, L. (1995). Earlylexical development: The contribution of parental label-ling and infants’ categorization abilities. Journal of ChildLanguage, 22, 325–343. doi:10.1017/S0305000900009818

Quam, C., & Swingley, D. (2014). Processing of lexicalstress cues by young children. Journal of ExperimentalChild Psychology, 123, 73–89.

R€am€a, P., Sirri, L., & Serres, J. (2013). Development of lex-ical–semantic language system: N400 priming effect forspoken words in 18- and 24-month old children. Brainand Language, 125, 1–10. doi:10.1016/j.bandl.2013.01.009

Ramey, C. T., & Ramey, S. L. (2004). Early learning andschool readiness: Can early intervention make a

difference? Merrill-Palmer Quarterly, 50, 471–491.doi:10.1353/mpq.2004.0034

Schafer, G., & Plunkett, K. (1998). Rapid word learningby 15-month-olds under tighly controlled conditions.Child Development, 69, 309–320. doi:10.1111/j.1467-8624.1998.tb06190.x

Steyvers, M., & Tenenbaum, J. B. (2005). The large-scalestructure of semantic networks: Statistical analyses anda model of semantic growth. Cognitive Science, 29, 41–78. doi:10.1207/s15516709cog2901_3

Styles, S. J., & Plunkett, K. (2009). How do infants build asemantic system? Language and Cognition, 1(1), 1–24.doi:10.1515/LANGCOG.2009.001

Torkildsen, J. v. K., Syversen, G., Simonsen, H. G., Moen,I., & Lindgren, M. (2007). Electrophysiological corre-lates of auditory semantic priming in 24-month-olds.Journal of Neurolinguistics, 20, 332–351. doi:10.1016/j.jneuroling.2007.02.003

Vitevitch, M. S., Luce, P. A., Pisoni, D. B., & Auer, E. T.(1999). Phonotactics, neighborhood activation, and lexi-cal access for spoken words. Brain and Language, 68,306–311. doi:10.1006/brln.1999.2116

Von Holzen, K., & Mani, N. (2012). Language nonselec-tive lexical access in bilingual toddlers. Journal of Experi-mental Child Psychology, 113, 569–586. doi:10.1016/j.jecp.2012.08.001

Willits, J. A., Wojcik, E. H., Seidenberg, M. S., & Saffran,J. R. (2013). Toddlers activate lexical semantic knowl-edge in the absence of visual referents: Evidence fromauditory priming. Infancy, 18, 1053–1075. doi:10.1111/infa.12026

Wojcik, E. H., & Saffran, J. R. (2013). The ontogeny of lex-ical networks toddlers encode the relationships amongreferents when learning novel words. PsychologicalScience, 24, 1898–1905. doi:10.1177/0956797613478198

Wojcik, E. H., & Saffran, J. R. (2015). Toddlers encodesimilarities among novel words from meaningful sen-tences. Cognition, 138, 10–20. doi:10.1016/j.cogni-tion.2015.01.015

Yurovsky, D., Bion, R. A. H., Smith, L. B., & Fernald, A.(2012, August). Mutual exclusivity and vocabulary struc-ture. Paper presented at the 34th annual meeting of theCognitive Science Society, Austin, TX.

16 Borovsky, Ellis, Evans, and Elman