Abstracts

Posters:

Revisiting the Dichotomy Between Inflectional and Derivational Morphology

Jett Hampton

Mentor: Deo Ngonyani

Textbook definitions make a sharp distinction between derivational and inflectional morphology. Haspelmath (2010) challenges this dichotomy and proposes that the two are rather a single component over a continuum. In particular, this research explores two purported features distinguishing derivations from inflections. (a) While derivations change the word concept, inflections change only word form; (b) Derivations may change the word class while inflections may not. Bantu noun classification provides empirical data in support of the continuum approach. Here I explore this question through data from Dhaisu, an East-African Bantu Language spoken in northeastern Tanzania. A comprehensive lexicon was constructed from various narratives and vocabulary collected during fieldwork by researchers supported by MSU’s Alliance for African Partnership (2017-2018). The data was subsequently archived on FieldWorks platform. Extensive research on Bantu nouns has revealed that nominal prefixes are indicators of noun class (non-sex based genders eg human, abstract, etc) and number. Both are inflectional categories. Further analysis of nouns in Dhaisu led to the identification of 18 distinct noun classes. The study finds that when nominal prefixes distinguish the plant from the tree (nandi `coconut’ → mnandi ‘coconut tree’), for example, they are both inflectional or derivational. It is a change of the concept but also grammatical features of number and class. When an adjective takes the prefix for abstract nouns (-rito ‘heavy’ → urito ‘weight’), it changes the word class to noun, a derivational operation. Therefore, while the dichotomy may hold true for languages like English, affixes in Dhaisu may be both.

Competition Between Past and Perfect

Kerry Berres, Darby Grachek, Jett Hampton, Sarah Jones, Hollie Nusbaum, Sarah Sirna

Mentors: Cristina Schmitt, Komeil Ahari, Alan Munn

The Present Perfect ("I have jumped") competes with the Simple Past ("I jumped") in many of its uses. Previous work suggests that the Simple Past is taking on more uses than before. For example, the Present Perfect and Simple Past are interchangeable in sentences like ("I've already eaten" vs. "I already ate"). However, some speakers can use both tenses interchangeably in sentences like "I lived in East Lansing since I was three" vs. "I've lived in East Lansing since I was three", while other speakers reject the past tense in this context. In this experiment, our goal is to determine whether Present Perfect in narratives will elicit speakers to use the perfect more often, so we can determine when the present perfect is necessary. Subjects will hear two stories: one story has as much Present Perfect as possible and the other has only simple past and present tenses. Participants will then have to retell the stories to a third party from memory. We hypothesize that in a narrative context where the perfect is used extensively, participants will be more likely to use the perfect to retell the story than the story that uses simple past and present. The results will tell us in what contexts they find it acceptable to use the perfect instead of simple past tense and vice versa.

Reliability of LENA in Identification of Child Vocalizations in Naturalistic Environments

Ellen Brooks and Alexis Yang

Mentors: Laura Dilley, Matt Lehet, Meisam Arjmandi, Nikaela Losievski

Different factors of early auditory environments play a critical role in language development in both children with normal and impaired hearing. These factors are especially important for children with cochlear implants. Language development can be assessed by quantifying child vocalizations. This metric indicates how much the child is attempting to communicate, directly correlating to language acquisition. LENA (Language Environment Analysis), a device worn by infants with cochlear implants, attempts to analyze vocalization information. With LENA, researchers are able to: (1) grasp how much a child communicates with surrounding individuals and (2) receive insight into expressive language development. Child vocalization is one of the powerful metrics. However, the reliability of LENA's other analyzations, such as maternal infant-directed speech, was shown to be misclassified 14% of the time. Errors in one of LENA's automated classifications can impact the accuracy of other measures, such as the number of conversational turns. In this study, human coders manually identify child vocalizations in short intervals of audio for comparison with LENA. Our results pertain heavily to the understanding of clinicians and researchers in using this device to analyze the language development of children with cochlear implants.

Reliability of Language Environment Analysis (LENA) System in Quantifying Conversational Turns in Verbal Communication of Child-Environment

Chitra Lakshumanan, Maddie Haar, Kayla Baumgartner, Sophie Ghoneim

Mentors: Laura Dilley, Matt Lehet, Meisam Arjmandi

Language acquisition in children is influenced by the quality and quantity of parental linguistic input, particularly for hearing-impaired children wearing cochlear implants (CIs). The number of conversational turns between children and their caregivers is strongly predictive of language outcomes. The LENA (Language ENvironment Analysis) is a wearable device used to study children’s linguistic interaction with adults. LENA captures 16-hour recordings of a child’s home environment and uses automatic speech processing to count conversational turns. To evaluate its usefulness as a clinical tool, this study analyzes the reliability of LENA in accurately measuring conversational turns. According to LENA, a conversational turn is defined as adult and target child speech segments occurring in succession and separated by no other live human speakers. Quantifying turns as segment pairs, LENA is able to identify the number of conversational turns between children and their caregivers. LENA’s misclassification of speakers and intervening audio can lead to several false positives or negatives within the data. Despite many clinicians and researchers using these metrics, there has not been a thorough analysis of how the accuracy of these metrics vary across individuals and speech environments. Human coders’ classifications of conversational turns within short chunks of audio were compared to LENA’s to obtain a measure of reliability. These results suggest that the effectiveness of LENA as a clinical tool for estimating conversational turns should be carefully evaluated to better understand the effect of conversational turns on language development of children with CIs and the etiology of its variability across children.

The Effect of Lenition on The Distribution of Quantity Words

Anthony Delsanter, Becky Lubera, Katelyn Weatherford

Mentor: Cristina Schmitt

In many Latin-American dialects of Spanish there is a process that weakens syllable-final [s] and this affects the realization of plural morphology. So, Las patas (the ducks) can have the -s produced as [s], [h] (an aspiration), or nothing at all, making it identical to the singular form "la pata". Different dialects have different constraints on where and when lenition mostly occurs and they also vary in the overall rates of lenition. In this project we ask how the variation in lenition in Chilean, Paraguayan and Argentinian Spanish impacts speakers use of "quantity" words such as cardinal numbers, quantifiers, and determiners in the noun phrases across dialects. We examine audio-recordings and transcripts of mother-child interactions from Chilean, Paraguayan and Argentinian speakers. We have two goals: first, we examine qualitative and quantitative properties of lenition between the different dialects in order to determine whether the described differences across dialects are also present in Child Directed Speech and in Child speech (4-5 year-olds); and second, we test if there is a correlation between the rates and type of lenition and the use of different types of quantity words by mothers and children. The idea is to determine whether lenition changes how the information about plurality is encoded in child-directed-speech and child speech.

Who is ‘She’, and Can Null Subjects Help Paraguayan Children Find Out?

Daniel Greeson, Megan Placko, and Becky Lubera

Mentor: Cristina Schmitt

Spanish allows for sentences without an overt subject, such as "Estoy cansado" ('Am tired') where "yo"/'I' is not pronounced. Null and overt pronouns are preferred in different contexts, but both are grammatical. This leaves children with the task of learning when to use which form. In Mexican Spanish, overt subject pronouns generally refer to a different person than the previous subject, while null subject pronouns are preferred when maintaining the same reference as the previous subject. Previous studies hypothesized that this contrast can be acquired if children first track null and overt pronouns in the domain of first- and second-person pronouns ("yo", 'I' and "tú", 'you'), then use this knowledge from the 1/2-person domain to help them interpret 3rd person pronouns. However, a potential problem lies in the fact that dialects of Spanish vary in the acceptability of overt subjects in different contexts. One such dialect is Paraguayan Spanish which allows overt pronouns in many more contexts than other varieties of Spanish. Because of this difference, it's possible that the learning path differs for Paraguayan Spanish-speaking children. Using an Argentinian corpus of Paraguayan mother-child interactions we examine the distribution of null/overt subject pronouns across 1st, 2nd, and 3rd person contexts, and also across contexts with and without a switch in pronoun reference. Our goal is to show how children who speak this dialect end up learning to interpret pronoun reference.

Talks:

The Effects of Proximity of Slowed Prior Speech Context on Function Word Perception

Joshua Zhao and Elizabeth Remy

Mentor: Laura Dilley

Dilley and Pitt (2010) demonstrated that slowed rate of speech within a contextual frame could decrease the likelihood that listeners perceive a co-articulated function word within a “target region” relative to a normal rate context. This is called the lexical rate effect (LRE). However, there is uncertainty over whether the quantity or the proximity of the slowed speech context is driving the effect. Two experiments examined how varying which syllables were slowed before the co-articulated function word in the target region would parametrically affect function word perception. The preceding context was divided into three regions. The first region included the preceding context excluding two syllables before the target region. The second region included the 2 nd syllable before the target region. The third region included the syllable before the target region. Both experiments had a condition slowing the entire preceding context and a condition with the normal rate. Experiment 1 slowed the first region in one condition and the first and second region in another condition. Experiment 2 slowed the second region in one condition and the second and third region in another condition. This allows us to disentangle the question

of quantity of slowed audio from proximity of slowed audio. The comparison between these experiments informs theories of how lexical and phonetic perception rely upon speech rate.

Hungry Caterpillars and Desperate Mice: How Children’s Books Reflect Their Understanding of the Discourse Information

Abby Jaroszewicz

Mentors: Cristina Schmitt, Alan Munn

In conversation, there’s a mixture of knowledge added by the speakers and knowledge that is implicit as part of the context (the common ground). Children and adults have different strategies to determine information that can be assumed in the common ground; a child might say “They ate lots of carrots” not realizing the person they’re talking to might not know who “they” refers to. When the pronoun’s antecedent is in the discourse, pragmatic principles are required to connect them. Parents and educators are becoming more aware vocabulary predicts school performance, and many children’s books are marketed to different age groups in terms of number of unique words. Such word counts don’t consider the complexity of aligning pronouns and antecedents or other kinds of discourse-dependent grammatical properties. This research examines children’s books in terms of ellipsis, pronouns, and comparatives to investigate the role of discourse-dependent properties in books aimed at young children.