Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Poster Slams

Elderly speakers describe non-core topics in a picture more frequently than young speakers: a topic annotation analysis in the Cookie Theft picture

Poster D64 in Poster Session D with Social Hour, Friday, October 7, 5:30 - 7:15 pm EDT, Millennium Hall

Jasmine Sun1, Sameer Pradhan2, Galit Agmon1, Naomi Nevler1, Sanjana Shellikeri1, Sharon Ash1, Mark Liberman2, Murray Grossman1, Sunghye Cho2; 1Penn Frontotemporal Degeneration Center, 2Linguistic Data Consortium

Introduction: Speakers’ linguistic behavior changes over time. Previous studies on aging have examined various linguistic aspects and shown, for example, that elderly speakers exhibit reduced fluency and long pause duration but produce more words compared to younger speakers. While many previous studies used brief picture descriptions in comparing elderly and young speakers, they did not compare how elderly and young speakers differ when describing topics in the picture. Picture description tasks involve speakers’ active selection process of which topic to describe in the picture, revealing which topic they consider important. We expected elderly and young speakers would differ in their topic choice, and this study examined the age group difference in topic selection using the Cookie Theft picture descriptions. Methods: We analyzed Cookie Theft picture descriptions produced by healthy elderly (n=45, mean age=67.6±8.7, 25 females (56%)) and young speakers (n=76, mean age=20±0.9, 35 females (46.1%)). Using the MAT annotation toolkit, we manually annotated non-overlapping spans of text that identified one of nine topics: six core to the picture, Stealing, Washing/Cleaning, Overflowing, Not Noticing, Helping, and Falling, and three non-core topics, Indoor, Outdoor, and Abstract. We obtained an inter-annotator of 88.52% across a 25 double annotated document subset comprising 525 classification pairs. The other files were annotated by one of the three annotators. After annotating the files, the number of each topic count was converted to counts per 100, controlling for the total number of topic tokens, and the results were compared by group using Wilcoxon signed-rank and chi-squared tests. Results: Elderly speakers (54.5±19.9%) produced core topics, which involved the characters in the picture, less frequently than young speakers (61.6±14.5%; W=1134.5, p=0.04), whereas they (45.5±19.9%) produced non-core topics, which were related to the background and outdoor scenery in the picture, more frequently than young speakers (38.4±14.5%; W=1836, p=0.04). Specifically, elderly speakers described the events of “kids stealing cookies” (W=1137.5, p=0.04), “girl helping the boy” (W=892.5, p=0.0005), “boy falling” (W=1041, p=0.01), “water overflowing” (W=1105.5, p=0.03) less frequently than young speakers. Elderly and young speakers did not differ in the number of unique topic types described (W=1476, p=0.95) or the total number of topic tokens described (W=1778.5, p=0.09). The most frequently first described event by both elderly and young speakers was an indoor-related topic, such as “this is a kitchen” (χ2=11.17, p=0.08). Conclusion: This study investigated how the two age groups differed in topic selection when describing pictures. Our results showed that elderly speakers described fewer core topics than younger speakers, demonstrating that the analysis of topic selection may be a useful tool for understanding the effect of aging in speakers’ linguistic behavior. In future work, we plan to train an automatic topic tagging system using this manually tagged dataset and apply the automatic system to the study of patients with neurodegeneration.

Topic Areas: Language Production, Meaning: Discourse and Pragmatics

SNL Account Login

Forgot Password?
Create an Account