Personal long-term memory aids

The prevalence and affordability of personal and environmental recording apparatuses are leading to increased documentation of our daily lives. This trend is bound to continue and it follows that academic, industry, and government groups are showing an increased interest in such endeavors for various purposes. In the present case, I assert that such documentation can be used to help remedy common memory problems. Assuming a long-term personal archive exists, when confronted with a memory problem, one faces a new challenge, that of finding relevant memory triggers. This dissertation examines the use of information-retrieval technologies on long-term archives of personal experiences towards remedying certain types of long-term forgetting. The approach focuses on capturing audio for the content. Research on Spoken Document Retrieval examines the pitfalls of information-retrieval techniques on error-prone speechrecognizer-generated transcripts and these challenges carry over to the present task. However, “memory retrieval” can benefit from the person's familiarity of the recorded data and the context in which it was recorded to help guide their effort. To study this, I constructed memory-retrieval tools designed to leverage a person’s familiarity of their past to optimize their search task. To evaluate the utility of these towards solving long-term memory problems, I (1) recorded public events and evaluated witnesses’ memory-retrieval approaches using these tools; and (2) conducted a longerterm memory-retrieval study based on recordings of several years of my personal and research-related conversations. Subjects succeeded with memory-retrieval tasks in both studies, typically finding answers within minutes. This is far less time than the alternate of re-listening to hours of recordings. Subjects’ memories of the past events, in particular their ability to narrow the window of time in which past events occurred, improved their ability to find answers. In addition to results from the memory-retrieval studies, I present a technique called “speed listening.” By using a transcript (even one with many errors), it allows people to reduce listening time while maintaining comprehension. Finally, I report on my experiences recording events in my life over 2.5 years. Thesis Supervisor: Walter Bender Title: Senior Research Scientist Doctoral Dissertation Committee Walter Bender Senior Research Scientist MIT Program in Media Arts and Sciences Christopher Schmandt Principal Research Scientist MIT Program in Media Arts and Sciences Rosalind W. Picard Associate Professor of Media Arts and Sciences MIT Program in Media Arts and Sciences