A person talks about a familiar face, and their speech becomes a mirror reflecting the workings of memory. In patients with temporal lobe epilepsy, this mirror sometimes grows cloudy: words come more slowly, phrases break off, pauses hang in the air. Doctors have long heard such complaints, but standard tests, where you have to name a picture or recall a list of words, often come back normal. The researchers decided to look into the mirror itself, not at its reflection on paper.
They invited 51 patients with temporal lobe epilepsy and 30 healthy volunteers. Each was shown photographs of famous people and asked to say whatever came to mind. The accounts were recorded and then run through a neural network that converted sound into text and separated the speaker's voice from the interviewer's remarks. From the text, lexical features were extracted: how many words in total, how often rare words appeared. From the audio, the length of phrases and pauses.
The results were surprising: the groups did not differ in vocabulary. Patients used just as many rare words as healthy participants. But speech rate gave them away completely. Patients' phrases were shorter (on average 5.54 seconds versus 6.59), their pauses were shorter (0.61 versus 0.67 seconds), and they fell silent more often (10.81 pauses per minute versus 9.29). The mirror was not just growing cloudy; it was starting to tremble.
Then an important detail emerged: lexical measures correlated with object-naming tests, while pause frequency correlated with the accuracy of biographical memories. That is, different facets of speech reflect different cognitive functions. Vocabulary is a shop window displaying the names of things. Rate and pauses are the mechanics by which memory pulls those names off the shelves.
Previously, such subtle impairments could only be noticed in a long conversation with an experienced neurologist. Now they can be measured automatically, from a recording of a conversation. This opens the way to remote monitoring: a patient talks about familiar faces at home, and an algorithm tracks how their speech changes month by month. The mirror of memory can be hung in every smartphone.
Word frequency was measured on the Zipf scale: a word with frequency 3 occurs about once per million words, and one with frequency 4 about once per ten thousand. Rare words turned out to be a marker of vocabulary.
Not published on arXiv. Here are its page, the PDF and the formalization: you can check it yourself, not only read our account.