How does it work? Lines are first analyzed for conformity to the SALT guidelines. Lines that meet criteria for SALT analysis (they may not be interrupted, abandoned, nor contain incomprehensible speech) are analyzed token by token. Analysis is performed using machine learning techniques. First, each token is classified as morphologically simplex or complex. Suffixes to complex words are then identified—particular care is taken to distinguish between the noun plural suffix (S) and the 3sg. active indicative suffix (3S), and between the possessive/"Saxon genitive" enclitic (Z) and the contracted form of is ('S)—and the root is identified. Finally, a finite-state tranducer is used to stem the inputs to derive the original stem form (e.g., dying is mapped to DIE/ING). The system is described in more detail in:

Kyle Gorman, Steven Bedrick, Géza Kiss, Eric Morley, Rosemary Ingham, Metrah Mohammad, Katina Papadakis, and Jan P.H. van Santen (2015). Automated morphological analysis of clinical language samples. In Proc. Computational Linguistics and Clinical Psychology Workshop, in press. [PDF].

What technologies are used? The system is written in Python 3 and uses the Flask web framework and the Thrax grammar development tools. The classifiers used are based on the averaged perceptron.

Where is the sample transcript from? It is based off of a transcript published in:

J. Weizenbaum. 1966. ELIZA: A computer program for the study of natural language communication between man and machine. Communications of the ACM 9(1): 36-45.

AutoSALT was developed by Kyle Gorman, Steven Bedrick and colleagues (including Emily Prud'hommeaux and Géza Kiss) at the Center for Spoken Language Understanding, part of the Institute on Development and Disability at the Oregon Health & Science University. AutoSALT development was funded in part by the National Institute on Deafness and Other Communication Disorders of the National Institutes of Health under awards R01DC007129-01 and R01DC012033, and by Autism Speaks under Innovative Technology for Autism Grant 2407. The content is solely the responsibility of the authors and does not necessarily represent the official views of the granting agencies or any other individual.

AutoSALT does not store SALT transcripts you submit for any reason.