Eric Ringger

The Topical Guide is a web application that facilitates the discovery of topical patterns and trends that give valuable insight into large document collections. It relies on probabilistic topic models, such as LDA, to reveal topical content without human assistance. The Topical Guide is a general, interactive tool for browsing the entire output of a topic model analysis along with the analyzed corpus.

CCASH (Cost-Conscious Annotation Supervised by Humans) is a web-based annotation framework. It is designed to be an environment for evaluating state-of-the-art and experimental techniques for efficient annotation and also for applying those techniques to real world annotation projects. While designing CCASH we had our eye particularly on Active Learning; however other techniques such as feature labeling and incorporating rich prior knowledge could also be incorporated into CCASH without too… Show more

CCASH (Cost-Conscious Annotation Supervised by Humans) is a web-based annotation framework. It is designed to be an environment for evaluating state-of-the-art and experimental techniques for efficient annotation and also for applying those techniques to real world annotation projects. While designing CCASH we had our eye particularly on Active Learning; however other techniques such as feature labeling and incorporating rich prior knowledge could also be incorporated into CCASH without too much trouble. Show less

The automatic discovering of noteworthy patterns and trends in large document collections is the goal of our text mining projects. We are interested not only in identifying such patterns and trends but also in revealing them to human users through useful user interfaces. Our hypothesis is that the process of discovering meaningful patterns is best accomplished by cooperation between automatic methods and human expertise. Our work involves models for document clustering and for topic modeling… Show more

The automatic discovering of noteworthy patterns and trends in large document collections is the goal of our text mining projects. We are interested not only in identifying such patterns and trends but also in revealing them to human users through useful user interfaces. Our hypothesis is that the process of discovering meaningful patterns is best accomplished by cooperation between automatic methods and human expertise. Our work involves models for document clustering and for topic modeling. The interplay between such models and interactive user interfaces is an area of current investigation. Show less

Recovering high quality digital text from modern machine-printed document images using Optical Character Recognition (OCR) is nearly a solved problem. However, recovering high quality digital text from historical document images is significantly more challenging. Our document recognition project focuses on the latter problem. The work encompasses efforts to combine multiple OCR hypotheses using multi-sequence alignment methods and machine learning to select the best hybrid transcription. Such… Show more

Recovering high quality digital text from modern machine-printed document images using Optical Character Recognition (OCR) is nearly a solved problem. However, recovering high quality digital text from historical document images is significantly more challenging. Our document recognition project focuses on the latter problem. The work encompasses efforts to combine multiple OCR hypotheses using multi-sequence alignment methods and machine learning to select the best hybrid transcription. Such hypotheses can come from multiple OCR engines or from a single OCR engine on different inputs. The potential for transcription improvement is substantial.
Furthermore, we are interested in mining patterns from historical texts. The work includes examinations of the impact of errors in document recognition on the performance of various probabilistic topic models.
Show less

Linguistically annotated corpora have proven useful in many applications in Natural Language Processing and in the Digital Humanities. Time and money are often lacking for extensive human annotation, so how can we annotate most efficiently? Members of the ALFA project are investigating ways of reducing the cost of labeling corpora with the help of high quality automatic annotators:
A high quality machine annotator can typically handle the easy annotation decisions, greatly reducing… Show more

Linguistically annotated corpora have proven useful in many applications in Natural Language Processing and in the Digital Humanities. Time and money are often lacking for extensive human annotation, so how can we annotate most efficiently? Members of the ALFA project are investigating ways of reducing the cost of labeling corpora with the help of high quality automatic annotators:
A high quality machine annotator can typically handle the easy annotation decisions, greatly reducing annotation cost.
Machine annotators can be trained cheaply using active learning---asking humans to annotate data deemed especially useful by the system. One of our primary contributions so far has been to make the active learner sensitive to the predicted cost of annotation incurred by the expert. We are also in the process of testing active learning in a context with multiple human annotators.
We are working on creating flexible machine assistants that quickly adapt when annotators change their annotation scheme: the structure or conventions governing the way they annotate.
We are developing a web annotation framework (CCASH) to apply these ideas and others in user studies (Syriac User Study) and in a large-scale annotation effort. Show less