Ilana Heintz

Graduate Student
Dept. of Linguistics
The Ohio State University
204 Oxley Hall
1712 Neil Ave.
Columbus, OH 43210
email: bromberg at ling dot osu dot edu
what's in a name?
my CV

• Language Modeling

  • Using statistically- and morphologically-informed techniques to reduce the out-of-vocabulary problem in Arabic. Dissertation research. Use the templatic morphology of Arabic to figure out where stems and affixes are, what the short vowels might be, and how to predict the next word in a sequence. Language modeling for eventual use in speech recognition, especially cross-dialectal.

• Automatic Speech Recognition I.

  • Exploring the use of joint versus independent models of phonological features as input to conditional random fields for phone recognition. Presented as a short paper at HLT-NAACL 2007.

• Automatic Speech Recognition II.

  • Using dimension reduction techniques from information retrieval to find new features for phone recognition. The new features should complement phonological features or phonetic features that we study because we know they are important in discriminating speech sounds. Submitted to ICASSP 2008.

• Automatic Speech Recognition III.

  • Combining multiple types of features into a Conditional Random Fields phone recognition system. Use features derived from support vector machines, neural networks, and TDNNs. Mix of binary and n-ary features. CRFs handle redundancy well, therefore they are a good choice in conducting this study. Which features complement each other? Which are only redundant? Part of the ASAT project, research performed in conjunction with Georgia Tech & Rutgers Universities. Presented at Interspeech 2007.

• Varied Techniques in Arabic Information Retrieval.

  • Which stemming methods and retrieval algorithms, and in what combination, work best for Information Retrieval, considering the specific morphological characteristics of Arabic? How can complex morphological information be better incorporated into a stemming model, and how does that interact with existing retrieval models?

• Systematicity in the Arabic Lexicon.

  • A study that finds a correspondence between the phonetic and semantic characteristics of frequent Arabic words. Presented at the Arabic Linguistics Symposium, 2005. Proceedings in press.

• Using information theoretic techniques to predict the epenthetic vowel.

  • With Beth Hume. Is the quality of the epenthetic vowel in a given language determined, at least in part, by its frequency and distribution?

• Ordering Sentences According to Topicality.

  • A short study pertaining to Natural Language Generation. Can LSA be used to determine the correct order of sentences of a short, single-topic article? (Answer: not very well, at least not LSA alone.) Presented at the Midwest Computational Linguistics Colloquium in 2006.