Rani Nelken

Arabic Diacritization Using Weighted Finite-State Transducers

Arabic is usually written without short vowels and additional diacritics, which are nevertheless important for several applications. We present a novel algorithm for restoring these symbols, using a cascade of probabilistic finite-state transducers trained on the Arabic treebank, integrating a word-based language model, a letter-based language model, and an extremely simple morphological model. This combination of probabilistic methods and simple linguistic information yields high levels of accuracy.

Mining Wikipedia Revision Histories for Improving Sentence Compression

A well-recognized limitation of research on supervised sentence compression is the dearth of available training data. We propose a new and bountiful resource for such training data, which we obtain by mining the revision history of Wikipedia for sentence compressions and expansions. Using only a fraction of the available Wikipedia data, we have collected a training corpus of over 380,000 sentence pairs, two orders of magnitude larger than the standardly used Ziff-Davis corpus. Using this new-found data, we propose a novel lexicalized noisy channel model for sentence compression, achieving improved results in grammaticality and compression rate criteria with a slight decrease in importance.

Abbreviated text input using language modeling

We address the problem of improving the efficiency of natural language text input under degraded conditions (for instance, on mobile computing devices or by disabled users), by taking advantage of the informational redundancy in natural language. Previous approaches to this problem have been based on the idea of prediction of the text, but these require the user to take overt action to verify or select the systems predictions. We propose taking advantage of the duality between prediction and compression . We allow the user to enter text in compressed form, in particular, using a simple stipulated abbreviation method that reduces characters by 26.4%, yet is simple enough that it can be learned easily and generated relatively fluently. We decode the abbreviated text using a statistical generative model of abbreviation, with a residual word error rate of 3.3%. The chief component of this model is an n -gram language model. Because the systems operation is completely independent from the users, the overhead from cognitive task switching and attending to the systems actions online is eliminated, opening up the possibility that the compression-based method can achieve text input efficiency improvements where the prediction-based methods have not. We report the results of a user study evaluating this method.

A Modal Interpretation of the Logic of Interrogation

We propose a novel interpretation of natural-language questions using a modal predicate logic of knowledge. Our approach brings standard model-theoretic and proof-theoretic techniques from modal logic to bear on questions. Using the former, we show that our interpretation preserves Groenendijk and Stokhofs answerhood relation, yet allows an extensional interpretation. Using the latter, we get a sound and complete proof procedure for the logic for free. Our approach is more expressive; for example, it easily treats complex questions with operators that scope over questions. We suggest a semantic criterion that restricts what natural-language questions can express. We integrate and generalize much previous work on the semantics of questions, including Beck and Sharvits families of subquestions, non-exhaustive questions, and multi-party conversations.

On the Ontological Nature of Syntactic Categories in Categorial Grammar

We address the ontological nature of syntactic categories in categorial grammar. Should we interpret categories as being a model of some actual grammatical reality, or are they merely part of an empirical model, one that attempts to capture the range of data, but without making any ontological commitments. This distinction, which is not often made, is important for determining the goals of formal grammar research. We evaluate this question within the context of a particular grammatical phenomenon, the modeling of premodifier ordering. We compare the categorial grammar treatment of this phenomenon with empirical statistical approaches to the same problem. We show that the whereas both models are equally empirically adequate, the statistical model is more generalizable and learnable, and thus advantageous as a realistic model.