The key features of the <term> system </term> include : ( i ) Robust efficient <term> parsing </term> of <term> Korean </term> ( a <term> verb final language </term> with <term> overt case markers </term> , relatively <term> free word order </term> , and frequent omissions of <term> arguments </term> ) .
The results of this experiment , along with a preliminary analysis of the factors involved in the decision making process will be presented here .
tech,9-1-H01-1049,bq <term> Listen-Communicate-Show ( LCS ) </term> is a new paradigm for <term> human interaction with data sources </term> .
We integrate a <term> spoken language understanding system </term> with <term> intelligent mobile agents </term> that mediate between <term> users </term> and <term> information sources </term> .
Using <term> LCS-Marine </term> , tactical personnel can converse with their logistics system to place a supply or information request .
We have demonstrated this capability in several field exercises with the Marines and are currently developing applications of this <term> technology </term> in <term> new domains </term> .
The <term> oracle </term> knows the <term> reference word string </term> and selects the <term> word string </term> with the best <term> performance </term> ( typically , <term> word or semantic error rate </term> ) from a list of <term> word strings </term> , where each <term> word string </term> has been obtained by using a different <term> LM </term> .
Actually , the <term> oracle </term> acts like a <term> dynamic combiner </term> with <term> hard decisions </term> using the <term> reference </term> .
The method amounts to tagging <term> LMs </term> with <term> confidence measures </term> and picking the best <term> hypothesis </term> corresponding to the <term> LM </term> with the best <term> confidence </term> .
The method amounts to tagging <term> LMs </term> with <term> confidence measures </term> and picking the best <term> hypothesis </term> corresponding to the <term> LM </term> with the best <term> confidence </term> .
We describe our use of this approach in numerous fielded <term> user studies </term> conducted with the U.S. military .
We take a selection of both <term> bag-of-words and segment order-sensitive string comparison methods </term> , and run each over both <term> character - and word-segmented data </term> , in combination with a range of <term> local segment contiguity models </term> ( in the form of <term> N-grams </term> ) .
I show that the <term> performance </term> of a <term> search engine </term> can be improved dramatically by incorporating an approximation of the <term> formal analysis </term> that is compatible with the <term> search engine </term> 's <term> operational semantics </term> .
<term> Techniques for automatically training </term> modules of a <term> natural language generator </term> have recently been proposed , but a fundamental concern is whether the <term> quality </term> of <term> utterances </term> produced with <term> trainable components </term> can compete with <term> hand-crafted template-based or rule-based approaches </term> .
<term> Techniques for automatically training </term> modules of a <term> natural language generator </term> have recently been proposed , but a fundamental concern is whether the <term> quality </term> of <term> utterances </term> produced with <term> trainable components </term> can compete with <term> hand-crafted template-based or rule-based approaches </term> .
The method combines <term> domain independent acoustic models </term> with off-the-shelf <term> classifiers </term> to give <term> utterance classification performance </term> that is surprisingly close to what can be achieved using conventional <term> word-trigram recognition </term> requiring <term> manual transcription </term> .
In our method , <term> unsupervised training </term> is first used to train a <term> phone n-gram model </term> for a particular <term> domain </term> ; the <term> output </term> of <term> recognition </term> with this <term> model </term> is then passed to a <term> phone-string classifier </term> .
The <term> model </term> is designed for use in <term> error correction </term> , with a focus on <term> post-processing </term> the <term> output </term> of black-box <term> OCR systems </term> in order to make it more useful for <term> NLP tasks </term> .
Overall <term> summarization </term> quality of the proposed <term> system </term> is state-of-the-art , with guaranteed <term> grammaticality </term> of the <term> system output </term> due to the use of a <term> constraint-based parser/generator </term> .
In this paper , we show how <term> training data </term> can be supplemented with <term> text </term> from the <term> web </term> filtered to match the <term> style </term> and/or <term> topic </term> of the target <term> recognition task </term> , but also that it is possible to get bigger performance gains from the <term> data </term> by using <term> class-dependent interpolation </term> of <term> N-grams </term> .
hide detail