Kneser-Ney smoothing with a correcting function for small data sets

Channel:
Microsoft Research
Subscribers:
344,000
Published on ● Video Link: https://www.youtube.com/watch?v=G8VBIEik6T0


Duration: 26:13
1,995 views
6

We present a technique which improves the Kneser- Ney smoothing algorithm on small data sets for bigrams and we develop a numerical algorithm which computes the parameters for the heuristic formula with a correction. We give motivation for the formula with correction on a simple example. Using the same example we show the possible difficulties one may run into with the numerical algorithm. Applying the algorithm to test data we show how the new formula improves the results on cross-entropy.



Other Videos By Microsoft Research


2016-09-06I-Room - Intelligent Collaborative Spaces for Emergency Response [1/3]
2016-09-06Query Lower Bounds for Matroids via Group Representations
2016-09-06New Algorithmic Ideas for Search, Ads, and Recommendations. [1/2]
2016-09-06A Family of License Languages
2016-09-06Approximation Algorithms for Discrete Stochastic Optimization Problems
2016-09-06Virtual Coupling Schemes for Position Coherency in Networked Haptic Virtual Environments
2016-09-06Dynamic Mechanism Design [1/2]
2016-09-06Design Thinking and Design Research [1/3]
2016-09-06Predicting Secret Keys via Branch Prediction
2016-09-06Pricing Games in Networks
2016-09-06Kneser-Ney smoothing with a correcting function for small data sets
2016-09-06The Limiting Shape of Internal DLA with Multiple Sources
2016-09-06The Case for Opportunistic Communication
2016-09-06Building natural language parsers
2016-09-06Coding Techniques for Data-Storage Systems
2016-09-06Behind the Code with Mohsen Agsen [1/5]
2016-09-06Critical percolation on finite graphs
2016-09-06Programming with Time Using Graphical Data Flow
2016-09-06Design and development of a content-based music search engine
2016-09-06How America is Losing Its Innovation Edge, Why it Matters and What We Can Do to Get it Back
2016-09-06Recurrence of the Simple Random Walk Path


Tags:
microsoft research