Methods and systems for predicting learning curve for statistical machine translation system
AssigneeXerox Corporation
InventorsPrasanth Kolachina, Nicola Cancedda, Marc Dymetman, Sriram Venkatapathy
Life of this patent
Estimating the term…
Abstract
The disclosed embodiments relate to a system and method for predicting the learning curve of an SMT system. A set of anchor points are selected. The set of anchor points correspond to a size of a corpus. Thereafter, a gold curve or a benchmark curve is fitted based on the set of anchor points to determine the BLEU score. Based on the BLEU score and a set of parameters associated with the first set of anchor points, a confidence score is computed.
Claims26
A computer implemented method of training a prediction model to estimate a learning curve indicative of training of a statistical machine translation (SMT), the computer implemented method comprising:
- selecting a first set of anchor points, wherein each of the first set of anchor points corresponds to a different number of monolingual segments, wherein the first set of anchor points are selectable by a user;
- computing a weight vector for each of the first set of anchor points based, at least in part, on a first set of parameters associated with each of the first set of anchor points;
- and
- computing a first confidence score for each of the first set of anchor points based on the weight vector.
A computer implemented method of estimating a learning curve indicative of training of a statistical machine translation (SMT), the computer implemented method comprising:
- selecting a second set of anchor points, wherein each of the second set of anchor points corresponds to a different number of bi-lingual segments, wherein the second set of anchor points are selectable by a user;
- fitting a second curve based on each of the second set of anchor points;
- and
- estimating a second set of parameters of the second curve to obtain the learning curve, wherein the second set of parameters include a value at an origin, a slope at the origin, and an asymptotic value.
A computer implemented method of training a prediction model to estimate a learning curve associated with training of a statistical machine translation (SMT), the computer implemented method comprising:
- selecting a first set of anchor points, wherein each of the first set of anchor points corresponds to a different number of monolingual segments, wherein the first set of anchor point are selectable by a user;
- selecting a second set of anchor points, wherein each of the second set of anchor points corresponds to a different number of bi-lingual segments, wherein the second set of anchor point are selectable by the user;
- fitting a second curve based on the second set of anchor points;
- extending the second curve to the first set of anchor points;
- computing a second confidence score for each of the first set of anchor points based, at least in part, on one or more parameters associated with the second curve;
- computing a weight vector for each of the first set of anchor points based on a first set of parameters associated with each of the first set of anchor points;
- and
- computing a first confidence score for each of the first set of anchor points based on the weight vector.
A prediction model server for estimating a learning curve indicative of training of a SMT, the system comprising:
- an anchor point selection module configured to select at least one of a first set of anchor points or a second set of anchor points;
- a curve fitting module configured to fit a second curve based on each of the second set of anchor points;
- an estimation module configured to extrapolate the second curve to the first set of anchor points;
- a scoring module configured to:
- compute a second confidence score for each of the first set of anchor points based, at least in part, on one or more parameters associated with the second curve;
- and
- compute a weight vector for each of the first set of anchor points based on a first set of parameters associated with each of the first set of anchor points;
- and
- wherein, the curve fitting module is further configured to determine a second set of parameters associated with the learning curve based, at least in part, on the second confidence score and the weight vector.
Dependencies are inferred from the claim text.