Difference between revisions of "User:Francis Tyers/Perceptron"

From Apertium
Jump to navigation Jump to search
Line 10: Line 10:
==Example==
==Example==


Here is a worked example of a perceptron applied to the task of lexical selection. Lexical selection is the task of choosing a target translation <code>t</code> for a given source word <code>s</code> in a context <code>C</code> out of a set of possible translations <code>T</code>. A perceptron makes a classification decision for a single class, so we need to train a separate perceptron for each possible target word selection.
Here is a worked example of a perceptron applied to the task of lexical selection. Lexical selection is the task of choosing a target translation <math>t*</math> for a given source word <math>s</math> in a context <math>C</math> out of a set of possible translations <math>T</math>. A perceptron makes a classification decision for a single class, so we need to train a separate perceptron for each possible target word selection.

In the example,

* <math>s</math> = estació
* <math>T</math> = {season, station}
* <math>t*</math> = season


===Features===
===Features===

The features we will be working with are ngram contexts around the "problem word".


===Training data===
===Training data===
Line 19: Line 27:
! !!
! !!
|-
|-
|1:sec || season
|-
|1:de 2:el 3:any || season
|-
|-
| <code>_ sec</code> || 1
|1:de 2:tren || station
|-
|-
| <code>_ de el any</code> || 1
|1:humit || season
|-
|-
|1:de 2:televisió || station
| <code>_ de tren</code> || 0
|-
|-
|1:de 2:línia || station
| <code>_ de el línia</code> || 0
|-
|-
| <code>_ humit</code> || 1
|1:plujós || season
|-
|-
| <code>_ plujós</code> || 1
|-1:un 1:a || station
|-
|-
| <code>un _ a</code> || 0
|1:naval || station
|-
|-
|}
|}

Revision as of 19:58, 8 November 2014

A perceptron is a classifier that

The classifier consists of:

  • Binary features
  • Weights

Example

Here is a worked example of a perceptron applied to the task of lexical selection. Lexical selection is the task of choosing a target translation for a given source word in a context out of a set of possible translations . A perceptron makes a classification decision for a single class, so we need to train a separate perceptron for each possible target word selection.

In the example,

  • = estació
  • = {season, station}
  • = season

Features

The features we will be working with are ngram contexts around the "problem word".

Training data

_ sec 1
_ de el any 1
_ de tren 0
_ de el línia 0
_ humit 1
_ plujós 1
un _ a 0

Feature vector

Weight vector

Retrieved from "https://wiki.apertium.org/w/index.php?title=User:Francis_Tyers/Perceptron&oldid=50886"