joshua.decoder.NbestMinRiskReranker Class Reference

Collaboration diagram for joshua.decoder.NbestMinRiskReranker:

List of all members.

Classes
class	RankerResult
class	RankerTask
Public Member Functions
	NbestMinRiskReranker (boolean produceRerankedNbest, double scalingFactor)
String	processOneSent (List< String > nbest, int sentID)
double	computeExpectedGain (int curHypLen, HashMap< String, Integer > curHypNgramTbl, List< HashMap< String, Integer >> ngramTbls, List< Integer > sentLens, List< Double > nbestProbs)
Static Public Member Functions
static void	computeNormalizedProbs (List< Double > nbestLogProbs, double scalingFactor)
static double	computeExpectedGain (String curHyp, List< String > nbestHyps, List< Double > nbestProbs)
static void	main (String[] args) throws IOException
Package Functions
void	getGooglePosteriorCounts (List< HashMap< String, Integer >> ngramTbls, List< Double > normalizedProbs, HashMap< String, Double > posteriorCountsTbl)
double	computeExpectedLinearCorpusGain (int curHypLen, HashMap< String, Integer > curHypNgramTbl, HashMap< String, Double > posteriorCountsTbl)
Package Attributes
boolean	produceRerankedNbest = false
double	scalingFactor = 1.0
final PriorityBlockingQueue < RankerResult >	resultsQueue
Static Package Attributes
static int	bleuOrder = 4
static boolean	doNgramClip = true
static boolean	useGoogleLinearCorpusGain = false
Static Private Member Functions
static double	addInLogSemiring (double x, double y, int addMode)

---

Detailed Description

this class implements: (1) nbest min risk (MBR) reranking using BLEU as a gain funtion.

This assume that the string is unique in the nbest list In Hiero, due to spurious ambiguity, a string may correspond to many possible derivations, and ideally the probability of a string should be the sum of all the derivataions leading to that string. But, in practice, one normally uses a Viterbi approximation: the probability of a string is its best derivation probability So, if one want to deal with spurious ambiguity, he/she should do that before calling this class

Author:

Zhifei Li, zhife.nosp@m.i.wo.nosp@m.rk@gm.nosp@m.ail..nosp@m.com

Version:

$LastChangedDate$

---

Constructor & Destructor Documentation

joshua.decoder.NbestMinRiskReranker.NbestMinRiskReranker	(	boolean	produceRerankedNbest,
		double	scalingFactor
	)

Here is the caller graph for this function:

---

Member Function Documentation

static double joshua.decoder.NbestMinRiskReranker.addInLogSemiring	(	double	x,
		double	y,
		int	addMode
	)		[static, private]

Here is the caller graph for this function:

double joshua.decoder.NbestMinRiskReranker.computeExpectedGain	(	int	curHypLen,
		HashMap< String, Integer >	curHypNgramTbl,
		List< HashMap< String, Integer >>	ngramTbls,
		List< Integer >	sentLens,
		List< Double >	nbestProbs
	)

Here is the call graph for this function:

Here is the caller graph for this function:

static double joshua.decoder.NbestMinRiskReranker.computeExpectedGain	(	String	curHyp,
		List< String >	nbestHyps,
		List< Double >	nbestProbs
	)		[static]

Here is the call graph for this function:

double joshua.decoder.NbestMinRiskReranker.computeExpectedLinearCorpusGain	(	int	curHypLen,
		HashMap< String, Integer >	curHypNgramTbl,
		HashMap< String, Double >	posteriorCountsTbl
	)		[package]

Here is the call graph for this function:

Here is the caller graph for this function:

static void joshua.decoder.NbestMinRiskReranker.computeNormalizedProbs	(	List< Double >	nbestLogProbs,
		double	scalingFactor
	)		[static]

based on a list of log-probabilities in nbestLogProbs, obtain a normalized distribution, and put the normalized probability (real value in [0,1]) into nbestLogProbs

Here is the call graph for this function:

Here is the caller graph for this function:

void joshua.decoder.NbestMinRiskReranker.getGooglePosteriorCounts	(	List< HashMap< String, Integer >>	ngramTbls,
		List< Double >	normalizedProbs,
		HashMap< String, Double >	posteriorCountsTbl
	)		[package]

Here is the caller graph for this function:

static void joshua.decoder.NbestMinRiskReranker.main

(

String[]

args

)

throws IOException [static]

Here is the call graph for this function:

String joshua.decoder.NbestMinRiskReranker.processOneSent	(	List< String >	nbest,
		int	sentID
	)

value in baselineScores will be changed to normalized probability

TODO: zhifei: now the re-ranking takes O(n^2) where n is the size of the nbest. But, we can significantly speed up this (leadding to O(n)) by first estimating a model on nbest, and then rerank the nbest using the estimated model.

Here is the call graph for this function:

Here is the caller graph for this function:

---

Member Data Documentation

int joshua.decoder.NbestMinRiskReranker.bleuOrder = 4 [static, package]

boolean joshua.decoder.NbestMinRiskReranker.doNgramClip = true [static, package]

boolean joshua.decoder.NbestMinRiskReranker.produceRerankedNbest = false [package]

final PriorityBlockingQueue<RankerResult> joshua.decoder.NbestMinRiskReranker.resultsQueue [package]

Initial value:

      new PriorityBlockingQueue<RankerResult>()

double joshua.decoder.NbestMinRiskReranker.scalingFactor = 1.0 [package]

boolean joshua.decoder.NbestMinRiskReranker.useGoogleLinearCorpusGain = false [static, package]