Apertium system architecture

The Apertium RBMT system uses a "pipeline", with individual stages performing specific tasks before that stage's output is passed along to the next stage. The page provides an overview of the pipeline, information about how the data in each stage is stored, and an example of data being passed through the pipeline to create an output translation.

The pipeline

The stages

Linguistic data

stage		introduced	filenames	mode	documentation
morphological tagger		2004	—	xxx-yyy-tagger, xxx-tagger	—
morphological analysis		2004	apertium-xxx.xxx.lexc and apertium-xxx.xxx.twol and apertium-xxx.xxx.twoc, OR apertium-xxx.xxx.dix	xxx-yyy-morph, xxx-morph
morphological disambiguation		2004, 2008	apertium-xxx.xxx.rlx	xxx-yyy-disam, xxx-disam	Constraint Grammar
discontiguous multiword assembly (optional)		2017	apertium-xxx-yyy.xxx-yyy.lsx	xxx-yyy-autoseq	Apertium separable
lexical transfer		2004	apertium-xxx-yyy.xxx-yyy.dix	xxx-yyy-biltrans	Bilingual dictionary
lexical selection		2012	apertium-xxx-yyy.xxx-yyy.lrx	xxx-yyy-lex	Lexical selection
anaphora resolution (optional)		2019, in progress	apertium-xxx-yyy.xxx-yyy.arx	xxx-yyy-anaphora	Anaphora Resolution Module
pre-transfer			—	xxx-yyy-pretransfer	—
shallow structural transfer	chunker	2006	apertium-xxx-yyy.xxx-yyy.t1x	xxx-yyy-chunker	Contributing to an existing pair#Adding structural transfer (grammar) rules
	interchunk	2006	apertium-xxx-yyy.xxx-yyy.t2x	xxx-yyy-interchunk
	postchunk	2006	apertium-xxx-yyy.xxx-yyy.t3x	xxx-yyy-postchunk
recursive structural transfer		2019, in progress	apertium-xxx-yyy.xxx-yyy.rtx	xxx-yyy-rectransfer	Apertium-recursive
discontiguous multiword disassembly (optional)		2017	apertium-xxx-yyy.yyy-xxx.lsx	xxx-yyy-revautoseq	Apertium separable
morphological generation		2004	apertium-yyy.yyy.lexc and apertium-yyy.yyy.twol and apertium-yyy.yyy.twoc, OR apertium-yyy.yyy.dix	xxx-yyy-dgen or xxx-yyy-gener or xxx-yyy-generador
post-generation		2004	apertium-yyy.post-yyy.dix	xxx-yyy-pgen	Post-generator

Apertium-internal

deformatter, reformatter

Example translation at each stage

Note that this example includes what all modules would do, even though the English-Spanish pair does not currently use all of them.

Input text

John said he took the big plant out to the yard.

Morphological analyzer

^John/John<np><ant><m><sg>$ ^said/say<vblex><past>/say<vblex><pp>$ ^he/prpers<prn><subj><p3><m><sg>$ ^took/take<vblex><past>$ ^the/the<det><def><sp>$ ^big/big<adj><sint>$ ^plant/plant<n><sg>/plant<vblex><inf>/plant<vblex><pres>/plant<vblex><imp>$ ^out/out<adv>/out<pr>$ ^to/to<pr>$ ^the/the<det><def><sp>$ ^yard/yard<n><sg>$^./.<sent>$

Morphological disambiguator

^John<np><ant><m><sg>$ ^say<vblex><past>$ ^prpers<prn><subj><p3><m><sg>$ ^take<vblex><past>$ ^the<det><def><sp>$ ^big<adj><sint>$ ^plant<n><sg>$ ^out<adv>$ ^to<pr>$ ^the<det><def><sp>$ ^yard<n><sg>$^./.<sent>$

Discontiguous multiword processing

^John<np><ant><m><sg>$ ^say<vblex><past>$ ^prpers<prn><subj><p3><m><sg>$ ^take# out<vblex><past>$ ^the<det><def><sp>$ ^big<adj><sint>$ ^plant<n><sg>$^./.<sent>$

Lexical transfer

^John<np><ant><m><sg>/John<np><ant><m><sg>$ ^say<vblex><past>/decir<vblex><past>$ ^prpers<prn><subj><p3><m><sg>/prpers<prn><tn><p3><m><sg>$ ^take# out<vblex><past>/sacar<vblex><past>$ ^the<det><def><sp>/el<det><def><GD><ND>$ ^big<adj><sint>/grande<adj><mf>$ ^plant<n><sg>/planta<n><f><sg>/fábrica<n><f><sg>/maquinaria<n><f><sg>$ ^to<pr>/a<pr>$ ^the<det><def><sp>/el<det><def><GD><ND>$ ^yard<n><sg>/patio<n><m><sg>$^.<sent>/.<sent>$

Lexical selection

^John<np><ant><m><sg>/John<np><ant><m><sg>$ ^say<vblex><past>/decir<vblex><past>$ ^prpers<prn><subj><p3><m><sg>/prpers<prn><tn><p3><m><sg>$ ^take# out<vblex><past>/sacar<vblex><past>$ ^the<det><def><sp>/el<det><def><GD><ND>$ ^big<adj><sint>/grande<adj><mf>$ ^plant<n><sg>/planta<n><f><sg>$ ^to<pr>/a<pr>$ ^the<det><def><sp>/el<det><def><GD><ND>$ ^yard<n><sg>/patio<n><m><sg>$^.<sent>/.<sent>$

Anaphora resolution

^John<np><ant><m><sg>/John<np><ant><m><sg>/$ ^say<vblex><past>/decir<vblex><past>/$ ^prpers<prn><subj><p3><m><sg>/prpers<prn><tn><p3><m><sg>/John<np><ant><m><sg>$ ^take# out<vblex><past>/sacar<vblex><past>/$ ^the<det><def><sp>/el<det><def><GD><ND>/$ ^big<adj><sint>/grande<adj><mf>/$ ^plant<n><sg>/planta<n><f><sg>/$ ^to<pr>/a<pr>/$ ^the<det><def><sp>/el<det><def><GD><ND>/$ ^yard<n><sg>/patio<n><m><sg>/$^.<sent>/.<sent>/$

Structural transfer

^John<np><ant><m><sg>$ ^decir<vblex><ifi><p3><sg>$ ^que<cnjsub>$ ^sacar<vblex><ifi><p3><sg>$ ^el<det><def><f><sg>$ ^planta<n><f><sg>$ ^grande<adj><mf><sg>$ ^a<pr>$ ^el<det><def><m><sg>$ ^patio<n><m><sg>$^.<sent>$

Morphological generator

John dijo que sacó la planta grande ~a el patio.

Post-generator

John dijo que sacó la planta grande al patio.

See also

• Modes