User:Francis Tyers/Apertium 4 - Revision history

Francis Tyers: /* Data creation and curation */

2020-06-27T12:12:36Z

Data creation and curation

2020-06-27T12:11:04Z

Engine

2020-06-25T11:20:50Z

2020-06-24T16:13:41Z

Research

2020-06-24T11:58:34Z

2020-06-24T07:52:13Z

Research

2020-06-24T00:13:45Z

Data creation

2020-06-24T00:13:31Z

Data creation

2020-06-23T23:54:37Z

2020-06-22T13:56:07Z

@@ Line 41: / Line 41: @@
 * Per-word lexical selection rule extraction
 * Integrate training into the make/test cycle, to avoid annotated data getting stale.
+* Get rid of the incubator/nursery/staging/etc. and move to a system of qualifying non-released language pairs by more objective means: coverage, diagnostics, WER, entries, etc.
 == End user ==

@@ Line 29: / Line 29: @@
 * Per session state, this could be stored in something like a special blank that could be updated. It might contain things like domain, etc.
 * [[Incorporating guessing into Apertium]] (no word left untranslated!)
+* [[Rethinking tokenisation in the pipeline]] support for languages that don't write spaces
 ; Neural system

@@ Line 6: / Line 6: @@
 * Make the different parts of the engine code more coherent in terms of repository (GitHub) modules
 * Better test coverage and testing of existing (particularly newer) modules
+* Good continuous integration, both of ML-based modules and rule-based ones.
 == Linguistic data ==
@@ Line 33: / Line 34: @@
 * [[apertium-neural]]: There should be a basic NMT implementation that functions in the Apertium ecosystem (C++,autotools,bash,apy,html-tools) for communities that want to build their own NMT systems and still take advantage of our ecosystem. We should be a one-stop shop for MT for marginalised langs.
-== Data creation ==
+== Data creation and curation ==
 * Automatic multiword extraction using parallel corpora
 * Recursive rule extraction
 * Per-word lexical selection rule extraction
+* Integrate training into the make/test cycle, to avoid annotated data getting stale.
 == End user ==

← Older revision		Revision as of 16:13, 24 June 2020
Line 50:		Line 50:

	* Neural MT benefits from decoupling TL generation from SL analysis. So that things generated are fluent, even if they don't correspond to the input. RBMT often has incoherent output because we don't take into account the target language. There has been research into using TLMs with RBMT, but mostly for N-best list reranking. Could we do something more clever? Unsupervised neural rewriting of RBMT output(?)		* Neural MT benefits from decoupling TL generation from SL analysis. So that things generated are fluent, even if they don't correspond to the input. RBMT often has incoherent output because we don't take into account the target language. There has been research into using TLMs with RBMT, but mostly for N-best list reranking. Could we do something more clever? Unsupervised neural rewriting of RBMT output(?)

	[[User:Khannatanmai\|Tanmai Khanna]] ([[User talk:Khannatanmai\|talk]]) Isn't that also one of the disadvantages of neural models with strong language models that prefer fluency over adequacy? At least in RBMT we can guarantee not giving up adequacy for fluency.

@@ Line 27: / Line 27: @@
 * Fully functional recursive transfer
 * Per session state, this could be stored in something like a special blank that could be updated. It might contain things like domain, etc.
+* [[Incorporating guessing into Apertium]] (no word left untranslated!)
 ; Neural system

← Older revision		Revision as of 07:52, 24 June 2020
Line 49:		Line 49:

	* Neural MT benefits from decoupling TL generation from SL analysis. So that things generated are fluent, even if they don't correspond to the input. RBMT often has incoherent output because we don't take into account the target language. There has been research into using TLMs with RBMT, but mostly for N-best list reranking. Could we do something more clever? Unsupervised neural rewriting of RBMT output(?)		* Neural MT benefits from decoupling TL generation from SL analysis. So that things generated are fluent, even if they don't correspond to the input. RBMT often has incoherent output because we don't take into account the target language. There has been research into using TLMs with RBMT, but mostly for N-best list reranking. Could we do something more clever? Unsupervised neural rewriting of RBMT output(?)

			[[User:Khannatanmai\|Tanmai Khanna]] ([[User talk:Khannatanmai\|talk]]) Isn't that also one of the disadvantages of neural models with strong language models that prefer fluency over adequacy? At least in RBMT we can guarantee not giving up adequacy for fluency.

@@ Line 4: / Line 4: @@
 == Software engineering ==
-* Make the different parts of the engine code more coherent in terms of modules
+* Make the different parts of the engine code more coherent in terms of repository (GitHub) modules
 * Better test coverage and testing of existing (particularly newer) modules

@@ Line 18: / Line 18: @@
 * Better support for Unicode in lttoolbox.
 * Use embeddings for morphological disambiguation and lexical selection
-* Pass the surface form until transfer (to allow modules to look up surface form embeddings)
+* [[Surface forms in the pipe]]: Pass the surface form until transfer (to allow modules to look up surface form embeddings)
 * Retire the HMM tagger
 * We currently have at least 3-4 classifiers (HMM, Perceptron, MaxEnt, ...). This is unecessary. We should have a single discriminative classifier for all classification needs (MLP?) modules should be frontends to it.
@@ Line 40: / Line 40: @@
 * Format handling
-* [[user-based chunking]]: Better treatment of "no-translate"
+* [[User-based chunking]]: Better treatment of "no-translate"
 * User dictionaries
 * Better handling of code-switching/mixed texts and informal text.