Elamur Groupantstranslation-meta-codecs
The groupant beside models/translation-meta-codecs.groupant in the Mova repository: what the code computes, what it never does, how to go back from its outputs, its invariants, cost and checks.
Facts
Implementation material. A Mova code model describes this code at one source revision: partial models, finite checks and their stated scope. It is not a reviewed general exposition, and its labels are shown as written, never raised.
Components
| component | label | where |
|---|---|---|
| approximant | partial | forward |
| ast | not materialized | |
| composition | partial | composition |
| cost | symbolic source derived | cost |
| descriptant | present | descriptant |
| invariant | partial | invariant |
| tests | see source local evidence | checks |
Obernants
The typed relations around the forward model: the behaviour complement (what the code never does), the reverse reconstruction (every input compatible with an output) and specialised views.
| id | type | label | where |
|---|---|---|---|
behavior-complement | behavior_complement | partial source derived model | behavior-complement |
reverse-reconstruction | reverse_reconstruction | partial source derived model | reverse-reconstruction |
owner-signature | representation_transport | partial source derived model | owner-signature |
head-coordinate | coordinate_transport | partial source derived model | head-coordinate |
logical-gate | logical_multiset_gate | partial source derived model | logical-gate |
normal-versus-residual | residual_conditioned_identity | partial source derived model | normal-versus-residual |
Model text
descriptant
translation-meta-codecs: wave32
Hebrew Reader train owns meta tokens through IBM then Strong override; owners feed HebEncoder train. meta MTok ctx_keys feed profile and HebEncoder; Fw/Rule remap their heads. Logic consumes event concepts via Lexicon and LCS. Strong is optional alignment input, not a translation oracle.
This source-bound partial group organizes existing Mova code. Native evidence and excluded effects are recorded in source sidecars and the root wave32 report. Registration is not a universal proof or whole-system runtime claim.
model
translation-meta-codecs
forward
Meta tokens and lexicon provide shared representation; profile and Hebrew encoder choose context chunks, IBM+Strong supply Hebrew ownership, Fw/Join/Reorder and Rule transform tokens, Logic compares mapped event facts. These are tagged paths, not one mandatory chain.
invariant
Valid 1-based heads/root0 and finite representable counters; metadata projection and retained residuals remain explicit. Deterministic ordered maps and strict comparisons fix ties on admitted finite domains.
cost
LCS O(nm); IBM Isum target(source+1)*log(row) plus rows; token transforms up to quadratic/cubic head and subtree scans; four-level context maps include key bytes. Real pipeline cost is sum along selected path, not all sibling helpers.
composition
Hebrew Reader train owns meta tokens through IBM then Strong override; owners feed HebEncoder train. meta MTok ctx_keys feed profile and HebEncoder; Fw/Rule remap their heads. Logic consumes event concepts via Lexicon and LCS. Strong is optional alignment input, not a translation oracle.
domain
Small synthetic tokens, initialized tables, finite nonnegative weights, valid coordinates; Rule Drop acyclic; LCS <=65535 and counts/arithmetic representable.
limitations
Real annotator/morph loading, corpus filesystem parsing, full training/evaluation and teacher paths remain unexecuted. Residual exactness is diagnostic reconstruction, not semantic understanding.
checks
- meta::tests::the_logic_gate_on_merges
- meta::tests::lexicon_tsv_back
- meta::tests::lcs_and_gaps
- meta::wave32::metadata_loss_and_lcs_ties
- profile::tests::typography_of_quotes_and_possessives
- profile::tests::the_logic_gate_on_overrides
- profile::wave32::contexts_votes_and_ties
- hebrew::wave32::gloss_chunks_and_owned_alignment
- heb_enc::tests::hebrew_from_the_meta_world_and_back
- ibm::tests::learns_a_toy_dictionary
- ibm::wave32::zero_prior_and_normalization
- fwt::tests::a_fronted_part_moves_after_its_root
- fwt::tests::drops_and_inserts
- logic::tests::every_kind_of_change_is_seen
- logic::tests::contradictions_and_their_control
- rules::tests::drop_reattaches_and_a_rule_that_fixes_gains
- strong::tests::strong_words_and_tokens
- strong::wave32::window_and_numeric_boundaries
catalog
behavior-complement
Valid head-remapping transforms do not retain deleted coordinates; exact residual gates do not entail exact normal translation. Logical-word equality does not imply equal propositions.
reverse-reconstruction
Compatible antecedents include lost surface spellings, metadata/switches, dropped tokens and hidden owner assignments; uniqueness requires full stored scripts and source-side parameters.
owner-signature
Reader owners map each meta token to optional Hebrew index; HebEncoder emits only first encounter per owner, losing later assignments.
head-coordinate
MTok uses1-based heads/root0; each structural transform must renumber positions after dropping/reordering, while span retains source coordinates.
logical-gate
Delta gate compares sorted logical_words of bare concept and chunk, preserving duplicates and normalizing selected numerals. Event comparison is separate: permutations preserve verb-multiset gaps, while LCS pairing (including repeated-verb ties) can change contradictions and distance.
normal-versus-residual
Hebrew measure computes ordinary text exact/chrf before diagnostic target residual reconstruction; gate_ok can hold for empty learned encoder.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/models/translation-meta-codecs.groupant/