Documentation Index
Fetch the complete documentation index at: https://mintlify.com/AngelMoralesChazari/TautoTeacher-2.0/llms.txt
Use this file to discover all available pages before exploring further.
core.lgs (located at src/main/resources/logicscript/core.lgs) is the central data file for TautoTeacher’s NLP pipeline. It is loaded at runtime from the classpath by LgsCargador.cargarConDiagnostico("logicscript/core.lgs"), which means you can edit and redeploy it without recompiling any Java code. The file currently defines irregular-verb lemmas, morphological suffix rules, exclusion lists, and positional sentence patterns that cover the full range of propositional connectives taught in the course.
Every .lgs file must declare its schema version on the first non-comment, non-blank line:
The parser recognizes version as metadata and skips it. Future schema changes will increment this number. Version mismatches are not currently enforced, but recording the version makes it easy to diagnose incompatibilities when upgrading TautoTeacher.
lemma directive
Syntax:
lemma <form> -> <canonical>
lemma entry maps one inflected or irregular surface form to its base (canonical) form. Before the pipeline assigns a proposition symbol, BaseConocimiento.canonicalizarFragmento looks up each word in the lemma table; if a match is found, the canonical form is used instead. This ensures that llueve and llueva both map to the same symbol p = llover rather than creating two separate propositions.
When to add a lemma
Add a lemma when… | Do NOT add a lemma when… |
|---|
The verb has an irregular stem change in the present tense (apruebo → aprobar) | The verb is a regular -ar verb (estudio, estudia, estudian) — the morphological normalizer handles these |
The word is a fixed noun used as a proposition (gorra → gorra, paraguas → paraguas) | The infinitive is already the surface form being used |
The form involves a highly irregular conjugation (voy → ir, tengo → tener) | A lexrule sufijo rule already covers the suffix class |
A locution needs a single canonical label (solea → hacer_sol) | The form appears in a position that the excluir list already protects |
Real examples from core.lgs
lemma llueve -> llover
lemma llevo -> llevar
lemma apruebo -> aprobar
lemma salgo -> salir
lemma duerme -> dormir
lexrule directive
Syntax variants:
lexrule excluir <word> [<word> ...]
lexrule sufijo <suffix> infinitivo ar|er|ir
lexrule sufijo <suffix> heuristica primera_persona
lexrule directives feed NormalizadorMorfologico, which is invoked by BaseConocimiento as a second-priority step when no lemma entry matches. Rules are evaluated in declaration order — the first rule whose suffix matches the end of the input word wins.
excluir subtype
lexrule excluir registers a list of words that must never be treated as inflected verbs by the morphological engine. Without this protection, nouns like paraguas (ends in -as) could be incorrectly reduced to a phantom infinitive.
From core.lgs:
lexrule excluir gorra sombrero paraguas calor frio sol cielo nube nubes lluvia examen clase
sufijo ... infinitivo subtype
Strips the declared suffix from the end of a word and appends the infinitive-class termination (ar, er, or ir). For example, the rule lexrule sufijo aba infinitivo ar converts estudiaba → estudi + ar → estudiar.
Three representative rules from core.lgs:
lexrule sufijo aba infinitivo ar
lexrule sufijo emos infinitivo er
lexrule sufijo imos infinitivo ir
sufijo ... heuristica primera_persona subtype
The first-person singular present tense in Spanish ends in -o across all three conjugation classes, so the target class cannot be determined from the suffix alone. This rule type activates a stem-analysis heuristic that inspects the root vowel pattern to decide between -ar, -er, and -ir.
lexrule sufijo o heuristica primera_persona
estudio → estudiar, aprendo → aprender, and vivo → vivir without requiring individual lemma entries for each verb.
pattern directive
Syntax:
pattern <NAME> <token-sequence> => <ir-type> left=N right=M [mid=K]
pattern declares a named template that SemanticMapper tries to match against the token list produced by NaturalLexer. Patterns are tested in declaration order; the first matching pattern wins. When a pattern matches, the engine builds an IR node of the specified type, using the token positions indicated by left, right, and (where required) mid to locate the operand literals.
Token types
| Token | Matches |
|---|
si | The word “si” (conditional) |
entonces | ”entonces” |
y | ”y” (conjunction) |
o | ”o” (disjunction) |
literal | Any span of text that is not a keyword — the actual proposition text |
solo_si | The phrase “solo si” |
a_menos_que | The phrase “a menos que” |
si_y_solo_si | The phrase “si y solo si” |
siempre_que | The phrase “siempre que” |
cuando | ”cuando” |
en_caso_de_que | The phrase “en caso de que” |
IR output types
| IR type | Meaning | Notes |
|---|
imp | Implication (→) | Requires left, right |
and | Conjunction (∧) | Requires left, right |
or | Disjunction (∨) | Requires left, right |
equiv | Biconditional (↔) | Requires left, right |
imp_and | (left ∧ mid) → right | Requires left, mid, right |
imp_or | left → (mid ∨ right) | Requires left, mid, right |
imp_or_ant | (left ∨ mid) → right | Requires left, mid, right |
imp_and_cons | left → (mid ∧ right) | Requires left, mid, right |
imp_unless | ¬left → right (“unless”) | Requires left, right |
left, right, mid index rules
The indices are zero-based positions in the matched token sequence (including keyword tokens, not just literals). For example, in the sequence [si, literal, entonces, literal], position 1 is the first literal and position 3 is the second literal.
Real examples from core.lgs
pattern SI_ENTONCES si literal entonces literal => imp left=1 right=3
pattern CONJUNCION literal y literal => and left=0 right=2
pattern DISYUNCION literal o literal => or left=0 right=2
pattern EQUIVALENCIA literal si_y_solo_si literal => equiv left=0 right=2
pattern SI_CONJ_Y_ENTONCES si literal y literal entonces literal => imp_and left=1 mid=3 right=5
pattern SI_ENTONCES_DISY_CONS si literal entonces literal o literal => imp_or left=1 mid=3 right=5
pattern SOLO_SI literal solo_si literal => imp left=0 right=2
pattern A_MENOS_QUE literal a_menos_que literal => imp_unless left=0 right=2
pattern CONSECUENTE_SI_ANTECEDENTE literal si literal => imp left=2 right=0
Pattern matching is positional and exact — the number of tokens in the declared sequence must match the number of tokens produced by the lexer for that block precisely. If a sentence produces even one extra or missing token, no pattern will fire and the segment falls back to a bare atom. Always verify pattern matches by inspecting the pasosDeAnalisis trace or running the regression harness after adding a new pattern.
Extending core.lgs safely:
- Add
lemma lines for any irregular verb forms that appear in your course sentences and fail to canonicalize correctly.
- Add
pattern lines for new sentence structures (e.g. a new connective phrasing) that the current patterns don’t cover. Give each pattern a descriptive SCREAMING_SNAKE_CASE name.
- After any edit, run the regression harness (
java -cp out tautoteacher2.logicscript.LogicScriptRegressionHarness) to confirm that all 40 existing cases still pass. Add a new test case to the harness for every new sentence structure you introduce.