Python ULF Library

Python library for interfacing with and manipulating unscoped episodic logical forms (ULF), based on the original Common Lisp implementation by Gene Kim.

Additional information can be found at the ULF project page.

Dependencies

• transduction
• memoization

Summary

Install the package using pip install ulflib.

Import the package using the following line.

from ulflib import ulflib

Documentation

Lexical match predicates

The following match functions are made available for matching individual lexical items (intended for use with the transduction package). Refer to the ULF annotation guidelines for additional details on the ULF lexical categories.

Upon importing this package, each lexical match function is registered with the transduction package and can be accessed in transduction rules using the corresponding predicates, e.g., !lex-noun-p or *lex-noun-p. The negated versions of the predicates are also registered, e.g., !not-lex-noun-p.

LEX_PREDS = [
  lex_noun_p,
  lex_rel_noun_p,
  lex_function_p,
  lex_pronoun_p,
  lex_verb_p,
  lex_adjective_p,
  lex_p_p,
  lex_p_arg_p,
  lex_ps_p,
  lex_pq_p,
  lex_prep_p,
  lex_pp_p,
  lex_mod_a_p,
  lex_mod_n_p,
  lex_mod_p,
  lex_rel_p,
  lex_det_p,
  lex_coord_p,
  lex_aux_s_p,
  lex_aux_v_p,
  lex_aux_p,
  lex_number_p,
  lex_name_p,
  lex_adv_a_p,
  lex_adv_s_p,
  lex_adv_e_p,
  lex_adv_f_p,
  lex_adv_formula_p,
  lex_adv_p,
  lex_x_p,
  lex_yn_p,
  lex_gr_p,
  lex_sent_p,
  lex_tense_p,
  lex_aspect_p,
  lex_detformer_p,
  litstring_p,
  lex_equal_p,
  lex_set_of_p,
  lex_noun_postmod_macro_p,
  lex_np_postmod_macro_p,
  lex_noun_or_np_postmod_macro_p,
  lex_macro_p,
  lex_macro_rel_hole_p,
  lex_macro_sub_hole_p,
  lex_macro_hole_p,
  lex_hole_variable_p,
  lex_verbaux_p,
  lex_pasv_p,
  lex_possessive_s_p,
  lex_invertible_verb_p,
  lex_comma_p,
  lex_elided_p,
  surface_token_p,
]

Phrasal match predicates

The following match functions are made available for matching phrasal ULF categories:

PHRASE_PREDS = [
  noun_p,
  adj_p,
  adj_premod_p,
  adj_postmod_p,
  adv_a_p,
  adv_e_p,
  adv_s_p,
  adv_f_p,
  adv_p,
  mod_a_p,
  mod_n_p,
  mod_a_former_p,
  mod_n_former_p,
  pp_p,
  term_p,
  verb_p,
  pred_p,
  det_p,
  aux_p,
  tensed_aux_p,
  tensed_verb_p,
  sent_p,
  tensed_sent_p,
  sent_punct_p,
  sent_mod_p,
  ps_p,
  noun_reifier_p,
  verb_reifier_p,
  sent_reifier_p,
  tensed_sent_reifier_p,
  advformer_p,
  detformer_p,
  modformer_p,
  preposs_macro_p,
  relativized_sent_p,
  p_arg_p,
  voc_p,
]

General match predicates

The following additional (uncategorized) match predicates are also defined:

GEN_PREDS = [
  plur_term_p,
  plur_partitive_p,
  plur_noun_p,
  plur_lex_noun_p,
  pasv_lex_verb_p,
  unknown_p,
  postmod_p,
  postmod_adj_p,
  verb_arg_p,
  verb_or_tensed_verb_p,
  sent_or_sent_mod_p,
  sent_or_tensed_sent_p,
  phrasal_sent_op_p,
  type_shifter_p,
  prog_marker_p,
  perf_marker_p,
  aux_or_head_verb_p,
  noun_or_adj_p,
  invertible_verb_or_aux_p
]

Search

The following functions can be used to search for the heads of verb phrases, noun phrases, and adjective phrases within a ULF, respectively:

find_vp_head(vp)
find_np_head(np)
find_ap_head(ap)

Additionally, the following functions find and replace the heads with some given value sub:

replace_vp_head(vp, sub)
replace_np_head(np, sub)
replace_ap_head(ap, sub)

The following match functions are also defined (mostly used internally by the above functions):

SEARCH_PREDS = [
  marked_conjugated_vp_head_p,
  vp_head_p,
  np_postmodification_head_p,
  ap_premodification_head_p,
  ap_postmodification_head_p
]

Suffix

The following functions are defined for manipulating the suffix of a ULF lexical item:

suffix_for_type(x)
"""Return the suffix for the type. If none found, return the type."""

add_suffix(word, suffix)
"""Take a word string and a suffix and merge them together."""

suffix_check(x, suffix)
"""Check if a symbol has the given suffix."""

split_by_suffix(x)
"""Split a symbol by its suffix."""

has_suffix(x)
"""Check if a symbol has a suffix."""

strip_suffix(str)
"""Strips the suffix, marked with '.', from a string, e.g., man.n -> man."""

Macro

The following top-level functions are defined for processing macros in ULFs:

add_info_to_sub_vars(ulf)
"""Add types, pluralization, etc. to the variables *h for sub macros."""

add_info_to_relativizers(ulf)
"""Add pluralization, etc. to the relativizers in relative clauses."""

apply_sub_macro(ulf, fail_on_bad_use=False)
"""Apply a sub macro."""

apply_rep_macro(ulf, fail_on_bad_use=False)
"""Apply a rep macro."""

apply_qt_attr_macro(ulf)
"""Apply a qt_attr macro."""

apply_substitution_macros(ulf)
"""Apply all substitution macros: sub, rep, qt-attr."""

Scoping

The functions within the scoping module can be used to scope tense, quantifiers, and coordinated expressions within a ULF formula:

from ulflib import scoping
scoping.scope(ulf)

If you wish to scope only a particular type of element (e.g., only tense), a list of keywords can be provided as an optional argument (where each member is tense, quan, or coord):

from ulflib import scoping
scoping.scope(ulf, types=['tense'])

Scoping relies on the following rules for determining whether an unscoped element is accessible within an embedding expression. To be accessible, the unscoped element must not be embedded by:

1. A larger expression binding a variable occurring free in the unscoped element.
2. A sentence modifier other than not.
3. A nonsubsective predicate modifier (not, nearly, ...; however, past/pres generally escape from these contexts).
4. An already scoped quantifier, or already scoped tense operator.
5. An unscoped conjunction or disjunction of sub-expressions.
6. A verbal sub-expression of the ulf, e.g., a subordinate or relative or conjoined verbal clause.

We can recursively use these accessibility rules even when we're no longer looking at a wff -- e.g., if some part of a wff contains a verbal sub-wff, any unscoped elements in that sub-wff are inaccessible at the level of the wff; similarly for unscoped sub-wff conjunctions/disjunctions, and certain modified and already scoped elements. However, we need to distinguish top-level and embedded candidate extraction, because at the top level extraction of unscoped elements from a clause is not blocked.

We don't extract definites or indefinites from scope islands. For definites, this is logically unnecessary (if they pick out some entity not dependent on variables of embedding quantifiers) and for indefinites we can "raise" them, if necessary, either by a physical transformation (including the possibility of removing an argument of a Skolem function, if we Skolemize) or by equating them to an external entity.