Cognitive Model of Time and Analysis of Natural Language Texts Naidenova, X.A., Garina, M.I.
-
Upload
yolanda-tippery -
Category
Documents
-
view
224 -
download
1
Transcript of Cognitive Model of Time and Analysis of Natural Language Texts Naidenova, X.A., Garina, M.I.
Cognitive Model of Time and Analysis of Natural Language Texts
Naidenova, X.A., Garina, M.I.
Cognitive model of time
Cognitive model of time includes:
The units of time (year, month, spring, minute); the time intervals and their properties: the beginning, the end,
duration, without beginning (opened in the past), without end (opened in the future), consisting of points;
Environment: nearest past, nearest future (about noon, soon after the beginning/the end, toward the evening);
Various relations between units and intervals: coincidence, contact, presedence, going after, intersection, inclusion, remoteness into the past/future time;
Degree of relations: the measures of remoteness, intersection and so on;
Comparison relations of interval duration: longer, less for long, shorter and so on;
Uncertain (fuzzy) relations: considerably later, once, early in the morning and so on;
Main cognitive constructions
Event (can be empty) time interval; Time interval ::= (the beginning --- the end) / unit of
time/ a set of units of time; a moment of time is the particular case of interval;
The beginning ::= the date/event; the end::= the date/event;
Interval has the duration; Event is associated with time interval. But the very
moment of time can be an event :«September began», «Days go»;
Time interval can be expressed via some events, for example, “at dawn, to the first volleys of artillery”, “long before the first sun rays”.
The Basic Cognitive Model of Time Periodicity
Event Time of event Duration
Methods
The date Unit of time Interval
The beginning The end
Event Event
The date The date
Unit of time Unit of time
Interval Interval
Time of event
Duration
Time of event
Duration
The Property of Cognitive Model of Time Obviously, this diagram is recursive, i.e., an
event is associated with the time interval, and a time interval can be expressed via some events, for example:
“at dawn”, “to the first volleys of artillery”, “long before the first sun rays”.
MethodsComputation Rules for DURATION: а) as the
difference between the end and the beginning with the precise time markers (the dates); б) as the time interval between events: since <EVENT1> to <EVENT2>; в) as a set of time units (900 days).
Rules for demonstrating the truthfulnes of different relations between time intervals: sequence (which is earlier, which is later, which will be, which is already past), simultaneity, inclusion into one and the same time interval, intersection, contact and so on.
Methods as meta-knowledge Methods are natural rules for analyzing the relationships
between time units/intervals; Methods include both computations and implicative
assertions of the general kind: E1,T.end < E2,T.begin
E1 R E2, where E1 and E2 are events, E.T is interval associated with event E, T.end and T.begin are the end
and beginning of interval T, respectively, and R is the
precedence relation; «T.begin ≤ T.end; T.end ≥ T.begin»;
Properties of Events
Events can be indeterminate (fuzzy) in the time (considerably later, once upon a time);
Events can flow in the time rapidly, slowly; Periodic events can be frequent or rare; The temporary properties of events can be
both objective and subjective (estimated).
It is still necessary to consider
A set of events can be associated with only one time interval;
An event can be expressed by both only one word and a set of proposals (maybe only one proposal);
An event and the time interval associated with it can be in different proposals;
It is still necessary to considerIt is possible that an active agent (including
temporary moment) cannot be determined without the aid of referential relation ;
There are the events attached (by default) to the time intervals (dawn, sunset, school-leaving ball, dinner, supper, breakfast, the beginning of workday).
The link of cognitive model of time with NLs The cognitive model of time does not depend on
language, but it is tuned into different natural languages. For this goal, the following levels of natural language
are considered: lexical, morphological, and syntactic ones.
Questions of constructing a translator of the cognitive model of time (its elements and the generalized assertions (meta - knowledge)) into language expressions for a given natural language and vice versa are examined.
Translator can be built as a trained system that learns by specially constructed phrases.
The Lexical Level A special case of time interval
is the name of time unit Example: TI = {century, year, month,
twenty-four hours, the morning, day, evening, the night, January, February, March, April, May, June, minute, second, winter, summer…}
Relationships between lexical units of NL
Classification («is-a»), composition («concsist-of»), part-whole, occurring in cycles, inclusion, sequence:
Year is (winter, spring, summer, autumn); Occurring in cycles: winter of one year follows
after autumn of previous year; Sequence: spring is after winter; «Part-Whole»: minute is a part of hour; Composition: twenty-four hours consist of
night and day;
Afternoon
Night Morning
Evening
January
February
March
April
May
June July
August
September
October
November
December
Year Month Time of dayDay
Explanations to the diagram The relation of classification is shown with
the aid of triangle connections while the relation of composition - by pointer with the rhomb.
If the relation of composition is determined between the intervals of upper level, then it is determined between the interval- descendants (for example, June consists of twenty-four hours).
Specific dimensionality can be determined only for the connections of the lower level (it cannot be said how many twenty-four hours year generally consists of, month generally consists of, but it can be said, how many twenty-four hours leap year consists of, month consists of, January consists of, etc)
Lexical level of time relationships. Table: « Time of event » (for Russian Language) from «Multidimensionality of the time ». Elkin, S. V. et al.)
1 Event 2 Duration of the event, durability
3 Long duration, short duration, short-term, momentary, transitory
Repetition of the event Process consisting of several different events
Event series
Multiple repetition of one and the same event
Daily, every weekIt is quarterly, monthly,Yearly
Once Once only, it is single-time
Time before the event Before, in advance, in good time, previously, before the appointed time, it is preliminary, it is premature, on the threshold of, it is earlier than, long before, thus far not, not in a long time, recently, as long as, the day before
Time after the event Later, afterward, it is later, then, after, hence, hence-forward, forth, in future, from now on, after all, immediately afterward, further, when
Association of the event with the time
In one’s life (from birth), originally, while
Indeterminate time
Affirmative time: sometimes, someday, in the course of time, then, once;
Negative time: never
Temporary pretexts
The approximate classification of Russian temporary pretexts has been given by Kreydlin, G.E. “Time through the prism of temporary pretexts” // The logical analysis of the language: Language and time. M. — 1997. - 352 p.
Relation Pretext Example Temporary marker, the event
Simultaneity
Extent
Duration
For,
During
For entire trip he said nothing
The time interval is attached to the event (trip)
Precedence Approximately We awaited approximately to midnight
Temporary marker: midnight;
Time: the indeterminate half-interval.
Syntactic Level
Relation Structure Syntactic diagram
Role in the sentence
Going after Immediately afterward<action>/<event>, <event>
Adverb with the pretext «afterward» <action>/<event>, <event>
Adverbial modifier
of time. Example: Immediately after wedding and parting words of parents.
Inclusion Including <the date>
Verbal Adverb <the date>
Adverbial modifier of time.
Example: Including 2010
Text processing Text processing system consists of Cognitive models of time and events that oriented to a
given domain application and the goals of text processing;
Translater that is adjusted to a given NL; Block of plausible (commonsense) reasoning which infers
consequences from established temporary relations between events in the text by means of meta-knowledge of cognitive models;
Dialogue Syntactical Analyzer for a given NL; Block of control or operational subsystem of the
translater.
CognitiveModel
Of Time
Cognitive Model
Of Event
Text
Plausible Inferrence
Translator, adjusted to NL
Syntactical Analyzer
Base of extracted events and time relations between
themControl
Text processing
Text interacts with the Translator and Syntactical Analyzer;
As a result, events, their time moments or interval are extracted, and then the conclusions about temporary relations between the events are inferred.
The work of Translator (1)
Translator first searches for the supporting (key) words (temporary markers), which are connected with the expression of time in the text.
Then translator, using lexical and syntactic models, attempts to determine the events, associated with the chosen time markers.
If it is necessary, then turning to Syntactic Analyzer follows.
The work of translator (2)
Translator can repeatedly be turned first to the text or first to the cognitive model, then to the syntactic analyzer in order purposefully to search for the required (according to the rules of cognitive model) linguistic constructions.
Formation of the Base of Events
Hypotheses about the events and the temporary supporting moments are erected as the list of the possible facts, extracted from the text.
The Base of Events is filled up with the copies of events with their time characteristics.
The work of Block of Plausible Reasoning
The block of plausible reasoning derives all consequences of the discovered facts (events, their properties, the relations between them).
Example of extracting events and time of events from the text
An example of the text analysis has been given from the narrative of V. Nekrasov “The entrenchments of Stalingrad”.
This example shows what we will have as a result of the event-temporary text analysis with the use of cognitive models of time and events
Sentence 1
The number of sentence
Event Time interval Inferred information
1 I do not recollect
Autumn Autumn consists of «September, October, November».
Sentence 2
The number of sentence
Event Time interval Inferred information
2 September passed
September It precedes “October”; consequently, “began October”.
Time interval «September» is an event. Predicate is expressed by the verb of passed time, whose semantics speaks that the time interval is finished, it left into the past. It is derived from the cognitive model of time that October goes after September, next month of autumn.
Sentence 3
The number of sentence
Event Time interval Inferred information
3 Event 1:
Fish laps in the Volga;
Event 2:
Circles disperse over the surface of water.
In the mornings Each day in the morning; October;Autumn.
Sentence 6
The number of sentence
Event Time interval Inferred data
6 It is tender = Left shore of the Volga is tender
At dawn;
To the first volleys of the artillery;
At dawn = early in the morning;
To (before) the first volleys of the artillery;
Event = the first volleys of the artillery;
6 First volleys of the artillery
At dawn; Early in the morning; October; Autumn.
Sentence 8
The number of sentence
Event Time interval Inferred information
8 Event: Х lasts; Х
= the fogFor a while For a while;
Early in the morning.
In this proposal there is no subject and therefore there is no acting agent. We establish it with the aid of the reference (analysis of the previous proposal).
Sentence 9
The number of sentence
Event Time interval Inferred information
9 Long-range gun
shootsLong before the first sun rays
9 Event: First sun rays
At dawn Early in the morning;
October; Autumn.
In proposal 9 there is no explicit indication of moment of time. But adverbial modifier of time «before the first sun rays» is associated with dawn, and dawn – with morning. That’s why we extract the event «the first sun rays» and associate it with « early in the morning ».
Sentences 11 and 12
The number of sentence
Event Time interval Inferred information
11 The day begins
The day Day comes after morning;
The beginning of the day.
12 The « frame » appears
At seven o’clock
Seven hours of the morning; Beginning of the day.
Sentence 19
The number of sentence
Event Time interval Inferred information
19 It will determine Entire day
19 It = the first ten of aircrafts
Entire day Entire day = from the morning to the evening
Sentence19 «It will determine entire day» requires the returning to the previous sentence in order to assiciate the word « it » with « the first ten of aircrafts ». This action requires the complete syntactic analysis of sentence 18.
Sentence 20
The number of sentence
Event Time interval Inferred information
20 Event 1:
They will bury those killed;
Event 2: repair the damaged guns;
Event 3 :
dig new slots and mud-huts;
Entire night After day;
From the evening to the morning;
From the sunset to the dawn;
Explanations to sentence 20
In proposal 20, we separate the fragment connected with the keywords “entire night”.
Subject and predicate determine event “we learn”. However time interval relates to direct object: «We learn, in what section entire night they will bury those killed, repair the damaged machine guns and guns, dig new slots and mud-huts».
Some observations
The dialogue between the cognitive model of time, the translator and the syntactic analyzer occurs.
The completeness and the accuracy of the extracted knowledge depends on the cognitive model of time, its completeness and accuracy.
We rested on the time marks. But it is possible to make events to be key markers.
What What is necessary to make? is necessary to make? ((11))
1.1. To build Cognitive Model To build Cognitive Model of Time as completely as of Time as completely as possible including all possible including all abstract elements, abstract elements, relations and methods. It relations and methods. It is also necessary to take is also necessary to take into account the into account the uncertainty of time uncertainty of time intervals.intervals.
What What is necessary to make? is necessary to make? ((2)2)1.1. To build the Cognitive Model of To build the Cognitive Model of
EventEvent. . 2.2. Cognitive elements of model are: FACTCognitive elements of model are: FACT, , PROCESSPROCESS, ,
ACTION,ACTION, RESULT, SUBJECTRESULT, SUBJECT,, OBJECT, PLACE of OBJECT, PLACE of EVENT, EVENT, TIME of EVENT,TIME of EVENT, CAUSAL LINKS between CAUSAL LINKS between EVENTS, PROPERTY of OBJECT (SUBJECT). EVENTS, PROPERTY of OBJECT (SUBJECT).
3.3. To refine this model through the To refine this model through the knowledge of a concrete domain knowledge of a concrete domain application (business, finances).application (business, finances).
To build model of all possible events in a given domain To build model of all possible events in a given domain application with a mechanism of plausible inference application with a mechanism of plausible inference over this models of events.over this models of events.
What What is necessary to make? is necessary to make? ((3)3) After the work on the cognitive models it After the work on the cognitive models it
is possible to be turned to creating the is possible to be turned to creating the translator. It will be necessary to translator. It will be necessary to determine in sentences the parts of determine in sentences the parts of speech (subject, predicate, object, speech (subject, predicate, object, adverbial modifiers). It is necessary to adverbial modifiers). It is necessary to recognize different kinds of sentencesrecognize different kinds of sentences..
The main thing, that the analysis of The main thing, that the analysis of sentences is governed with the aid of the sentences is governed with the aid of the cognitive models. The conclusions are cognitive models. The conclusions are done at the level of cognitive models and done at the level of cognitive models and the concrete queries to the syntactic the concrete queries to the syntactic analyzer can be formed via cognitive analyzer can be formed via cognitive models.models.
Thanks for the attention