A Vietnamese Text-based Conversational Agent

Natural language question analysis component is the first component in any question answering systems.. In our experiment, the open-domain systems focus on retrievingand ranking related

1 Introduction 1

1.1 A Semantic Approach for Question Analysis 1

1.2 A Vietnamese Text-based Conversational Agent 2

1.3 Thesis Organisation 3

2 Literature review 4 2.1 Text-based conversational agents 4

2.1.1 Using keywords for pattern matching 4

2.1.2 Using the sentence similarity measure for pattern matching 7

2.2 FrameScript Scripting Language 9

2.3 Question answering systems 12

3 Our Question Answering System Architecture 15 3.1 Vietnamese Question Answering System 15

3.1.1 Natural language question analysis component 16

3.1.1.1 Intermediate representation of an input question 16

3.1.1.2 Question analysis 17

3.1.2 Answer retrieval component 18

3.2 Using FrameScript for question analysis 19

3.2.1 Preprocessing module 19

3.2.2 Syntactic analysis module 20

3.2.3 Semantic analysis module 22

4 Text-based Conversational Agent for Vietnamese 24 4.1 Overview of architecture 24

4.2 Determining separate contexts 25

4.3 Identifying hierarchical contexts 27

vi

TABLE OF CONTENTS vii

5.1 Experimental results

for Vietnamese text-based conversational agent 29

5.2 Question Analysis for English 31

5.3 Discussion 33

A Scripting patterns

2.1 O’Shea et al.’s conversational agent framework 7

2.2 Aqualog’s architecture 14

3.1 Architecture of our question answering system 16

3.2 Architecture of the natural language question analysis componentusing FrameScript 19

4.1 Architecture of our Vietnamese text-based conversational agent 25

viii

List of Tables

4.1 Script examples of “subjects” 26

4.2 Transformations between contexts 27

4.3 Order of transformation rules 28

4.4 Ordered transformation between contexts 28

5.1 List of transformations among contexts 30

5.2 Unsatisfying analysis 30

5.3 The satisfied degree of students 31

5.4 Number of rules corresponding with each question-structure type 31

5.5 Number of rules with conditional responses 32

5.6 Number of questions corresponding with each question-structure type 32 5.7 Error results 32

ix

x

Chapter 1

Introduction

The goal of question answering systems is to give answers to the user’s questionsinstead of ranked lists of related documents as used by most current search engines(Hirschman and Gaizauskas, 2001) Natural language question analysis component

is the first component in any question answering systems This component creates

an intermediate representation of the input question, which is expressed in naturallanguage, to be utilized in the rest of the system

For the task of translating a natural language question into an explicit ate representation of the complexity in question answering systems, most publishedworks so far use rule-based approach to the best of our knowledge Some questionanswering systems such as (Lopez et al., 2007; Phan and Nguyen, 2010) manuallydefined a list of sequence rule structures to analyze questions However, in theserule-based approaches, manually creating the rules is error-prone and expensive intime and effort

intermedi-In this thesis, we present an approach to return an intermediate representation

of question via FrameScript scripting language (McGill et al., 2003) Natural guage questions will be transformed into intermediate representation elements whichinclude the construction type of question, question class, keywords in question andsemantic constraints between them Framescript allows users to intuitively writerules to directly extract the output tuple

lan-1

1.2 A Vietnamese Text-based Conversational Agent

A text-based conversational agent is a program allowing the conversational actions between human and machine by using natural language through text Thetext-based conversational agent uses scripts organized into contexts comprising hier-archically constructed rules The rules consist of patterns and associated responses,where the input is matched based on patterns and the corresponding responses aresent to user as output

inter-We focus on the analysis of input text in building a conversational agent cently, the input analysis over user’s statements have been developed following twomain approaches: using keywords (ELIZA (Weizenbaum, 1983), ALICE (Wallace,

Re-2001), ProBot (Sammut, 2001)) and using similarity measures (O’Shea et al., 2010;

Graesser et al., 2004; Traum, 2006) for pattern matching The approaches usingkeywords usually utilize a scripting language to match the input statements, whilethe other approaches measure the similarity between the statements and patternsfrom the agent’s scripts

In this thesis, we introduce a Vietnamese text-based conversational agent chitecture on a specific knowledge domain Our system aims to direct the user’sstatement into an appropriate context The contexts are structured in a hierarchy ofscripts consisting of rules in FrameScript language (McGill et al.,2003) In addition,our text-based conversational agent was constructed to integrate in a Vietnamesequestion answering system Our conversational agent provides not only informationrelated to user’s statement but also provides necessary knowledge to support ourquestion answering system when it is unable to find an answer

ar-The knowledge domain we used to build our text-based conversational agent isthe academic regulation at Vietnam National University, Hanoi (VNU) The aca-demic regulation book helps students to know the course programs, the regulation ofexaminations, the discipline at VNU However, most students don’t prefer readingthe academic regulation book Therefore, our contribution creates an interactionchannel to offer the necessary information to students Once students give theirstatements that they are interested in the academic regulation, our text-based con-versational agent responses these statements by providing the related information indetail Furthermore, our conversation agent also interacts with students by offeringthe option to ask if students want to know other information

Literature review

In this chapter, we review related works using text-based approaches for tional agent (CA) Section 2.1describes the approaches constructing rules to matchuser’s natural language utterances in the ways of using keywords (in section 2.1.1)and using a sentence similarity measure (in section 2.1.2) In addition, section 2.2

conversa-covers the basic knowledge background about FrameScript scripting language that

we have been working on, while section 2.3 presents reviews about the questionanswering systems driving specific-domains

2.1.1 Using keywords for pattern matching

ELIZA (Weizenbaum,1983) was one of the earliest text-based conversational agentsbased on a simple pattern matching by using the identification of keywords fromuser’s statement Then ELIZA transforms the user’s statement to an appropriaterule and generates output response The procedure that ELIZA responds to an userinput to give an appropriate output consists of five steps

• Identify the important keywords appearing in user’s statement

• Define some minimal context within which selected keyword occurs

• Determine an appropriate transformation rule

• Generate the responses when the input text contained no keywords

4

2.1 Text-based conversational agents 5

• Provide a facilitate editing for scripts on the script writing level

Transformation rules are used to serve decomposing a data string according tocertain criteria and reassembling a decomposed string according to certain assemblyspecifications Therefore, the input are analyzed based on the decomposition rulestriggered by keywords, and responses are generated against the reassembly rulesassociated with selected decomposition rules For example, encountering the inputsentence:

“It seems that you like me”

this sentence is decomposed into the four parts:

(1) It seems that (2) you (3) like (4) me

by using the decomposition rule:

(0 YOU 1 ME)

The associated response might then be:

“What makes you think I like you”

by using the reassembly rule:

(WHAT MAKES YOU THINK I 3 YOU)

An integer 0 in the decomposition rule will match more words and a non-zero integer

“n” appearing in a decomposition rules indicates that exactly “n” words will bematched, while an integer 3 in the above reassembly rule shows that the third part

of the decomposed sentence is inserted in its place to reply the input sentence Ifeach word is defined in a dictionary of keywords by scanning an input sentence fromleft to right, then only decomposition rules containing that keyword need to be tried

An ELIZA script consists mainly of a set of list structures as following:

ALICE (Wallace, 2001) is a text-based conversational agent as chat robot lizing an XML language called Artificial Intelligence Markup Language (AIML).AIML files consist of category tags representing rules; each category tag contains apair of pattern and template tag The entire category is stored in a tree The systemsearches the pattern according with an user input by using depth-first search in thetree, and produces the appropriate template as a response For example, a categorybelow:

<topic name=“MOVIES”>

<category>

<pattern>YES</pattern>

<that>DO YOU LIKE ROMANTIC MOVIES</that>

<template>What is your favourite romantic movie?</template>

</category>

<category>

<pattern>YES</pattern>

<that>DO YOU LIKE ACTION MOVIES</that>

<template>What is your favourite action movie?</template>

</category>

When the client says yes, the program must discover the robot’s previous utterance

If the robot asked “Do you like romantic movies?”, the response sent to reply is

“What is your favourite romantic movie?”

AIML is clever and simple, and easy for implementation and a good start forbeginners writing simple bots However, it is difficult to write and debug more

2.1 Text-based conversational agents 7

discriminating patterns, and it is very hard to know all the transformations availablebecause AIML depends on self-modifying the input

Sammut (Sammut, 2001) presented a text-based CA called ProBot that is able

to extract data from users ProBot’s scripts are typically organized into cal contexts consisting of a number of organized rules to handle unexpected inputs.Concurrently, McGill et al (McGill et al.,2003) derived from ProBot’s scripts (Sam-mut,2001) build the rule system in FrameScript scripting language (in section2.2).FrameScript (McGill et al., 2003) provides for the rapid prototyping of conversa-tional interfaces and simplifies the writing of scripts

hierarchi-2.1.2 Using the sentence similarity measure

for pattern matching

O’Shea et al (O’Shea et al.,2008,2010) proposed a text-based conversational agentframework (shown in figure2.1) using semantic analysis All patterns in scripts arethe natural language sentences The pattern matching uses a sentence similaritymeasure (Li et al., 2006) to calculate the similarity between sentences from scriptsand user input The highest ranked sentence is selected and its associated response

is sent as output

Figure 2.1: O’Shea et al.’s conversational agent framework

Scripts used in framework consist of contexts relating to a specific topic of sation Each context contains one or more rules, and each rule uses “s” to represent

conver-a nconver-aturconver-al lconver-anguconver-age sentence conver-and “r” to represent conver-a response stconver-atement For exconver-ample,considering a following rule:

<Rule_01>

s: I’m a student

r: Which university do you study?

With a user’s statement:

“I am a master student” or

“I am a phd student”

This input and the natural language sentences from the scripts are received in order

to send the sentence similarity measure Then sentence similarity measure calculates

a firing strength for each sentence pair to rank the sentences In this above example,the highest ranked sentence selected is “I’m a student” and its associated responsesent to user is “Which university do you study?”

The advantages of using a sentence similarity measure for pattern-matching isthat rule structures are simplified and reduced in size and complexity By contrast,this approach can’t retrieve some information from an input to insert into responselike using keywords for presented section 2.1.1

Graesser et al (Graesser et al., 2004) presented a conversational agent calledAUTOTUTOR matching input statements in the use of Latent Semantic Analysis.Traum (Traum, 2006) adapted the effective question answering characters (Leuski

et al.,2006) to build a conversational agent also employing Latent Semantic Analysisfor pattern matching

2.2 FrameScript Scripting Language 9

FrameScript (McGill et al., 2003) is a language for creating multi-modal user terfaces It employs from Sammut’s Probot (Sammut, 2001) to enable rule-basedprogramming, frame representations and simple function evaluation The Frame-Script scripting language also proposes a set of tools to represent knowledge andinteracting with users and external devices

in-Each script in FrameScript (McGill et al.,2003) includes a list of rules matchedagainst user input and used to give the appropriate response Rules are grouped intoparticular contexts of the form: context_name :: rule_set The scripting rules in theFrameScript language consist of patterns and responses with the form:

alter-[response 1 | response 2 | | another response],

where each response is given in turn every time the pattern is matched and thesequence repeats when the last response is output Alternatives have the form sur-rounded by braces:

{response 1 | response 2 | | another response},

in which any response may be chosen randomly for user output

In addition, responses utilize the ‘#’ to perform some action such as ing the current context For example, #goto(a_script) transforms a conversation orinteraction from one context to another Similarly, ‘∧’ is used to perform actions, ex-

chang-cept that when the following expression is evaluated it is inserted into response notthrown away And some response expression may be dependent on some conditionsholding true in the constructed form below:

Furthermore, some pattern elements create a numbered match component when

a pattern matches These component are segments of the input that can be referred

to in a response using ‘∧’ Pattern elements that identify match components arewild-card (*, and ∼), alternatives and non-terminals When ‘∧’ is followed by aninteger then the numbered pattern component associated with that integer is placed

in the output response Encountering an example as following:

{My name is | I’m} * ==>

[ Hello∧2 How old are you? ]

I am <Number> years old ==>

[∧(∧1 <= 20) –> Are you a student?

| How do you do? ]

The transcript of dialogue is shown below illustrating the above example:User: My name is X

CA: Hello X How old are you?

User: I am 19 years old

CA: Are you a student?

An input received from user is given to a domain in order to ensure that the input

is matched against the correct scripts Script can be registered as topic in a domain

to become the current script and process the input When a script is registered as

a topic, the domain uses the script’s trigger to determine whether or not an inputactivates that topic If a topic doesn’t have a trigger, any input will activate it.When a topic’s trigger matches the input, it becomes the current context and thecurrent topic

2.2 FrameScript Scripting Language 11

Example ::

domain example

trigger{* {Hi | hi | Hello | hello} *}

* {Hi | hi | Hello | hello} * ==> [Hi there!]

When writing complex scripts where scripts have similar behaviours, FrameScript

is possible to use inheritance to enable rule to be shared between scripts Moreover,FrameScript allows defining failsafes for scripts A failsafe is another script whoserules would be used if an input matches incorrectly any of rules for a script

The order in which domains attempt to determine rules that the input should

be matched is:

1 triggers of the topics

2 the current context

3 the failsafe of the current context

4 the current topic

5 the failsafe of the current topic

6 the failsafe for the domain

When an input is compared to the rules of a script, the input is first compared tothe rules specifically defined by the script If none of these rules match, the input ismatched against the rules of the script’s parents The rules of the scripts are tried

in top to bottom order

2.3 Question answering systems

Kinds of question answering systems range from closed-domain systems (aiming toanswer questions in a specific domain) to open-domain systems (aiming to answer all

of asked questions) In our experiment, the open-domain systems focus on retrievingand ranking related documents corresponding with the input, while the close-domainsystems focus on analysis natural language questions to extract reliable terms.Additionally, natural language question analysis component is the first compo-nent in any question answering systems This component creates an intermediaterepresentation of the input question, which is expressed in natural language, to beutilized in the rest of the system The basis of the question parser is question clas-sification that can be defined as the task of mapping a given question to one of

k classes based on the possible types of the answers (Li and Roth, 2002b) quently, natural language questions analysis techniques are used to identify keywordsand semantic relations in input questions

Subse-Therefore, our related works come from reviewing question answering systemsagainst the question analysis approaches in specific domain driven ones

Pattern-matching based systems

Close-domain question answering systems are usually linked to relational databasesand called natural language interfaces to databases A natural language interface to

a database (NLIDB) is a system that allows the users to access information stored in

a database by typing questions using natural language expressions (Androutsopoulos

et al., 1995)

Early NLIDB systems used pattern-matching technique to process user’s tions and generate corresponding answers (Sneiders,2002) presented a NLIDB sys-tem by using question patterns covering conceptual model of the database Theinput is converted into SQL query by using defined templates that contain entityslots – free space for data instances representing the primary concepts of the ques-tion Some other open-domain systems presented in (Wu et al., 2003;Saxena et al.,

ques-2007) used pattern-matching techniques to respond user’s requests

The main advantage of pattern-matching approach is its simplicity, and the tem can be able to perform well in certain applications However, the one’s shallow-ness would often lead to bad results

sys-2.3 Question answering systems 13

Semantic-based systems

Later NILDBs respond user’s question by using semantic grammar to parse theinput into syntax tree and mapping the tree to a database query In semantic-basedsystems, the grammar’s categories (i.e the non-leaf nodes appearing in the parsetree) have not to correspond to syntactic concepts (Androutsopoulos et al., 1995).Semantic constraints are usually enforced by choosing semantic grammar categories,

in which the grammar’s categories can also be chosen to ease the mapping from thesyntax tree to database objects

Nguyen and Le (Nguyen and Le,2008) introduced a NLIDB question answeringsystem in Vietnamese employing semantic grammars Their system includes twomain modules: QTRAN and TGEN QTRAN (Query Translator) maps a naturallanguage question to an SQL query while TGEN (Text Generator) generates answersbased on the query result tables QTRAN uses limited context-free grammars toanalyze user’s question into syntax tree via CYK algorithm The syntax tree isthen converted into an SQL query by using a mapping dictionary to determinenames of attributes in Vietnamese, names of attributes in the database and names

of individuals stored in these attributes

The PRECISE system (Popescu et al., 2003) maps the natural language tion to a unique semantic interpretation by analyzing some lexicons and semanticconstraints (Stratica et al., 2003) described a template-based system to translateEnglish question into SQL query by matching the syntactic parse of the question to

ques-a set of fixed semques-antic templques-ates Some other systems bques-ased on semques-antic grques-ammques-arrules such as Planes (Waltz, 1978), Eufid (Templeton and Burger, 1983) Semanticgrammar-based approaches were considered as an engineering methodology, whichallows semantic knowledge to be easily included in the system

Annotation-based systems

Recently, some question answering systems that used semantic annotations ated high results in natural language question analysis A well known annotationbased framework is GATE (General Architecture for Text Engineering) (Cunning-ham et al., 2002) which have been used in many question answering systems likeOntology-based AquaLog (Lopez et al., 2007) and QuestIO (Damljanovic et al.,

gener-2008) systems, and Galea’s open-domain system (Galea, 2003), especially for thenatural language question analysis component

Aqualog (Lopez et al., 2007) shown in figure 2.2 is an ontology-based questionanswering system for English and is the basis for the development of our system Anatural language question is mapped to a set of representation based on the inter-mediate triple that is called a Query-Triple through the Linguistic Component byusing Java Annotation Patterns Engine (JAPE) grammars in GATE (Cunningham

et al., 2002) The Relation Similarity Service takes a Query-Triple and processes

it to provide queries with respect to the input ontology called Onto-Triple ThenAqualog uses Onto-Triple to return an answer for users

Figure 2.2: Aqualog’s architecture

In our experiment, we reported an approach to convert Vietnamese natural guage questions into intermediate representation element in query-tuples (Question-structure, Question-class, Term1, Relation, Term2, Term3) based on semantic annota-tions via JAPE grammars (Nguyen et al., 2009) The selected query-tuple type ismore complex aiming to cover a wider variety of question types in different languages

lan-In addition, we proposed a language-independent approach to acquire JAPE rules

in a systematic manner which avoids unintended interaction among rules (Nguyen

et al., 2011) (Phan and Nguyen, 2010) presented an approach to syntactically andsemantically map Vietnamese questions into triple-like of Subject, Verb and Object

in also utilizing JAPE grammars

The START (Katz, 1997; Katz et al., 2006) question answering system alsoused natural language annotations (Katz, 1997) without utilizing GATE A lexicaldatabase WordNet (Fellbaum,1998) is important natural language application Afterthe appearance of WordNet, almost question answering systems used it to provideinformation for analyzing questions

an-a nan-aturan-al lan-anguan-age phran-ase an-and elements in the ontology The communican-ation betweenthe front-end and back-end is an intermediate representation of the question, whichcaptures the semantic structure of the users’ question.

Furthermore, we focus on describing a rule-based approach to directly extract anintermediate representation elements of question via FrameScript scripting language(McGill et al.,2003) (in section 3.2)

The architecture of our question answering system is shown in figure3.1 It includestwo components: the Natural language question analysis and the Answer retrieval.The question analysis component takes the user’s question as an input and re-turns a query-tuple representing the question in a compact form The role of thisintermediate representation is to provide structured information of the input ques-tion for later processing such as retrieving answers

The answer retrieval component includes two main modules: Ontology mapping

15

and Answer extraction It takes an intermediate representation produced by thequestion analysis component and an ontology as its input to generate semanticanswers.

Figure 3.1: Architecture of our question answering system

3.1.1 Natural language question analysis component

3.1.1.1 Intermediate representation of an input question

The intermediate representation used in our approach aims to cover a wider variety

of question types It consists of a question-structure and one or more query-tuple inthe following format:

( question-structure, question-class, T erm1, Relation, T erm2, T erm3 )

where T erm1 represents a concept (object class), T erm2 and T erm3, if exist,represent entities (objects), Relation (property) is a semantic constraint betweenterms in the question This representation is meant to capture the semantics of thequestion

Simple questions corresponding to basic constructions only have one query-tuple

3.1 Vietnamese Question Answering System 17

and its question-structure is the query-tuple’s question-structure More complexquestions such as composite questions are constructed by several sub-questions,each sub-question is described by a separate question-structure, and the question-structure capture this composition attribute This representation is chosen so that

it can represent a richer set of question types Therefore, some terms or relation inthe query-tuple can be missed Composite questions such as:

“list all students in the Faculty of Information Technology whose hometown isHanoi?”

has question structure of type And with two query-tuples where ? represents amissed element: ( UnknRel , List , students , ? , Faculty of Information Technology,

? ) and ( Normal , List , students, hometown, Hanoi, ? )

The definitions of the following question categories of HowWhy, YesNo, What,When, Where, Who, Many, ManyClass, List and Entity, and question-structures ofNormal, UnknTerm, UnknRel, Definition, Compare, ThreeTerm, Clause, Combine,And, Or, Affirm, Affirm_3Term, Affirm_MoreTuples could be found in appendixes

be reused for further processing in subsequent modules New modules are cally designed to handle Vietnamese questions using JAPE grammars over existinglinguistic annotations

specifi-There are three modules that we use to get an intermediate representation ofuser’s question including: preprocessing, syntactic analysis and semantic analysis.The preprocessing module generates TokenVn annotations representing a Viet-namese word with features such as part-of-speech to identify question-words andcomparing-phrases or special-words by using JAPE rules

The syntactic module is responsible for identifying noun phrases, questionphrases and relation phrases between noun phrases or noun phrases and questionphrases The different modules communicate through the annotations, for example,

this module uses the TokenVn annotations, which is the result of the preprocessingmodule.

The semantic analysis module identifies the question structure and producesthe query-tuples as the intermediate representation (question-structure, question-class, Term1, Relation, Term2, Term3) of the input question using the annotationsgenerated by the previous modules Using results from two front modules, we useJAPE grammars to detect the question structure and corresponding terms and re-lations

We create the intermediate representation of input question in hard-wire mannerlinking every detected pattern via JAPE grammars to Java source codes to extractcorresponding elements It takes a lot of time and effort when appearing new pat-terns Therefore, we will present a knowledge acquisition approach to directly get

an intermediate representation of question by using FrameScript scripting language(McGill et al., 2003) in the section 3.2 to resolve above mentioned problems

3.1.2 Answer retrieval component

The answer retrieval component includes two main modules: Ontology Mapping andAnswer Extraction as shown in figure 3.1 It takes an intermediate representationproduced by the question analysis component and an ontology as its input to gen-erate a semantic answer

The task of the Ontology Mapping module is to map terms and relations inthe query-tuple to concepts, instances and relations in the ontology by using stringnames If exact match is not possible, we use string distance algorithm (Cohen

et al.,2003; Vargas-Vera and Motta,2004; Lopez et al.,2007) to find near-matchedelements in the Ontology with the similarity measure above a certain threshold Incase ambiguity is still present, the system interacts with the users by presentingdifferent options to get the correct ontology element

For each query-tuple, the result of the Mapping Ontology module is an tuple where the terms and relations in the query-tuple are now their correspondingelements in the ontology How the Mapping Ontology module finds correspondingelements in the ontology depends on the question structure For example, when thequery-tuple contains Term1, Term2 and Relation with Term3 missing It first tries tomatch Term1and Term2with concepts or instances in the target Ontology After that,the set of potential relations in the Ontology contains only relations between the

ontology-3.2 Using FrameScript for question analysis 19

two mapped conceptsinstances The ontology Relation is then identified in a similarmanner as mapping term to a concept or an instance With the ontology-tuple, theAnswer Extraction module find all individuals of the corresponding ontology concept

of Term1, having the ontology Relation with the individual corresponding to Term2.Depending on the question-structure and question-class, the best semantic answerwill be returned

Unlike existing approaches for English (Lopez et al., 2007) and Vietnamese (insection 3.1.1) where the intermediate representation couldn’t be extracted directly

in rules, we will describe an approach to directly extract the representation of user’squestion using FrameScript scripting language (McGill et al., 2003)

To set the context for the discussion of question analysis, our approach presented

in figure3.2also consists of three sub-modules: preprocessing, syntactic analysis andsemantic analysis

Figure 3.2: Architecture of the natural language question analysis component usingFrameScript

3.2.1 Preprocessing module

The preprocessing module identifies part-of-speech tags in a user’s question Afterthat, we use part-of-speech tags to create basic scripts for detecting words The ba-sic scripts Noun, Verb, Determiner, Adjective, Adverb, Conjunction and Prepositionare used to identify corresponding nouns, verbs, determiners, adjectives, adverbs,

conjunctions and prepositions In fact, these scripts will be used in creating rules inthe syntactic and semantic analysis modules at later stages.

3.2.2 Syntactic analysis module

This module is responsible for determining noun phrases, question phrases and lation phrases between noun phrases or noun phrases and question phrases

re-Concepts and entities are normally expressed in noun phrases Therefore, it isimportant that we can reliably identify noun phrases in order to generate the query-tuple Hence we build the script called NounPhrase (such as a sample script below) tospecify patterns of noun phrases by utilizing scripts generated from the preprocessingmodule

| <Determiner> <Composite> <Noun> } ;;

For example, given the following question: “which projects are about ontologies and