Natural Language Generation Demos Basics of NLG NLG concepts - PowerPoint PPT Presentation

Natural Language Specific Goals Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG Goal: ( compare WA , VA ) NLG concepts Issues in NLG Given: loc ( WA , NORTHWEST ) ∧ loc ( VA , EASTCOAST ) NLG subtasks Architecture of NLG systems Washington State is located in the northwest, while Two-step architecture Three-step architecture Virginia is on the east coast. Hw7 11/50

Natural Language Specific Goals Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG Goal: ( compare WA , VA ) NLG concepts Issues in NLG Given: loc ( WA , NORTHWEST ) ∧ loc ( VA , EASTCOAST ) NLG subtasks Architecture of NLG systems Washington State is located in the northwest, while Two-step architecture Three-step architecture Virginia is on the east coast. Hw7 Washington State is located in the northwest and Virginia on the east coast. 11/50

Natural Language Some generation applications Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG NLG concepts Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 12/50

Natural Language Some generation applications Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos The output component of a machine translation system Basics of NLG NLG concepts Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 12/50

Natural Language Some generation applications Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos The output component of a machine translation system Basics of NLG NLG concepts Interface to a database system Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 12/50

Natural Language Some generation applications Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos The output component of a machine translation system Basics of NLG NLG concepts Interface to a database system Issues in NLG NLG subtasks Interface to an expert system (math tutor) Architecture of NLG systems Two-step architecture Three-step architecture Hw7 12/50

Natural Language Some generation applications Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos The output component of a machine translation system Basics of NLG NLG concepts Interface to a database system Issues in NLG NLG subtasks Interface to an expert system (math tutor) Architecture of NLG systems Autogeneration of help pages for a software system Two-step architecture Three-step architecture Hw7 12/50

Natural Language Some generation applications Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos The output component of a machine translation system Basics of NLG NLG concepts Interface to a database system Issues in NLG NLG subtasks Interface to an expert system (math tutor) Architecture of NLG systems Autogeneration of help pages for a software system Two-step architecture Three-step architecture Robot-human communication Hw7 12/50

Natural Language Some generation applications Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos The output component of a machine translation system Basics of NLG NLG concepts Interface to a database system Issues in NLG NLG subtasks Interface to an expert system (math tutor) Architecture of NLG systems Autogeneration of help pages for a software system Two-step architecture Three-step architecture Robot-human communication Hw7 Data summarization (stock/weather reports) 12/50

Natural Language Some generation applications Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos The output component of a machine translation system Basics of NLG NLG concepts Interface to a database system Issues in NLG NLG subtasks Interface to an expert system (math tutor) Architecture of NLG systems Autogeneration of help pages for a software system Two-step architecture Three-step architecture Robot-human communication Hw7 Data summarization (stock/weather reports) Text summarization 12/50

Natural Language Comparison to NLU Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Similarities Basics of NLG NLG concepts Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 13/50

Natural Language Comparison to NLU Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Similarities Basics of NLG NLG concepts Issues in NLG both utilize a lexicon, grammar, and knowledge NLG subtasks representation Architecture of NLG systems Two-step architecture Three-step architecture Hw7 13/50

Natural Language Comparison to NLU Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Similarities Basics of NLG NLG concepts Issues in NLG both utilize a lexicon, grammar, and knowledge NLG subtasks representation Architecture of NLG systems have same “endpoints” (internal computational Two-step architecture Three-step representation, natural language utterances). architecture Hw7 13/50

Natural Language Comparison to NLU Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Similarities Basics of NLG NLG concepts Issues in NLG both utilize a lexicon, grammar, and knowledge NLG subtasks representation Architecture of NLG systems have same “endpoints” (internal computational Two-step architecture Three-step representation, natural language utterances). architecture Hw7 both symbolic and stochastic (corpus-based) methods apply 13/50

Natural Language Comparison to NLU Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Differences Demos Basics of NLG NLG concepts Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 14/50

Natural Language Comparison to NLU Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Differences Demos NLU is about hypothesis management, where as NLG is Basics of NLG about choice ; When producing NL utterances from NLG concepts Issues in NLG non-linguistic material, everything bit of information to NLG subtasks Architecture of be encoded in the output has to be chosen along the NLG systems Two-step architecture way. Three-step architecture Hw7 14/50

Natural Language Comparison to NLU Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Differences Demos NLU is about hypothesis management, where as NLG is Basics of NLG about choice ; When producing NL utterances from NLG concepts Issues in NLG non-linguistic material, everything bit of information to NLG subtasks Architecture of be encoded in the output has to be chosen along the NLG systems Two-step architecture way. Three-step architecture parsing, or most shallow NLU tasks, ignore certain Hw7 aspects of meaning; NLG utilizes elaborated meaning representations 14/50

Natural Language Comparison to NLU Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Differences Demos NLU is about hypothesis management, where as NLG is Basics of NLG about choice ; When producing NL utterances from NLG concepts Issues in NLG non-linguistic material, everything bit of information to NLG subtasks Architecture of be encoded in the output has to be chosen along the NLG systems Two-step architecture way. Three-step architecture parsing, or most shallow NLU tasks, ignore certain Hw7 aspects of meaning; NLG utilizes elaborated meaning representations research in NLG has focused more on texts than has NLU 14/50

Natural Language Texts Generation Scott Farrar CLMA, University of Washington far- For NLG, we often need to deal with structures larger than rar@u.washington.edu the sentence. Demos Basics of NLG NLG concepts Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 15/50

Natural Language Texts Generation Scott Farrar CLMA, University of Washington far- For NLG, we often need to deal with structures larger than rar@u.washington.edu the sentence. Demos Basics of NLG Definition NLG concepts Issues in NLG NLG subtasks Text : A unit of language larger than the individual clause Architecture of (spoken or written). This is referred to as a discourse in NLG systems Two-step architecture J&M. Three-step architecture Hw7 15/50

Natural Language Texts Generation Scott Farrar CLMA, University of Washington far- For NLG, we often need to deal with structures larger than rar@u.washington.edu the sentence. Demos Basics of NLG Definition NLG concepts Issues in NLG NLG subtasks Text : A unit of language larger than the individual clause Architecture of (spoken or written). This is referred to as a discourse in NLG systems Two-step architecture J&M. Three-step architecture Hw7 a phone conversation 15/50

Natural Language Texts Generation Scott Farrar CLMA, University of Washington far- For NLG, we often need to deal with structures larger than rar@u.washington.edu the sentence. Demos Basics of NLG Definition NLG concepts Issues in NLG NLG subtasks Text : A unit of language larger than the individual clause Architecture of (spoken or written). This is referred to as a discourse in NLG systems Two-step architecture J&M. Three-step architecture Hw7 a phone conversation a recipe 15/50

Natural Language Texts Generation Scott Farrar CLMA, University of Washington far- For NLG, we often need to deal with structures larger than rar@u.washington.edu the sentence. Demos Basics of NLG Definition NLG concepts Issues in NLG NLG subtasks Text : A unit of language larger than the individual clause Architecture of (spoken or written). This is referred to as a discourse in NLG systems Two-step architecture J&M. Three-step architecture Hw7 a phone conversation a recipe a weather report 15/50

Natural Language Texts Generation Scott Farrar CLMA, University of Washington far- For NLG, we often need to deal with structures larger than rar@u.washington.edu the sentence. Demos Basics of NLG Definition NLG concepts Issues in NLG NLG subtasks Text : A unit of language larger than the individual clause Architecture of (spoken or written). This is referred to as a discourse in NLG systems Two-step architecture J&M. Three-step architecture Hw7 a phone conversation a recipe a weather report a paragraph in War and Peace 15/50

Natural Language Texts Generation Scott Farrar CLMA, University of Washington far- For NLG, we often need to deal with structures larger than rar@u.washington.edu the sentence. Demos Basics of NLG Definition NLG concepts Issues in NLG NLG subtasks Text : A unit of language larger than the individual clause Architecture of (spoken or written). This is referred to as a discourse in NLG systems Two-step architecture J&M. Three-step architecture Hw7 a phone conversation a recipe a weather report a paragraph in War and Peace an entire novel 15/50

Natural Language Text coherence Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Such texts exhibit a certain structure and we can speak of Demos their well-formedness, just like any other unit of language. Basics of NLG NLG concepts Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 16/50

Natural Language Text coherence Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Such texts exhibit a certain structure and we can speak of Demos their well-formedness, just like any other unit of language. Basics of NLG NLG concepts Issues in NLG NLG subtasks Definition Architecture of NLG systems A coherent text is one whose parts are interrelated in Two-step architecture Three-step meaningful way. An incoherent text is one whose parts do architecture not bind together in a naturalistic manner. Hw7 16/50

Natural Language Text coherence Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Such texts exhibit a certain structure and we can speak of Demos their well-formedness, just like any other unit of language. Basics of NLG NLG concepts Issues in NLG NLG subtasks Definition Architecture of NLG systems A coherent text is one whose parts are interrelated in Two-step architecture Three-step meaningful way. An incoherent text is one whose parts do architecture not bind together in a naturalistic manner. Hw7 16/50

Natural Language Text coherence Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Such texts exhibit a certain structure and we can speak of Demos their well-formedness, just like any other unit of language. Basics of NLG NLG concepts Issues in NLG NLG subtasks Definition Architecture of NLG systems A coherent text is one whose parts are interrelated in Two-step architecture Three-step meaningful way. An incoherent text is one whose parts do architecture not bind together in a naturalistic manner. Hw7 Pronominalization 16/50

Natural Language Text coherence Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Such texts exhibit a certain structure and we can speak of Demos their well-formedness, just like any other unit of language. Basics of NLG NLG concepts Issues in NLG NLG subtasks Definition Architecture of NLG systems A coherent text is one whose parts are interrelated in Two-step architecture Three-step meaningful way. An incoherent text is one whose parts do architecture not bind together in a naturalistic manner. Hw7 Pronominalization Temporal expressions 16/50

Natural Language Text coherence: pronominal expressions Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos James Riddle “Jimmy” Hoffa (born February 14, 1913, Basics of NLG disappeared July 30, 1975), was an American labor leader. NLG concepts Issues in NLG As the president of the International Brotherhood of NLG subtasks Teamsters from the mid-1950s to the mid-1960s, Hoffa Architecture of NLG systems wielded considerable influence. After he was convicted of Two-step architecture Three-step attempted bribery of a grand juror, he served nearly a architecture decade in prison. He is also well-known in popular culture for Hw7 the mysterious circumstances surrounding his unexplained disappearance and presumed death. His son James P. Hoffa is the current president of the Teamsters. 17/50

Natural Language Text coherence: pronominal expressions Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos James Riddle “Jimmy” Hoffa (born February 14, 1913, Basics of NLG disappeared July 30, 1975), was an American labor leader. NLG concepts Issues in NLG As the president of the International Brotherhood of NLG subtasks Teamsters from the mid-1950s to the mid-1960s, Hoffa Architecture of NLG systems wielded considerable influence. After he was convicted of Two-step architecture Three-step attempted bribery of a grand juror, he served nearly a architecture decade in prison. He is also well-known in popular culture for Hw7 the mysterious circumstances surrounding his unexplained disappearance and presumed death. His son James P. Hoffa is the current president of the Teamsters. 18/50

Natural Language Text coherence: temporal expressions Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu During Buck’s time in jail, Clyde had been the driver in a Demos store robbery. The wife of the murder victim, when shown Basics of NLG NLG concepts photos, picked Clyde as one of the shooters. On August 5, Issues in NLG NLG subtasks 1932, while Bonnie was visiting her mother, Clyde and two Architecture of associates were drinking alcohol at a dance in Stringtown, NLG systems Two-step architecture Oklahoma (illegal under Prohibition). When they were Three-step architecture approached by sheriff C.G. Maxwell and his deputy, Clyde Hw7 opened fire, killing deputy Eugene C. Moore. That was the first killing of a lawman by what was later known as the Barrow Gang, a total which would eventually amount to nine slain officers. 19/50

Natural Language Text coherence: temporal expressions Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu During Buck’s time in jail, Clyde had been the driver in a Demos store robbery. The wife of the murder victim, when shown Basics of NLG NLG concepts photos, picked Clyde as one of the shooters. On August 5, Issues in NLG NLG subtasks 1932, while Bonnie was visiting her mother, Clyde and two Architecture of associates were drinking alcohol at a dance in Stringtown, NLG systems Two-step architecture Oklahoma (illegal under Prohibition). When they were Three-step architecture approached by sheriff C.G. Maxwell and his deputy, Clyde Hw7 opened fire, killing deputy Eugene C. Moore. That was the first killing of a lawman by what was later known as the Barrow Gang, a total which would eventually amount to nine slain officers. 20/50

Natural Language Evaluation Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Problem : evaluation of NLG systems is always a great Basics of NLG NLG concepts challenge (no gold standard) Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 21/50

Natural Language Evaluation Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Problem : evaluation of NLG systems is always a great Basics of NLG NLG concepts challenge (no gold standard) Issues in NLG NLG subtasks Architecture of NLG systems Evaluation techniques: Two-step architecture Three-step architecture Hw7 21/50

Natural Language Evaluation Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Problem : evaluation of NLG systems is always a great Basics of NLG NLG concepts challenge (no gold standard) Issues in NLG NLG subtasks Architecture of NLG systems Evaluation techniques: Two-step architecture Three-step architecture Turing Test (subjective) Hw7 21/50

Natural Language Evaluation Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Problem : evaluation of NLG systems is always a great Basics of NLG NLG concepts challenge (no gold standard) Issues in NLG NLG subtasks Architecture of NLG systems Evaluation techniques: Two-step architecture Three-step architecture Turing Test (subjective) Hw7 task-oriented (expensive) 21/50

Natural Language Evaluation Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Problem : evaluation of NLG systems is always a great Basics of NLG NLG concepts challenge (no gold standard) Issues in NLG NLG subtasks Architecture of NLG systems Evaluation techniques: Two-step architecture Three-step architecture Turing Test (subjective) Hw7 task-oriented (expensive) statistical comparison with real texts (untrustworthy) 21/50

Natural Language Turing Test for evaluation of an NLG system Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG NLG concepts 5 Indistinguishable from human Issues in NLG NLG subtasks 4 Most likely human Architecture of NLG systems 3 Maybe human or machine Two-step architecture Three-step 2 Most likely machine architecture Hw7 1 Definitely machine 22/50

Natural Language Grice’s Conversational Maxims Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG NLG concepts Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 23/50

Natural Language Grice’s Conversational Maxims Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG Quantity : Make your contribution as informative as is NLG concepts required Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 23/50

Natural Language Grice’s Conversational Maxims Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG Quantity : Make your contribution as informative as is NLG concepts required Issues in NLG NLG subtasks Quality : Do not say what you believe is false. Do not Architecture of NLG systems say that for which you lack adequate evidence. Two-step architecture Three-step architecture Hw7 23/50

Natural Language Grice’s Conversational Maxims Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG Quantity : Make your contribution as informative as is NLG concepts required Issues in NLG NLG subtasks Quality : Do not say what you believe is false. Do not Architecture of NLG systems say that for which you lack adequate evidence. Two-step architecture Three-step architecture Relation : Be relevant. Hw7 23/50

Natural Language Grice’s Conversational Maxims Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG Quantity : Make your contribution as informative as is NLG concepts required Issues in NLG NLG subtasks Quality : Do not say what you believe is false. Do not Architecture of NLG systems say that for which you lack adequate evidence. Two-step architecture Three-step architecture Relation : Be relevant. Hw7 Manner : Be perspicuous, avoid obscurity of expression, avoid ambiguity, be brief, be orderly. 23/50

Natural Language Statistical evaluation scenarios Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG Comparison with humanly produced texts: NLG concepts Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 24/50

Natural Language Statistical evaluation scenarios Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG Comparison with humanly produced texts: NLG concepts Issues in NLG NLG subtasks text length, mean length of utterance (MLU) Architecture of NLG systems Two-step architecture Three-step architecture Hw7 24/50

Natural Language Statistical evaluation scenarios Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG Comparison with humanly produced texts: NLG concepts Issues in NLG NLG subtasks text length, mean length of utterance (MLU) Architecture of average number of embedded clauses (and other NLG systems Two-step architecture complex structures) Three-step architecture Hw7 24/50

Natural Language Statistical evaluation scenarios Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG Comparison with humanly produced texts: NLG concepts Issues in NLG NLG subtasks text length, mean length of utterance (MLU) Architecture of average number of embedded clauses (and other NLG systems Two-step architecture complex structures) Three-step architecture diction (number of word types) Hw7 24/50

Natural Language Statistical evaluation scenarios Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG Comparison with humanly produced texts: NLG concepts Issues in NLG NLG subtasks text length, mean length of utterance (MLU) Architecture of average number of embedded clauses (and other NLG systems Two-step architecture complex structures) Three-step architecture diction (number of word types) Hw7 number of long distance dependencies 24/50

Natural Language General evaluation criteria Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Despite the challenges, we can posit some fundamental Basics of NLG NLG concepts criteria for evaluating NLG systems: Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 25/50

Natural Language General evaluation criteria Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Despite the challenges, we can posit some fundamental Basics of NLG NLG concepts criteria for evaluating NLG systems: Issues in NLG NLG subtasks content : does the output contain appropriate/enough Architecture of NLG systems information? Two-step architecture Three-step architecture Hw7 25/50

Natural Language General evaluation criteria Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Despite the challenges, we can posit some fundamental Basics of NLG NLG concepts criteria for evaluating NLG systems: Issues in NLG NLG subtasks content : does the output contain appropriate/enough Architecture of NLG systems information? Two-step architecture Three-step architecture organization : is the discourse structure realistic? Hw7 25/50

Natural Language General evaluation criteria Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Despite the challenges, we can posit some fundamental Basics of NLG NLG concepts criteria for evaluating NLG systems: Issues in NLG NLG subtasks content : does the output contain appropriate/enough Architecture of NLG systems information? Two-step architecture Three-step architecture organization : is the discourse structure realistic? Hw7 correctness : are there grammatical or stylistic errors? 25/50

Natural Language General evaluation criteria Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Despite the challenges, we can posit some fundamental Basics of NLG NLG concepts criteria for evaluating NLG systems: Issues in NLG NLG subtasks content : does the output contain appropriate/enough Architecture of NLG systems information? Two-step architecture Three-step architecture organization : is the discourse structure realistic? Hw7 correctness : are there grammatical or stylistic errors? textual flow : is the language choppy or smooth? 25/50

Natural Language NLG subtask: Non-linguistic Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG NLG concepts There are several tasks for a full generation system: Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 26/50

Natural Language NLG subtask: Non-linguistic Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG NLG concepts There are several tasks for a full generation system: Issues in NLG NLG subtasks content determination : task of deciding what Architecture of NLG systems information is to be communicated Two-step architecture Three-step architecture Hw7 26/50

Natural Language NLG subtask: Non-linguistic Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG NLG concepts There are several tasks for a full generation system: Issues in NLG NLG subtasks content determination : task of deciding what Architecture of NLG systems information is to be communicated Two-step architecture Three-step architecture discourse structuring : deciding how to package the Hw7 ‘chunks’ of content 26/50

Natural Language NLG subtask: Linguistic Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG NLG concepts Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 27/50

Natural Language NLG subtask: Linguistic Generation Scott Farrar lexicalization : determine the particular words and CLMA, University of Washington far- construction types to use rar@u.washington.edu Demos Basics of NLG NLG concepts Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 27/50

Natural Language NLG subtask: Linguistic Generation Scott Farrar lexicalization : determine the particular words and CLMA, University of Washington far- construction types to use rar@u.washington.edu aggregation : decide how much information to include Demos in each sentence/phrase Basics of NLG NLG concepts Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 27/50

Natural Language NLG subtask: Linguistic Generation Scott Farrar lexicalization : determine the particular words and CLMA, University of Washington far- construction types to use rar@u.washington.edu aggregation : decide how much information to include Demos in each sentence/phrase Basics of NLG NLG concepts referring expression generation : determine what Issues in NLG NLG subtasks expressions should be used to refer to the entities in the Architecture of domain NLG systems Two-step architecture Three-step architecture Hw7 27/50

Natural Language NLG subtask: Linguistic Generation Scott Farrar lexicalization : determine the particular words and CLMA, University of Washington far- construction types to use rar@u.washington.edu aggregation : decide how much information to include Demos in each sentence/phrase Basics of NLG NLG concepts referring expression generation : determine what Issues in NLG NLG subtasks expressions should be used to refer to the entities in the Architecture of domain NLG systems Two-step architecture Three-step linguistic realization : task of converting abstract architecture Hw7 representations of sentences to real text; linearization. 27/50

Natural Language NLG subtask: Linguistic Generation Scott Farrar lexicalization : determine the particular words and CLMA, University of Washington far- construction types to use rar@u.washington.edu aggregation : decide how much information to include Demos in each sentence/phrase Basics of NLG NLG concepts referring expression generation : determine what Issues in NLG NLG subtasks expressions should be used to refer to the entities in the Architecture of domain NLG systems Two-step architecture Three-step linguistic realization : task of converting abstract architecture Hw7 representations of sentences to real text; linearization. structure realization : task of converting abstract structures such as paragraphs and sections into markup symbols understood by the document presentation component (e.g., HTML, L A T EX) 27/50

Natural Language NLG subtask: Linguistic Generation Scott Farrar lexicalization : determine the particular words and CLMA, University of Washington far- construction types to use rar@u.washington.edu aggregation : decide how much information to include Demos in each sentence/phrase Basics of NLG NLG concepts referring expression generation : determine what Issues in NLG NLG subtasks expressions should be used to refer to the entities in the Architecture of domain NLG systems Two-step architecture Three-step linguistic realization : task of converting abstract architecture Hw7 representations of sentences to real text; linearization. structure realization : task of converting abstract structures such as paragraphs and sections into markup symbols understood by the document presentation component (e.g., HTML, L A T EX) Need to modularize the above tasks appropriately, given the goal of the NLG system and the available resources. 27/50

Natural Language Lexicalization Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG NLG concepts Issues in NLG (1) a. Mary’s car NLG subtasks b. the car owned by Mary Architecture of NLG systems Two-step architecture Three-step architecture Hw7 28/50

Natural Language Lexicalization Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG NLG concepts Issues in NLG (3) a. Mary’s car NLG subtasks b. the car owned by Mary Architecture of NLG systems Two-step architecture (4) a. the ship’s cargo hold Three-step architecture b. the cargo hold which is part of the ship Hw7 28/50

Natural Language Agrregation Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG (5) a. I am Ron Paul. I am a rogue Republican. NLG concepts Issues in NLG b. My name is Ron Paul and I’m a rogue Republican. NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 29/50

Natural Language Agrregation Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG (7) a. I am Ron Paul. I am a rogue Republican. NLG concepts Issues in NLG b. My name is Ron Paul and I’m a rogue Republican. NLG subtasks Architecture of NLG systems (8) a. The course number is ling571. Two-step architecture Three-step b. The course is difficult. The course is open to architecture Hw7 undergraduates. c. Ling571 is difficult, but open to undergraduates. 29/50

Natural Language Agrregation: other examples Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG NLG concepts Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 30/50

Natural Language Agrregation: other examples Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos John’s bicycle is red Basics of NLG NLG concepts Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 30/50

Natural Language Agrregation: other examples Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos John’s bicycle is red Basics of NLG Mary’s bicycle is yellow NLG concepts Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 30/50

Natural Language Agrregation: other examples Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos John’s bicycle is red Basics of NLG Mary’s bicycle is yellow NLG concepts Issues in NLG NLG subtasks Tom’s bicycle is blue Architecture of NLG systems Two-step architecture Three-step architecture Hw7 30/50

Natural Language Agrregation: other examples Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos John’s bicycle is red Basics of NLG Mary’s bicycle is yellow NLG concepts Issues in NLG NLG subtasks Tom’s bicycle is blue Architecture of NLG systems Lisa’s bicycle is red Two-step architecture Three-step architecture Hw7 30/50

Natural Language Agrregation: other examples Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos John’s bicycle is red Basics of NLG Mary’s bicycle is yellow NLG concepts Issues in NLG NLG subtasks Tom’s bicycle is blue Architecture of NLG systems Lisa’s bicycle is red Two-step architecture Three-step becomes : architecture Hw7 30/50

Natural Language Agrregation: other examples Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos John’s bicycle is red Basics of NLG Mary’s bicycle is yellow NLG concepts Issues in NLG NLG subtasks Tom’s bicycle is blue Architecture of NLG systems Lisa’s bicycle is red Two-step architecture Three-step becomes : architecture Hw7 John and Lisa have red bicycles. 30/50

Natural Language Agrregation: other examples Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos John’s bicycle is red Basics of NLG Mary’s bicycle is yellow NLG concepts Issues in NLG NLG subtasks Tom’s bicycle is blue Architecture of NLG systems Lisa’s bicycle is red Two-step architecture Three-step becomes : architecture Hw7 John and Lisa have red bicycles. Tom’s and Mary’s bicycles are blue and yellow respectively. 30/50

Natural Language Lexical aggregation Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Basics of NLG NLG concepts Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 31/50

Natural Language Lexical aggregation Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Ericsson made profit in 2004 Basics of NLG NLG concepts Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 31/50

Natural Language Lexical aggregation Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Ericsson made profit in 2004 Basics of NLG NLG concepts Nokia made profit in 2004 Issues in NLG NLG subtasks Architecture of NLG systems Two-step architecture Three-step architecture Hw7 31/50

Natural Language Lexical aggregation Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Ericsson made profit in 2004 Basics of NLG NLG concepts Nokia made profit in 2004 Issues in NLG NLG subtasks Siemens made profit in 2004 Architecture of NLG systems Two-step architecture Three-step architecture Hw7 31/50

Natural Language Lexical aggregation Generation Scott Farrar CLMA, University of Washington far- rar@u.washington.edu Demos Ericsson made profit in 2004 Basics of NLG NLG concepts Nokia made profit in 2004 Issues in NLG NLG subtasks Siemens made profit in 2004 Architecture of NLG systems ATT made profit in 2004 Two-step architecture Three-step architecture Hw7 31/50