| { | |
| "paper_id": "1988", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T07:45:28.542449Z" | |
| }, | |
| "title": "Functional Descriptions as a Formalism for Linguistic Knowledge Representation in a Generation Oriented Approach", | |
| "authors": [ | |
| { | |
| "first": "Miyo", | |
| "middle": [], | |
| "last": "Otani", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "[email protected]" | |
| }, | |
| { | |
| "first": "Nathalie", | |
| "middle": [], | |
| "last": "Simonin", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "[email protected]" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "This paper describes how linguistic knowledge has been declaratively formalised using Functional Descriptions (FDs), in the generation module of the SAGE system. SAGE (Sentence Analysis and GEneration) is the Natural Language Frontend of the Dialogue Manager of the Esprit I project Esteam-316. The present implementation has the advantage of being based on two principles, which are the dynamic checking of constraints and the functional unification of knowledge and dynamic objects. In order to provide these functionalities to the former formalism of FD, we have introduce the notions of Syntactic Components and of Current Syntactic Component. The whole sentence is built step by step in a complex tree-like structure. The generation interpreter is able to move upward and downward inside this tree. In addition, our system validates the use of a Lexicon-Grammar (drawn from the LADL studies) for sentence-generation. The target language is English, but all of the knowledge bases have been developed in such a way that the generation process is able to support a change of language.", | |
| "pdf_parse": { | |
| "paper_id": "1988", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "This paper describes how linguistic knowledge has been declaratively formalised using Functional Descriptions (FDs), in the generation module of the SAGE system. SAGE (Sentence Analysis and GEneration) is the Natural Language Frontend of the Dialogue Manager of the Esprit I project Esteam-316. The present implementation has the advantage of being based on two principles, which are the dynamic checking of constraints and the functional unification of knowledge and dynamic objects. In order to provide these functionalities to the former formalism of FD, we have introduce the notions of Syntactic Components and of Current Syntactic Component. The whole sentence is built step by step in a complex tree-like structure. The generation interpreter is able to move upward and downward inside this tree. In addition, our system validates the use of a Lexicon-Grammar (drawn from the LADL studies) for sentence-generation. The target language is English, but all of the knowledge bases have been developed in such a way that the generation process is able to support a change of language.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": "The Generation Module described here belongs to the SAGE system, which parses and generates sentences. SAGE is the Natural Language Frontend of the Dialogue Manager of the Esprit I project ESTEAM-316. ESTEAM-316 is an Advice-Giving system, and its testbed application field is private investment. Several examples given in this paper will be thus related to the financial domain.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Preamble", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "One of the tasks of the generation process is to determine with which syntactic structure an idea will be expressed. Our synthesis module therefore generates a linguistic structure made of nested Syntactic Components (SCs) described by Functional Descriptions (FDs). An SC is characterised by its meaning slot and is composed of Syntactic sub-Components (sub-SCs). The value of the meaning slot is an instance of a semantic concept and is called a token. For instance, a clause SC is composed of the following sub-SCs: a subject, a verb, some complements and adverbials.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Preamble", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "After an introduction to FDs in section 2, section 3 explains how linguistic knowledge has been structured using FDs. Section 4 describes how our generation module handles this knowledge whereas section 5 draws conclusions on the choices made in our implementation.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Preamble", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "The theory of Functional Unification and Functional Descriptions (FDs) has been developed by Martin Kay [Kay 81 ] and is based on the formalism of Functional Grammar developed by Simon Dik [Dik 78 ]. Briefly, a FD is an object described by a set of slots, i.e. an attribute to which a value is associated. The value of the slot is either atomic (integer, real, or string quoted between \" \"), or non-atomic (ordered list of objects quoted between ( ), non-ordered set of objects quoted between { }, nested FD or symbols). Two other kinds of values which are paths and links to other objects have been introduced. The paths are either compiled when the object is loaded in memory, or dynamically computed when the loaded object is handled for instance by the generation process. They are quoted between < > [Fimbel & al 85] . The links are a kind of pointer to another object: they are invisible to the user and may be handled only by the interpreter ( i.e. a program written in C).", | |
| "cite_spans": [ | |
| { | |
| "start": 104, | |
| "end": 111, | |
| "text": "[Kay 81", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 189, | |
| "end": 196, | |
| "text": "[Dik 78", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 805, | |
| "end": 821, | |
| "text": "[Fimbel & al 85]", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Functional Unification and Functional Descriptions", | |
| "sec_num": "2.1" | |
| }, | |
| { | |
| "text": "We wanted our generation system to be declarative and reusable for other applications. For this, we needed to handle dynamic objects in different kinds of rules, especially in the generation grammar. These rules are described by FDs divided in three parts: a condition slot, a body of several slots that may be functionally unified with the current SC, and several slots of conclusion. Conditions and conclusions are lists of FDs containing premise and action slots. We may assess that there is an or logical operator between these FDs within the list. Condition, body and conclusions are not compulsory in a rule FD. However, no more than one condition is allowed, whereas there may be several conclusions. The operands of premises and actions are either constants (FD or any object known by the system) or dynamic objects specified by paths and links.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Constraints: Conditions and Actions", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "One of the most important links used by our system is a link pointing to the SC currently computed, which we shall symbolise by current_SC. Another one is a link from a sub-SC to its parent SC: a path like < current_SC parent meaning > will be interpreted as the meaning of the parent of the current SC. Examples of premises and actions will appear below.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Constraints: Conditions and Actions", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "A rule is relevant if one of the FDs of the condition slot is verified. If so, the body of the rule is unified to the current SC. Then all of the conclusions of the rule are evaluated: this means the actions of the first FD of a conclusion slot whose premises are verified are activated.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Constraints: Conditions and Actions", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "One might wonder about the usefulness of actions in a condition FD: this proved to be helpful in order to modify and prepare the current SC before it is unified with the body of the rule.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Constraints: Conditions and Actions", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "The following section details how the new FD formalism is used by SAGE to develop linguistic knowledge bases.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Constraints: Conditions and Actions", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "The purpose of the lexicon-grammar is to define syntactic properties completely: using it, we are able to take into account a wide range of constructions of a given language. Our lexicon-grammar is based on the theory developed by Maurice Gross [Gross 75 ] and the studies carried out by the LADL on French constructions. To give an idea of this knowledge base, we give the information stored for the direct object of the verb want below.", | |
| "cite_spans": [ | |
| { | |
| "start": 245, | |
| "end": 254, | |
| "text": "[Gross 75", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Lexicon-Grammar: syntactic valencies", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "\u2022 The standard construction is [ Subject + Verb + Direct Object ];", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Lexicon-Grammar: syntactic valencies", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "\u2022 The direct object may be a human being as in \"The mother wants her child\", a nonhuman entity as in \"He wants time\", or a thai-clause as in \"Mary wants that John settle down in Paris\";", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Lexicon-Grammar: syntactic valencies", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "\u2022 The thai-clause is reduced to the following forms:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Lexicon-Grammar: syntactic valencies", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "\u2022 [ Noun group + Adjective ] or NAdj, if the omitted verb is be: e.g \"The teacher wants the exercise ready for tomorrow';", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Lexicon-Grammar: syntactic valencies", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "\u2022 [ Verb in the complete infinitive form + complements ] or ToVinf0, if the concept of the subject of this clause is the same as that of the subject of want: e.g. \"Mary wants to settle down in Paris\";", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Lexicon-Grammar: syntactic valencies", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "\u2022 [ Noun group + Verb in the complete infinitive form + complements ] or NToVinf when the two subjects are different: e.g. \"Mary wants her friends to settle down in Paris\";", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Lexicon-Grammar: syntactic valencies", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "\u2022 The whole clause may be transformed into the passive form.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Lexicon-Grammar: syntactic valencies", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "The semantic distributions are also defined in these FDs: they state to which semantic classes the token of a sub-SC may belong, according to is_a slots. For instance, in our lexicon-grammar, want allows a human being or a non-human item as the direct object.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Lexicon-Grammar: syntactic valencies", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "In our formalism, this information is specified as shown in Figure 2 . The \"+\" symbol states that the corresponding form or semantic distribution is allowed. \"-\" would stand for forbidden constructions and \"?\" would introduce constructions that are acceptable for the parser, but dubious and forbidden in a generation processing.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 60, | |
| "end": 68, | |
| "text": "Figure 2", | |
| "ref_id": "FIGREF1" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Lexicon-Grammar: syntactic valencies", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "The slot interrogative_pronoun defines the interrogative pronouns allowed for referring to the sub-SC in a Wh-questions.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Lexicon-Grammar: syntactic valencies", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "Semantic distributions specified in the lexicon-grammar may be very detailed according to the use of such or such verb or noun. For instance, the subject of to graze may be a sheep, and that of to eat may be a human being. This information is very useful in synthesizing a pronoun: it helps to check whether a pronoun is ambiguous. If the speaker talks about a cow and a sheep, then an \"it\" before the verb to graze probably is the sheep, and an \"it\" before to browse refers to the cow: needless to say these distributions may greatly help the parser too.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Lexicon-Grammar: syntactic valencies", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "Given that identifiers of a token depend only on the concept of which it is an instance, the allocation of the slots in the sub-SCs raises the problem of the link between those specific identifiers and standard sub-SCs. This is solved with a Linguistic Definition (LD), which makes explicit the mapping between slots and Syntactic sub-Components (sub-SCs). A concept is thus associated with several LDs and the generator is able to choose among them according to such or such linguistic constraints.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Linguistic Definitions: a mapping from semantics slots to syntactic components", | |
| "sec_num": "3.2" | |
| }, | |
| { | |
| "text": "For instance, the concept of *transaction may be generated in a clause with the verb to buy, or with the verb to sell. In several cases, when the token of transaction is nested in another one, the noun phrase structure may be best adapted for instance as a subject of a sentence as in \"Her purchase of an expensive cottage remained unknown for a long week\". 1 The LD corresponding to the mapping from an instance of the concept *transaction into the syntactic structure of the verb sell is described in Figure 3 .", | |
| "cite_spans": [ | |
| { | |
| "start": 358, | |
| "end": 359, | |
| "text": "1", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 503, | |
| "end": 511, | |
| "text": "Figure 3", | |
| "ref_id": "FIGREF4" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Linguistic Definitions: a mapping from semantics slots to syntactic components", | |
| "sec_num": "3.2" | |
| }, | |
| { | |
| "text": "[ Assuming that Figure 4 is the initial SC, and that the SC will be unified 2 with the LD of Figure 5 . Concept names are marked with a star * for the sake of readability.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 16, | |
| "end": 24, | |
| "text": "Figure 4", | |
| "ref_id": "FIGREF3" | |
| }, | |
| { | |
| "start": 93, | |
| "end": 101, | |
| "text": "Figure 5", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Linguistic Definitions: a mapping from semantics slots to syntactic components", | |
| "sec_num": "3.2" | |
| }, | |
| { | |
| "text": "The following two paragraphs give examples of how syntactic codes are specified using our \"rule FD formalism\".", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Specification of syntactic coding", | |
| "sec_num": "3.3" | |
| }, | |
| { | |
| "text": "The syntactic forms seen in \u00a73.1 must be specified using a rule format. infinitive form + complements ] is allowed if the subject of the current SC equals the subject of the main clause i.e. of the parent SC. In the body of ToVinf0 (see Figure 6 ), the verb is set to the complete infinitive form, and the subject is erased because not expressed in an infinitive clause. ", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 237, | |
| "end": 245, | |
| "text": "Figure 6", | |
| "ref_id": "FIGREF6" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Standard syntactic forms", | |
| "sec_num": "3.3.1" | |
| }, | |
| { | |
| "text": "A slot interrogative_pronoun has been added in each sub-SC of the lexicon-grammar items, in order to generate the appropriate pronoun in a Wh-question. This is spectacular when the pronoun is not directly predictable from the syntactic function the sub-SC (subject, object or adverbials). For instance, starting from \"The fund will be available in 2 years.\", we may generate \" Under what delay will the fund be available?\".", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Interrogative pronouns", | |
| "sec_num": "3.3.2" | |
| }, | |
| { | |
| "text": "In the case given in Figure 2 , the objectl SC may be transformed using the code wh_what described by Figure 7 . ", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 21, | |
| "end": 29, | |
| "text": "Figure 2", | |
| "ref_id": "FIGREF1" | |
| }, | |
| { | |
| "start": 102, | |
| "end": 110, | |
| "text": "Figure 7", | |
| "ref_id": "FIGREF8" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Interrogative pronouns", | |
| "sec_num": "3.3.2" | |
| }, | |
| { | |
| "text": "Besides the KBs mentioned above, which are the Linguistic Definitions and the Lexicon-Grammar, there are mainly the generation grammar, the semantic dictionary and the lexicon of words. The last two will not be described here:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Other Knowledge Bases", | |
| "sec_num": "3.4" | |
| }, | |
| { | |
| "text": "\u2022 The semantic dictionary is a semantic net of FDs linked with is_a slots. Briefly, the concepts are defined with FDs divided in three parts: 1) the semantic net link descriptions; 2) a schemata which lists the slots that characterize a token, i.e. an instance of the concept; and 3) the list of LDs that the generation module has at its disposal for the given token. \u2022 The lexicon is a standard dictionary of English words with the indication of syntactic categories (noun, preposition, auxiliary, verb, etc) and of conjugation models (the verb to eat obeys the same conjugation rules as to sing).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Other Knowledge Bases", | |
| "sec_num": "3.4" | |
| }, | |
| { | |
| "text": "The following section details the grammar structure and the generation process itself.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Other Knowledge Bases", | |
| "sec_num": "3.4" | |
| }, | |
| { | |
| "text": "For a given generation rule, the grammar specifies under what conditions it may be applied using the slot condition, what rules are to be chosen for the synthesis of each sub-SC in the FD of the sub-SC of the body of the rule, and what actions are to be carried out on the current SC (such as choosing the number and person of a verb according to the subject within a clause). In our implementation, neither condition nor conclusion slots are needed in the clause rule. When provided, they indicate in which cases a grammar rule is relevant to the current SC, and if so, what actions are to be undertaken on the SC.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Generation strategies 4.1 Generation grammar rules", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "The names of the sub-SCs are made explicit using the slot list_of_subSCs. Sub-SCs are not all compulsory and may be deleted in the SC if no meaning is given to them after the unification of the LD (see \u00a73.2). The parent of the sub-SCs is the current SC: it is specified so by the path < current_SC > in the slot parent. Figure 8 shows also how the slot grammar_rules specifies the grammar rules allowed for each sub-SC. One sub-SC may be synthesized simply using another grammar rule. For instance, objects of a clause may be synthesized as personal pronoun using gram_pers_pron rule or as reflexive pronoun using gram_reflexive rule. Most of the time, the sub-SCs are generated using LD and lexicon-grammar as stated by the symbol linguistic_synthesis. The generation rules are tried in the order specified by grammar_rules.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 320, | |
| "end": 328, | |
| "text": "Figure 8", | |
| "ref_id": "FIGREF9" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Generation strategies 4.1 Generation grammar rules", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "The order slots specify the order of the sub-SCs in the generated sentence, and may be modified by actions. For instance, if the object2 sub-SC is synthesized as a personal pronoun, then it is put before objectl by decreasing its order value form 210 to 110. The order values are interpreted by the morphological generator i.e. all sub-SCs of an SC are sorted according to their order, before the corresponding words are generated. ", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Generation strategies 4.1 Generation grammar rules", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "The generation process is top-down, with backtracking. It recursively builds a complex object of several nested SCs. At the beginning of the generation process, the first SC is an object similar to:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Interpretation in the generation process", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "[ meaning = token_to_be_generated ]", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Interpretation in the generation process", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "The current SC is supplemented with sub-SCs in a loop: according to the concept of the token in the meaning slot, the generation interpreter chooses a LD, then a syntactic structure in the lexicon-grammar. These two FDs are functionally unified with the current SC.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Interpretation in the generation process", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "Then, one syntactic form like noun_phrase or ToVinf0 (see \u00a73.3.1) is chosen for the current SC according to the grammar rule given in the LD and the validity condition of the syntactic form. These forma are compulsory only if the type of the current SC is of clause or noun phrase. In the former case, the generator needs either a complete clause form or a reduced one and chooses among the lists of reduction_form or clause_form; in the second case, it requires a noun_phrase_form. If found, the form is functionally unified with the current SC. Figure 9 shows the SC corresponding to You want time after functional unifications. The concept *user is generated in a dialog pronoun into the second person (you) because the Natural Language Front-End is integrated with a Person-Machine Dialogue application.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 547, | |
| "end": 555, | |
| "text": "Figure 9", | |
| "ref_id": "FIGREF11" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Interpretation in the generation process", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "At this stage of the process, the generator may add several modifiers to the current level, adverbials in clauses, or adjectives in noun groups: these adjuncts are also carried through functional unifications since the modifiers are also described in a FD just like any LD. Henceforth, the modifiers are handled in the same way as the sub-SCs defined by the LD.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Interpretation in the generation process", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "Lastly, the grammar rule specified by the LD is applied. First, the body of the grammar rule is functionally unified to the current SC. Secondly, the sub- This is where our declarative KBs based on Functional Descriptions prove to be efficient.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Interpretation in the generation process", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "The same heuristic based on series of functional unifications is used for totally different structures such as noun phrase or clause. Therefore, this loop is allowed to be totally recursive.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Interpretation in the generation process", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "Then the conclusion slots of the grammar rule are evaluated, if there are any.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Interpretation in the generation process", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "Transformations are processed whenever they are needed, as for questions (which puts the verb in the interrogative form and inserts an auxiliary verb before the subject), or negations or passive transformations.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Interpretation in the generation process", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "If a failure occurs during this loop, for instance if one of the sub-SC has not been generated completely, then backtracking is carried out by choosing another LD and/or another grammar rule.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Interpretation in the generation process", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "The main feature of our Natural Language Front-End is that the Parsing and Generation processes are carried out using the same linguistic knowledge bases. On the other hand, parsing and generation grammar formalism differ because of their dedicated heuristics. Unlike parsing, our generation process is not a sequence of \"left-to-right\" procedures [Danlos 87a ] [Danlos 87b ]. Moreover, a given heuristic of clause transformation is strongly dedicated to a parsing or to a generation process: during generation, the order of the objects in a clause may be modified after they have been synthesized into personal pronouns (see \u00a74.2); during parsing, it is not possible to recognize an objectl and an object2 unless the aphorisms are solved.", | |
| "cite_spans": [ | |
| { | |
| "start": 348, | |
| "end": 359, | |
| "text": "[Danlos 87a", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 362, | |
| "end": 373, | |
| "text": "[Danlos 87b", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Conclusion on parsing and generation grammars", | |
| "sec_num": "4.3" | |
| }, | |
| { | |
| "text": "Rules may be declaratively defined using condition and conclusion slots. This has been made possible with the link current_SC and with the slot parent referring to the embedding SC, allowing dynamic paths like < current_SC parent meaning >.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Declaration and interpretation", | |
| "sec_num": "5.1" | |
| }, | |
| { | |
| "text": "Yet some of the generation inferences are invisible to the user for being integrated in the C programs. For instance, the choice of a syntactic form (ToVinf0, NAdj, etc) is not declared in a grammar rule (clause or noun phrase rule) but is processed by a specific C function called before the grammar rule is activated. This implicit inference might become visible and might be declared in the generation grammar if an action choose_a_form_among were introduced, with the following format: choose_a_form_among = < current_SC noun_phrase_form >. Other premises or actions may be added to the inference engine in the same way.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Declaration and interpretation", | |
| "sec_num": "5.1" | |
| }, | |
| { | |
| "text": "The concept of link in FD is useful for handling objects defined dynamically with paths, and also for backtracking.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Backtracking", | |
| "sec_num": "5.2" | |
| }, | |
| { | |
| "text": "If a grammar rule fails for instance, then we must be able to recover a former state of the current SC i.e. to destroy slots that have been added by functional unification. So each SC is actually handled as a link to a FD. Before any inference is carried out, the FD of each SC is copied and stored in a backup FD. In case of failure of the generation process, the backtracking process makes the SC links point to the latter stored FDs: the same links (i.e. SC pointers) are thus still relevant.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Backtracking", | |
| "sec_num": "5.2" | |
| }, | |
| { | |
| "text": "Presently, we are able to generate sentences with infinitive clauses, imperative clauses, Yes-No questions, several Wh-questions (where the interrogative pronoun is bound either to the main clause or to a nested clause). Here is a table of several generated sentences together with an average response time they require on a SUN 3/50. We believe that we brought the theory of Functional Description and Functional Unification some promising features, especially concerning link handling and rule description. The constitution of declarative knowledge using dynamic operands and declarative rules is a principle that may be applied in other fields of Natural Language Processing. This allow the sharing of linguistic knowledge bases by parser and generation module, even though there are distinct parsing and generation grammars, and dedicated interpreters in C.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Results and performance", | |
| "sec_num": "5.3" | |
| }, | |
| { | |
| "text": "Moreover, we believe that FDs as described in our present paper may be useful for realms of Knowledge based systems other than Natural Language Processing.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Results and performance", | |
| "sec_num": "5.3" | |
| }, | |
| { | |
| "text": "We do not take into account pragmatic rules concerning focus, intention of the speaker, etc for choosing among several LDs. 2 in the meaning of functional unification. See[Kay 81].", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "Syntactic Component for Generation", | |
| "authors": [ | |
| { | |
| "first": "Laurence", | |
| "middle": [], | |
| "last": "Danlos", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "French", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "English", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1987, | |
| "venue": "Natural Language Generation: New Results in Artificial Intelligence, Psychology and Linguistics", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Laurence Danlos, A French and English Syntactic Component for Genera- tion, Natural Language Generation: New Results in Artificial Intelligence, Psychology and Linguistics, Kempen G. ed, Dordrecht/Boston, Martinus Nijhoff Publishers, 1987.", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "The linguistic basis of text generation", | |
| "authors": [ | |
| { | |
| "first": "Laurence", | |
| "middle": [], | |
| "last": "Danlos", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": null, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Laurence Danlos, The linguistic basis of text generation, Cambridge Uni- versity Press.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Functional Grammar, Publications in Language Science", | |
| "authors": [ | |
| { | |
| "first": "Simon", | |
| "middle": [], | |
| "last": "Dik", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1978, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Simon Dik, Functional Grammar, Publications in Language Science, Foris Publications, Dordrecht, Holland, 1978.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "Using a Text Model for Analysis and Generation", | |
| "authors": [ | |
| { | |
| "first": "Eric", | |
| "middle": [], | |
| "last": "Fimbel", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Herbert", | |
| "middle": [], | |
| "last": "Grosco", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jean-Marie", | |
| "middle": [], | |
| "last": "Lancel", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Nathalie", | |
| "middle": [], | |
| "last": "Simonin", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": null, | |
| "venue": "Proceedings of ACL'85", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "226--231", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Fimbel & al 85] Eric Fimbel, Herbert Grosco, Jean-Marie Lancel, Nathalie Simonin, Us- ing a Text Model for Analysis and Generation, Proceedings of ACL'85, Geneva, pp 226-231.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "M\u00e9thodes en syntaxe", | |
| "authors": [ | |
| { | |
| "first": "Maurice", | |
| "middle": [], | |
| "last": "Gross", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1975, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Maurice Gross, M\u00e9thodes en syntaxe, R\u00e9gime des constructions compl\u00e9tives, Hermann, 1975.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Unification Grammars", | |
| "authors": [ | |
| { | |
| "first": "Martin", | |
| "middle": [], | |
| "last": "Kay", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1981, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Martin Kay, Unification Grammars, Xerox Publication, 1981.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "Nathalie Simonin, A Grammar used for Parsing and Generation", | |
| "authors": [], | |
| "year": 1986, | |
| "venue": "11th International Conference on Computational Linguistics, Proceedings, Coling'86", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Lancel & al 86] Jean-Marie Lancel, Francois Rousselot, Nathalie Simonin, A Grammar used for Parsing and Generation, 11th International Conference on Com- putational Linguistics, Proceedings, Coling'86, August 1986.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "Sentence Parsing and Generation with a Semantic Dictionary and a Lexicon-Grammar", | |
| "authors": [ | |
| { | |
| "first": "Jean-Marie", | |
| "middle": [], | |
| "last": "Lancel", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Miyo", | |
| "middle": [], | |
| "last": "Otani", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Nathalie", | |
| "middle": [], | |
| "last": "Simonin", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1988, | |
| "venue": "13th International Conference on Computational Linguistics, Proceedings, Coling'88", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "& al 88] Jean-Marie Lancel, Miyo Otani, Nathalie Simonin, Laurence Danlos SAGE: a Sentence Parsing and Generation System, \"Sentence Parsing and Gener- ation with a Semantic Dictionary and a Lexicon-Grammar\", 13th Interna- tional Conference on Computational Linguistics, Proceedings, Coling'88, August 1988.", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Sentence Generation: from semantic representations to sentences throughout Linguistic Definitions and Lexicon-Grammar", | |
| "authors": [ | |
| { | |
| "first": "Miyo", | |
| "middle": [], | |
| "last": "Otani", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jean-Marie", | |
| "middle": [], | |
| "last": "Lancel", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1988, | |
| "venue": "Proceedings of Ecai '88", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Miyo Otani, Jean-Marie Lancel, Sentence Generation: from semantic rep- resentations to sentences throughout Linguistic Definitions and Lexicon- Grammar, Proceedings of Ecai '88, August 1988.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "FIGREF0": { | |
| "num": null, | |
| "text": "Figure 1is an example of a FD.[ age = 26 height_in_meter = 1.83 name = \"Jimmy\" chronology_of_professional_positions = ( software_engineer project_manager ) hobbies = { skiing rock_climbing movies } present_position = [ function = project_manager affiliation = Cap_Sogeti_Innovation ] ] An example of FD.", | |
| "uris": null, | |
| "type_str": "figure" | |
| }, | |
| "FIGREF1": { | |
| "num": null, | |
| "text": "Syntactic_want \u2194 [ verb = [ word = want ] subject = [ noun_phrase_form = { ( noun_phrase + ) } distribution = { ( *human + ) } interrogative-pronoun = ( wh_who )] object1 = [ noun_phrase_form = { ( noun_phrase + )} reduced_clause_form = { ( ToVinf0 + ) ( NAdj + )} clause-form = { ( NToVinf + )} distribution = { ( *human + ) ( *non_human + ) } interrogative-pronoun = ( wh_what wh_who ) ] transformation = ( passive_transformation ) ... ] Characterization of want in the Lexicon-grammar.", | |
| "uris": null, | |
| "type_str": "figure" | |
| }, | |
| "FIGREF2": { | |
| "num": null, | |
| "text": "subject = [ meaning = < current_SC meaning seller > ] objectl = [ meaning = < current_SC meaning object > ] object2 = [ meaning = < current_SC meaning buyer > ] gramma_-rule = gram_clause } Figure 3: A Linguistic Definition.", | |
| "uris": null, | |
| "type_str": "figure" | |
| }, | |
| "FIGREF3": { | |
| "num": null, | |
| "text": "meaning = [ instance_of = *transaction buyer = *user object = [ instance_of = *house type = *cottage ] ] An initial SC.", | |
| "uris": null, | |
| "type_str": "figure" | |
| }, | |
| "FIGREF4": { | |
| "num": null, | |
| "text": "the path evaluation will result in the SC ofFigure 5. After path evaluation, the sub-SCs with no meaning slot are deleted as in the case of the object2 sub-SC in", | |
| "uris": null, | |
| "type_str": "figure" | |
| }, | |
| "FIGREF5": { | |
| "num": null, | |
| "text": "In our system, the condition slot of the symbol ToVinf0 states that the construction [ Verb in the complete [ meaning = [ instance_of = *transaction seller = *user object = [ instance_of = *house type = *cottage ] ] subject = [ meaning = *user ] objectl = [ meaning = [ instance_of = *house type = *cottage ] ] grammar_rule = gram_clause ] Figure 5: An SC after an LD has been unified.", | |
| "uris": null, | |
| "type_str": "figure" | |
| }, | |
| "FIGREF6": { | |
| "num": null, | |
| "text": "ToVinf0 \u2192 [ condition = ( [ equal = ( < current_SC subject meaning > < current_SC parent subject meaning > ) ] ) subject = [ structure = erased ] verb = [ form = complete_infinitive ] ] Description of the Syntactic form ToVinf0.", | |
| "uris": null, | |
| "type_str": "figure" | |
| }, | |
| "FIGREF7": { | |
| "num": null, | |
| "text": "wh_what \u2192 [ condition = ( [ not_a = ( < current_SC meaning > *human ) ] ) pronoun = [ word = what ] ]", | |
| "uris": null, | |
| "type_str": "figure" | |
| }, | |
| "FIGREF8": { | |
| "num": null, | |
| "text": "Description of the Syntactic code wh_what.", | |
| "uris": null, | |
| "type_str": "figure" | |
| }, | |
| "FIGREF9": { | |
| "num": null, | |
| "text": "is an example of what may be specified for the clause grammar rule.", | |
| "uris": null, | |
| "type_str": "figure" | |
| }, | |
| "FIGREF10": { | |
| "num": null, | |
| "text": "gram_clause \u2194 [ list_of_subSCs = ( subject verb objectl object2 adverbial_date ) rule_type = clause_type subject = [ parent = < current_SC > order = 50 grammar_rules = ( gram_pers_pron linguistic_synthesis ) ] verb = [ parent = < current_SC > order = 60 grammar_rules = ( gram_verbe ) ] objectl = [ order = 200 parent = < current_SC > grammar_rules = ( gram_reflexive gram_pers_pron linguistic_synthesis ) } object2 = [ order = 210 parent -< current_SC > grammar_rules = ( gram_reflexive gram_pers_pron linguistic_synthesis ) ] adverbial_date = [ parent = < current_SC > order =400 grammar_rules = ( gram_pers_pron linguistic_synthesis ) ] ] An example of a clause grammar rule.", | |
| "uris": null, | |
| "type_str": "figure" | |
| }, | |
| "FIGREF11": { | |
| "num": null, | |
| "text": "SCs are synthesized either by another grammar rule such as the one allowing the pronominalization of sub-SCs, or by the [ meaning = [ instance_of = *want actor = *user object = [ instance_of = *time ] ] subject = [ meaning = *user distribution = {(*human +)} noun_phrase = {(noun_phrase +)} ] verb = [ word = want ] objectl = [ meaning = [ instance_of = *time ] distribution = {(*non_human +) (*human +)} reduction-form = {(ToVinf0 +) (NAdj +)} clause_form = {(NToVinf +)} noun_phrase_form = {(noun_phrase +)} ] transformation = {(passivel 100)} ] Syntactic Component of You want time.generation process described in the present paragraph \u00a74.2, according to the grammar_rules list. It is during that phase that the sub-SCs become the current SC in turn.", | |
| "uris": null, | |
| "type_str": "figure" | |
| } | |
| } | |
| } | |
| } |