A parallel approach to syntax for generation

doi:10.1016/0004-3702(92)90017-R

Artificial Intelligence

Volume 57, Issues 2–3, October 1992, Pages 183-225

https://doi.org/10.1016/0004-3702(92)90017-R Get rights and content

Abstract

To produce good utterances from nontrivial inputs a natural language generator should consider many words in parallel, which raises the question of how to handle syntax in a parallel generator. If a generator is incremental and centered on the task of word choice, then the role of syntax is merely to help evaluate the appropriateness of words. One way to do this is to represent syntactic knowledge as an inventory of “syntactic constructions” and to have many constructions active in parallel at run-time. If this is done then the syntactic form of utterances can be emergent, resulting from synergy among constructions, and there is no need to build up or manipulate representations of syntactic structure. This approach is implemented in FIG, an incremental generator based on spreading activation, in which syntactic knowledge is represented in the same network as world knowledge and lexical knowledge.

References (38)

R.A. Brooks
Intelligence without representation
Artif. Intell.
(1991)
G.E. Hinton
Preface to the Special Issue on Connectionist Symbol Processing
Artif. Intell.
(1990)
G. Kempen et al.
An incremental procedural grammar for sentence formulation
Cogn. Sci.
(1987)
G. van Noord
An overview of head-driven bottom-up generation
P.E. Agre et al.
Pengi: an implementation of a theory of activity
B.K. Baars
The competing plans hypothesis: an heuristic viewpoint on the causes of errors in speech
W.L. Chafe
The deployment of consciousness in the production of a narrative
L. Danlos
Conceptual and linguistic decisions in generation
K.J. De Smedt
Incremental sentence generation: a computer model of grammatical encoding
J.A. Feldman et al.
Computing with structured connectionist networks
Commun. ACM
(1988)

C.J. Fillmore

The mechanisms of “Construction Grammar”

C.J. Fillmore

On grammatical constructions, course notes

(1989)

W. Finkler et al.

POPEL-HOW: A distributed parallel model for incremental natural language production with feedback

M. Gasser

A connectionist model of sentence generation in a first and second language

E.H. Hovy

P.S. Jacobs

A knowledge-based approach to language generation

D. Jurafsky

Representing and integrating linguistic knowledge

J. Kalita et al.

Generation of simple sentences in English using the connectionist model of computation

H. Karlgren

Why trees in syntax?

Stud. Linguist.

(1976)

Cited by (7)

Gaze durations during speech reflect word selection and phonological encoding
2001, Cognition
Speakers produced the sentence frame The A and the B are above the C to describe three pictured objects while their eye movements were monitored. Object B or C varied in codability (the number of alternative names for it) and in the frequency of its dominant name. Codability is known to affect speed of word selection, and word frequency, speed to retrieve a word's pronunciation (phonological encoding). Speakers gazed longer at lower codability and lower frequency objects before naming them. However, the codability and frequency of B and C did not affect when speakers began naming A, even when utterances were perfectly fluent. The results indicate that speakers began “The A…” once they had a name prepared for A, before selecting names for B and C. Similar gaze patterns during less constrained scene description tasks in other studies suggest that speakers often incrementally select and phonologically encode nouns in fluent utterances.
Syntactic priming in spoken sentence production - An online study
2001, Cognition
Citation Excerpt :
Such data are in line, moreover, with corpus linguistic observations of naturally occurring speech which provide evidence of the use of a phrasal grammatical encoding scope during speech production (Clark & Wasow, 1998; Garrett, 1975, 1976, 1980). Importantly, such a view of the scope of syntactic planning accords well with incremental models of speech production which have emphasized that speech production, rather than being holistic so that the grammatical structure of an entire sentence is generated simultaneously, is piecemeal so that the grammatical structure of a sentence is broken down into discrete phrasal chunks and formulated at successive points in time (De Smedt, 1994, 1996; Ferreira, 1996; Kempen, 1987; Kempen & Hoenkamp, 1987; Levelt, 1989; Ward, 1992; Wheeldon, Meyer, & Smith, in press; Wundt, 1900). In short, the present study offers a first look at the phenomenon of syntactic persistence in the context of an online experimental design.
Six experiments investigate syntactic priming online via a picture description task in which participants produce target sentences whose initial phrase is syntactically similar or dissimilar to that of the prime sentence produced on the previous trial. In the first experiment it is shown that a syntactically related prime sentence speeds onset latencies to a subsequent target sentence by approximately 50 ms relative to a syntactically unrelated prime sentence. In the second experiment, the cost of the process of lemma access is factored out via a picture previewing technique but a priming effect is still obtained demonstrating that the effect is not a product of the priming of lemma access processes. In Experiment 3, the related and unrelated prime trials feature the same picture display but the 50 ms facilitation effect is still observed indicating that the effect does not result from the priming of visual perception of the picture movements. This is further strengthened in Experiment 4 which uses written prime sentences rather than a picture description task on the prime trial and still obtains a facilitation effect. In Experiment 5, the effect disappears when the participants are instructed to name the movements but not the objects depicted in the array and this is interpreted as evidence against the view that the effect results from the conceptualization of the events depicted by the array. In the final experiment, the scope of the syntactic persistence effect is investigated by priming sentences with initial phrases of varying syntactic complexity. Significant priming is only observed for an initial phrase featuring two nouns – a finding consistent with the view that the syntactic persistence effect applies only to the generation of the first phrase of an utterance prior to speech onset. The implications of these results are analyzed in the final discussion section.
High level processing scope in spoken sentence production
1999, Cognition
Five experiments investigate the scope of conceptual and grammatical encoding during spoken sentence production. An online picture description task is employed in which participants generate a variety of sentences in response to an array of moving pictured objects. Experiment 1, demonstrates longer onset latencies for single clause sentences beginning with a complex phrase (e.g. The dog and the kite move above the house) than for matched single clause sentences beginning with a simple phrase (e.g. The dog moves above the kite and the house). This finding suggests that more time is dedicated to the processing of the first phrase of an utterance than the remainder prior to speech onset. Experiments 2 and 3, compare the production of single and double clause sentences. The main effect of Experiment 1 is replicated. However, the data also suggest that some time is dedicated to the processing of elements within the second clause prior to speech onset. In Experiment 4, when participants are allowed to preview pictures prior to movement and timer onset the effect of initial phrase complexity is significantly reduced indicating that the latency effects observed previously primarily reflect lemma access. Finally, Experiment 5 demonstrates that this reduction is greater for nouns within the first phrase than for nouns beyond it. We conclude from these experiments that, prior to speech onset, lemma access is completed for the first phrase of an utterance and that high level processing is initiated but not completed for the remainder of a sentence beyond the first phrase.
Language control and code-switching
2018, Languages
A control process model of code-switching
2014, Language, Cognition and Neuroscience
Architectures for natural language generation: Problems and perspectives
1996, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

View all citing articles on Scopus

View full text

A parallel approach to syntax for generation

Abstract

Artif. Intell.

Artif. Intell.

Cogn. Sci.

Pengi: an implementation of a theory of activity

The competing plans hypothesis: an heuristic viewpoint on the causes of errors in speech

The deployment of consciousness in the production of a narrative

Conceptual and linguistic decisions in generation

Incremental sentence generation: a computer model of grammatical encoding

Computing with structured connectionist networks

Commun. ACM