Toward a Functional Model of Human Language Processing

Toward a Functional Model of Human Language Processing Jerry Ball 1 , Mary Freiman 2 , Stuart Rodgers 3 & Christopher Myers 1 Air Force Research Laboratory 1 , L3 Communications 2 , AGS TechNet 3 Jerry.Ball@mesa.afmc.af.mil, Mary.Freiman@mesa.afmc.af.mil, Stu@agstechnet.com, Christopher.Myers@mesa.afmc.af.mil Abstract This paper describes a computational cognitive model of human language processing under development in the ACT- R cognitive architecture. The paper begins with the context for the research, followed by a discussion of the primary theoretical and modeling commitments. The main theoretical commitment is to develop a language model which is at once functional and cognitively plausible. The paper continues with a description of the word recognition subcomponent of the language model which uses a perceptual span and ACT- R‟s spreading activation mechanism to activate and select the lexical unit that most closely matches the perceptual input. Next we present a description of the linguistic structure building component of the model which combines parallel, probabilistic processing with serial, pseudo- deterministic processing, including a non-monotonic context accommodation mechanism. A description of the mapping of linguistic representations into a situation model, follows. The paper concludes with a summary and conclusions. Keywords: human language processing (HLP); functional; cognitively plausible; pseudo-deterministic. Introduction The capability to model the cognitive processes associated with language is a long sought-after goal of cognitive science. Computational cognitive process models help researchers to not only understand language processes in their own right, but to determine how they affect and interact with other cognitive processes (e.g., reasoning, decision-making, situation assessment, etc.). Scaled-up versions of these models also support the development of cognitive agents with communicative capabilities based on human linguistic processes (Ball et al., 2009; Douglass, Ball & Rodgers, 2009). In this paper we present a “snapshot” of a functional language comprehension model under development within the ACT-R architecture (Anderson, 2007). The model implements a referential and relational theory of human language processing (Ball, 2007; Ball, Heiberg & Silber, 2007) within ACT-R 1 . A key commitment of the language comprehension research is development of a model which is at once cognitively plausible and functional. We believe that adherence to well-established cognitive constraints will At the time of publication the model contained 6,395 declarative memory elements and 548 production rules which cover a broad range of grammatical constructions. facilitate the development of functional models by pushing development in directions that are more likely to be successful. There are short-term costs associated with adherence to cognitive constraints; however, we have already realized longer-term benefits. For example, the integration of a word recognition capability with ACT-R‟s perceptual system and higher-level linguistic processing has facilitated the recognition and processing of multi-word expressions and multi-unit words in ways that are not available to systems with separate word tokenizing and part of speech tagging processes. Using an available tokenizer and part of speech tagger would have initially facilitated development, but the cognitive implausibility of using staged tokenizing and part of speech tagging led us to reject this approach. The benefits that we have realized as a result of this decision are described below. Theoretical & Modeling Commitments There is extensive psycholinguistic evidence that human language processing is incremental and interactive (Gibson & Pearlmutter, 1998; Altmann, 1998; Tanenhaus et al., 1995; Altmann & Steedman, 1988). Garden-path effects, although infrequent, strongly suggest that processing is essentially serial at the level of phrasal and clausal analysis (Bever, 1970). Lower level processes of word recognition suggest parallel, activation-based processing mechanisms (McClelland & Rumelhart, 1981; Paap et al., 1982). Summarizing the psycholinguistic evidence, Altmann & Mirkovic (2009, p. 605) claim “The view we are left with is a comprehension system that is „maximally incremental‟; it develops the fullest interpretation of a sentence fragment at each moment of the fragment‟s unfolding”. These cognitive constraints legislate against staged analysis models. All levels of analysis must at least be highly pipelined together, if not, in addition, allowing feedback from higher to lower levels. They also suggest the need for hybrid systems which incorporate a mixture of parallel and serial mechanisms, with lower levels of processing being primarily parallel, probabilistic and interactive, while higher levels of analysis are primarily serial, deterministic and incremental. To adhere to and take advantage of these cognitive constraints, we have developed a pseudo-deterministic human language processing model—i.e. a model that presents the appearance and efficiency of serial, deterministic processing, but uses a non-monotonic context