Niyogi, P., & Berwick, R. C. (1996). A Language Learning Model for Finite Parameter Spaces. In M. R. Brent (Ed.), Computational Approaches to Language Acquisition (pp. 161-193).  Home/Search   Document Not in Database   Summary   Related Articles   Check

....of the search and the constraining topology of the parametric theory do little to prevent learning times from rivaling those of exhaustive search. In fact, the use of parametric theory to constrain the direction of search can be a significant hindrance, making learning impossible in some cases (Niyogi Berwick, 1996). For further discussion of the limitations of the TLA, see (Fodor, 1998) A more complex form of random search is genetic algorithms. Such algorithms have been applied to many types of problems, including grammar learning (Clark, 1992; Clark Roberts, 1993; Pulleyblank Turkel, 1998) This type ....

Niyogi, P., & Berwick, R. C. (1996). A Language Learning Model for Finite Parameter Spaces. In M. R. Brent (Ed.), Computational Approaches to Language Acquisition (pp. 161-193).

....of input sentences consumed by the STL before convergence on the target grammar can be derived from a relatively straightforward Markov analysis. Importantly, the formulation most useful to analyze performance does not require states which represent the grammars of the parameter space (contra Niyogi and Berwick (1996)) Instead, each state of the system depicts the number of parameters that have been set, t, and the state transitions represent the probability that the STL will adopt some number of new parameter values, w, on the basis of the current state and whatever usable parametric information is revealed ....

P. Niyogi and R.C. Berwick. 1996. A language learning model for finite parameter spaces. Cognition, 61:161--193.

.... model of language learning because it predicts that the child will take a memoryless walk through grammar space starting from a random complete grammar and may not converge to g t 2 UG even when exposed to a fair sample of triggering data from g t (e.g. Brent, 1996; Briscoe, 1999, 2000a; Niyogi and Berwick, 1996). Here I outline a framework for thinking about grammatical learning in the context of language change and some desiderata for the LA drawing on and extending Briscoe (2000b) Then I briefly present one model of the LA which instantiates the framework and satisfies these desiderata, simplifying ....

Niyogi, P. and Berwick, R.C. (1996) `A language learning model for finite parameter spaces', Cognition, vol.61, 161--193.

....maxima and subset superset relations which may cause a learner to converge to 2 See, e.g. Pinker (1994) or Aitchison (1996) for recent positive summaries and discussion of this evidence. See Sampson (1989) for a dissenting view. 2 an incorrect grammar (Clark, 1992; Gibson and Wexler, 1994; Niyogi and Berwick, 1996; Wexler and Manzini, 1987) Gibson and Wexler (1994) formalize the concept of a trigger (e.g. Lightfoot, 1992:13f) as a simple (unembedded or degree 0) sentence of primary linguistic data which signals the value of some parameter and can serve to guide the learner to the target grammar. The ....

.... grammars will require time proportional to their number (e.g. Clark, 1992) while the number of positive samples of the language, and hence amount of time required for convergence to a target grammar, can be arbitrarily long depending on the distribution of trigger types in the language (e.g. Niyogi and Berwick, 1996). Brent (1996) argues that Markovian memoryless procedures of the type introduced by Gibson and Wexler (1994) and further investigated by Niyogi and Berwick (1996) are psychologically implausible because they predict that a child may repeatedly revisit the same hypothesis and or jump randomly ....

[Article contains additional citation context not shown here]

Niyogi, Partha and Robert Berwick (1996) `A language learning model for finite parameter spaces', Cognition, vol.61, 161--193.

.... timescale (e.g. Hurford, 1987; Kirby, 1998; Steels, 1998) and continued to coevolve with the LAD (e.g. Briscoe, 1997, 1998, 2000a) The model of the LAD presented here builds on and extends previous work in the parameter setting framework (e.g. Chomsky, 1981; Clark, 1992; Gibson and Wexler, 1994; Niyogi and Berwick, 1996; Briscoe, 1997, 1998, 2000a) by developing a Bayesian account of parameter setting, and embedding this within a more general theory of language acquisition in which it is not essential that the hypothesis space of grammars is finite. The Bayesian account of parameter setting can explain the ....

....if necessary, updated parameters from a randomly drawm sequence of unembedded or singly embedded (potential) triggers, t from L(g t ) with V CA(t) preassigned. The predefined proper subset of triggers used constituted a uniformly distributed fair sample capable of distinguishing each g 2 G (e.g. Niyogi and Berwick, 1996). The first figure in Table 2 shows the mean number of potential triggers required by the learners to converge on each of the eight languages. These figures are each calculated from 1000 trials and rounded to the nearest integer. Presentation of 150 sentence types for each trial ensured ....

[Article contains additional citation context not shown here]

Niyogi, P. and Berwick, R.C. (1996) `A language learning model for finite parameter spaces', Cognition, vol.61, 161--193.

....yields the left branching derivation and interpretation where the second adverb has wide scope. Such complications are not relevant here. 5 This formal setting and associated assumptions concerning models of language acquisition is based on recent work on formal learnability see, for example, Niyogi (1996) for a particular thorough treatment though its antecedents go back at least to Wexler and Culicover (1980) The assumption that the learner has access to a fair and effective sample circumvents most of the substantive issues of language acquisition. However, our focus here is not on these ....

....acquisition. However, our focus here is not on these issues but rather on how the process of acquisition influences language change. 6 Gibson and Wexler (1994) following earlier work on formal learnability, incorporate both these requirements into the Trigger Learning Algorithm. However, Niyogi and Berwick (1996), Frank and Kapur (1996) Dresher (1999) and Fodor (1998) have all questioned one or both and made alternative proposals. 7 Following Gibson and Wexler (1994) I will write SVO, using S(ubject) O(bject) etc to informally indicate relevant aspects of both SF and LF. I also use the same ....

[Article contains additional citation context not shown here]

Niyogi, P. and Berwick, R.C. (1996) `A language learning model for finite parameter spaces', Cognition, vol.61, 161--193.

.... to address the in principle viability of connectionist networks in handling recursion, in much the same way as simple artificial languages have been used, for example, to assess the feasibility of symbolic parameter setting approaches to the learning of linguistic structure (Gibson Wexler, 1994; Niyogi Berwick, 1996). The structure of this paper is as follows. We begin by distinguishing varieties of recursion in natural language, considering the three kinds of recursion discussed in Chomsky (1957) We then summarize past connectionist research dealing with natural language recursion. Next, we introduce three ....

....this approach could be generalized to model the acquisition of the full complexity of natural language. Note, however, that this limitation applies equally well to symbolic approaches to language acquisition (e.g. Anderson, 1983) including parameter setting models (e.g. Gibson Wexler, 1994; Niyogi Berwick, 1996), and other models which assume an innate universal grammar (e.g. Berwick Weinberg, 1984) Turning to linguistic issues, the better performance of the SRN on cross dependency recursion compared with center embedding recursion may reflect the fact that the difference between learning limited ....

Niyogi, P. & Berwick, R.C. (1996). A language learning model for finite parameter spaces.

.... timescale (e.g. Hurford, 1987; Kirby, 1998; Steels, 1998) and continued to coevolve with the LAD (e.g. Briscoe, 1997, 1998a,b) The model of the LAD presented here builds on and extends previous work in the parameter setting framework (e.g. Chomsky, 1981; Clark, 1992; Gibson and Wexler, 1994; Niyogi and Berwick, 1996; Briscoe, 1997, 1998a,b) by developing a Bayesian account of parameter setting, and embedding this within a more general theory of language acquisition in which it is not essential that the hypothesis space of grammars is finite. The Bayesian account of parameter setting can explain the ....

....on each of the eight languages. These figures are each calculated from 1000 trials and rounded to the nearest integer. 120 sentence types were required to guarantee convergence with p 0:99 on all languages tested for the default learner. This figure rose to 150 for the unset learner. See Niyogi and Berwick, 1996 for detailed discussion of high probability convergence from finite data. As can be seen, the unset learner is equally effective on all eight languages, however, the preferences incorporated into the default learner s initial p setting make languages compatible (e.g. SVO) or partially compatible ....

[Article contains additional citation context not shown here]

Niyogi, P. and Berwick, R.C. (1996) `A language learning model for finite parameter spaces', Cognition, vol.61, 161--193.

.... variety of explanations have been offered for the emergence of an innate LAD with such properties based on saltation (Berwick, 1998; Bickerton, 1990, 1998) or genetic assimilation (Pinker and Bloom, 1990; Kirby, 1998) Formal models of parameter setting (e.g. Clark, 1992; Gibson and Wexler, 1994; Niyogi and Berwick, 1996; Brent, 1996) have demonstrated that development of a psychologically plausible and effective parameter setting algorithm, even for minimal fragments of UG, is not trivial. The account developed in Briscoe (1997, 1998, 1999a,b,c,d) and outlined here improves the account of parameter setting, and ....

....if necessary, updated parameters from a randomly drawm sequence of unembedded or singly embedded (potential) triggers, t from L(g t ) with V CA(t) preassigned. The predefined proper subset of triggers used constituted a uniformly distributed fair sample capable of distinguishing each g 2 G (e.g. Niyogi and Berwick, 1996). The first figure in Table 2 shows the mean number of potential triggers required by the learners to converge on each of the eight languages. These figures are each calculated from 1000 trials and rounded to the nearest integer. Presentation of 150 sentence types for each trial ensured ....

[Article contains additional citation context not shown here]

Niyogi, P. and Berwick, R.C. (1996) `A language learning model for finite parameter spaces', Cognition, vol.61, 161--193.

....of UG. Formal models of parameter setting have been developed for small fragments, but even the search spaces defined by these models contain local maxima and subset superset relations which may cause a learner to converge to an incorrect grammar (Clark, 1992; Gibson and Wexler, 1994; Niyogi and Berwick, 1996; Wexler and Manzini, 1987) Gibson and Wexler (1994) formalize the concept of a trigger (e.g. Lightfoot, 1992:13f) as a simple (unembedded or degree 0) sentence of primary linguistic data which signals the value of some parameter and can serve to guide the learner to the target grammar. The ....

.... grammars will require time proportional to their number (e.g. Clark, 1992) whilst the number of positive samples of the language and hence amount of time required for convergence to a target grammar can be arbitrarily long depending on the distribution of trigger types in the language (e.g. Niyogi and Berwick, 1996). The model presented in section 2 addresses these issues via a modified parameter setting procedure, which can learn more complex grammars than those investigated by Gibson and Wexler, and which is more directed and therefore less psychologically implausible than the Markovian memoryless ....

[Article contains additional citation context not shown here]

Niyogi, P. and Berwick, R.C. (1996) `A language learning model for finite parameter spaces', Cognition, vol.61, 161--193.

.... timescale (e.g. Hurford, 1987; Kirby, 1998; Steels, 1998) and continued to coevolve with the LAD (e.g. Briscoe, 1997, 1998a,b) The model of the LAD presented here builds on and extends previous work in the parameter setting framework (e.g. Chomsky, 1981; Clark, 1992; Gibson and Wexler, 1994; Niyogi and Berwick, 1996; Briscoe, 1997, 1998a,b) by developing a Bayesian account of parameter setting, and embedding this within a more general theory of language acquisition in which it is not essential that the hypothesis space of grammars is finite. The Bayesian account of parameter setting can explain the ....

....the (formal) framework countenances such apparently sub optimal learners, but whether they would be preferentially selected during evolution. The pidgin creole transformation (e.g. Bickerton, 1984) suggests that in unusual circumstances, children are superset language learners (see below) See Niyogi and Berwick (1996) and Muggleton (1996) for further discussion of high probability convergence from finite data. i.e. that VCA(t) is always given) However, this is an unrealistic assumption. Even if we allow that a learner will only alter parameter settings given a trigger, that is, a determinate SF:LF pairing, ....

[Article contains additional citation context not shown here]

Niyogi, P. and Berwick, R.C. (1996) `A language learning model for finite parameter spaces', Cognition, vol.61, 161--193.

....exponential behaviour is more often the outcome observed as shown below. Figures 3.5 3.7. show how maturation time has an effect from various initial conditions. Figure 3.5. shows the situation where initially the whole population speaks VOS V2. The result is that eventually VOS V2 takes over (Niyogi Berwick 1995) Transition from Transition to 0 B B B B B B B B B B B B L 0 L 1 L 2 L 3 L 4 L 5 L 6 L 7 L 0 1 3 2 3 L 1 1 L 2 3 5 2 5 L 3 1 L 4 1 L 5 1 L 6 1 L 7 1 1 C C C C C C C C C C C C A Table 3.3. Transition matrix for a learner, assuming a population of speakers of SOV V2. This matrix shows the probabilities relating to ....

Niyogi, P. & Berwick, R. C. (1996b), `A language learning model for finite parameter spaces', Cognition nn(n), yyy--zzz.

No context found.

Niyogi, P. and R. Berwick (1996a). A language learning model for finite parameter space. Cognition 61: 162-193.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC