Abstract
Intuitively, a learning algorithm is robust if it can succeed despite adverse conditions. We examine conditions under which learning algorithms for classes of formal languages are able to succeed when the data presentations are systematically incomplete; that is, when certain kinds of examples are systematically absent. One motivation comes from linguistics, where the phonotactic pattern of a language may be understood as the intersection of formal languages, each of which formalizes a distinct linguistic generalization. We examine under what conditions these generalizations can be learned when the only data available to a learner belongs to their intersection. In particular, we provide three formal definitions of robustness in the identification in the limit from positive data paradigm, and several theorems which describe the kinds of classes of formal languages which are, and are not, robustly learnable in the relevant sense. We relate these results to classes relevant to natural language phonology.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
The notion of robustness studied here is different from the one studied by Case et al. [3]. There, a class is “robustly learnable” if and only if its effective transformations are learnable too. As such, their primary interest is classes “outside the world of the recursively enumerable classes.” This paper uses the term “robustly learnable” to mean learnable despite the absence of some positive evidence.
- 2.
Alexander Clark (personal communication) provides a counterexample. Let \(C = \{ L_\infty , L_1, \dots \}\) where \(L_n = \{ a^m : 0< m < n \} \cup \{b^{n+1}\}\) and \(L_\infty = a^+ \cup \{ b\}\). Let \(D = \{ a^*\}\). Both classes are ilpd-learnable but \(\{ L_C \cap L_D : L_C\in C, L_D\in D\}\) is not.
- 3.
Technically, local classes need to be augmented with symbols marking word edges.
- 4.
Note that this is a stronger guarantee than consistency.
References
Angluin, D.: Inductive inference of formal languages from positive data. Inf. Control 45(2), 117–135 (1980)
Blum, L., Blum, M.: Toward a mathematical theory of inductive inference. Inf. Control 28(2), 125–155 (1975)
Case, J., Jain, S., Stephan, F., Wiehagen, R.: Robust learning-rich and poor. J. Comput. Syst. Sci. 69(2), 123–165 (2004)
Clark, A., Lappin, S.: Linguistic Nativism and the Poverty of the Stimulus. Wiley-Blackwell (2011)
Eyraud, R., Heinz, J., Yoshinaka, R.: Efficiency in the identification in the limit learning paradigm. In: Heinz, J., Sempere, J.M. (eds.) Topics in Grammatical Inference, pp. 25–46. Springer, Heidelberg (2016). https://doi.org/10.1007/978-3-662-48395-4_2
Freivalds, R., Kinber, E., Wiehagen, R.: On the power of inductive inference from good examples. Theoret. Comput. Sci. 110(1), 131–144 (1993)
Fulk, M., Jain, S.: Learning in the presence of inaccurate information. Theoret. Comput. Sci. 161, 235–261 (1996)
Gold, E.M.: Language identification in the limit. Inf. Control 10(5), 447–474 (1967)
Haines, L.H.: On free monoids partially ordered by embedding. J. Combinatorial Theory 6(1), 94–98 (1969)
Heinz, J.: String extension learning. In: Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics, pp. 897–906. Association for Computational Linguistics, Uppsala, Sweden (July 2010)
Heinz, J.: The computational nature of phonological generalizations. In: Hyman, L., Plank, F. (eds.) Phonological Typology, Phonetics and Phonology, vol. 23, chap. 5, pp. 126–195. Mouton de Gruyter (2018)
Heinz, J., Kasprzik, A., Kötzing, T.: Learning in the limit with lattice-structured hypothesis spaces. Theoret. Comput. Sci. 457, 111–127 (2012)
Heinz, J., Rawal, C., Tanner, H.G.: Tier-based strictly local constraints for phonology. In: Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Short Papers, vol. 2, pp. 58–64. Association for Computational Linguistics, Portland (2011)
Jain, S.: Program synthesis in the presence of infinite number of inaccuracies. J. Comput. Syst. Sci. 53, 583–591 (1996)
Jain, S., Lange, S., Nessel, J.: On the learnability of recursively enumerable languages from good examples. Theoret. Comput. Sci. 261, 3–29 (2001)
Jain, S., Osherson, D., Royer, J.S., Sharma, A.: Systems That Learn: An Introduction to Learning Theory, 2nd edn. The MIT Press (1999)
Lambert, D.: Grammar interpretations and learning TSL online. In: Proceedings of the Fifteenth International Conference on Grammatical Inference. Proceedings of Machine Learning Research, vol. 153, pp. 81–91, August 2021
Lambert, D.: Relativized adjacency. Journal of Logic, Language and Information, May 2023
Lambert, D., Rawski, J., Heinz, J.: Typology emerges from simplicity in representations and learning. J. Lang. Modelling 9(1), 151–194 (2021)
McNaughton, R., Papert, S.A.: Counter-Free Automata. MIT Press (1971)
Osherson, D.N., Stob, M., Weinstein, S.: Systems That Learn. MIT Press, Cambridge (1986)
Pin, J.E.: Profinite methods in automata theory. In: 26th International Symposium on Theoretical Aspects of Computer Science STACS 2009, February 2009
Rogers, J., et al.: On languages piecewise testable in the strict sense. In: Ebert, C., Jäger, G., Michaelis, J. (eds.) MOL 2007/2009. LNCS (LNAI), vol. 6149, pp. 255–265. Springer, Heidelberg (2010). https://doi.org/10.1007/978-3-642-14322-9_19
Simon, I.: Piecewise testable events. In: Brakhage, H. (ed.) GI-Fachtagung 1975. LNCS, vol. 33, pp. 214–222. Springer, Heidelberg (1975). https://doi.org/10.1007/3-540-07407-4_23
Smetsers, R., Volpato, M., Vaandrager, F., Verwer, S.: Bigger is not always better: on the quality of hypotheses in active automata learning. In: Clark, A., Kanazawa, M., Yoshinaka, R. (eds.) The 12th International Conference on Grammatical Inference. Proceedings of Machine Learning Research, vol. 34, pp. 167–181. PMLR, Kyoto, Japan, 17–19 Sep 2014
Valiant, L.G.: A theory of the learnable. Commun. ACM 27(11), 1134–1142 (1984)
Acknowledgements
We acknowledge support from the Data + Computing = Discovery summer REU program at the Institute for Advanced Computational Science at Stony Brook University, supported by the NSF under award 1950052.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2023 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this paper
Cite this paper
Kaelbling, P., Lambert, D., Heinz, J. (2023). Robust Identification in the Limit from Incomplete Positive Data. In: Fernau, H., Jansen, K. (eds) Fundamentals of Computation Theory. FCT 2023. Lecture Notes in Computer Science, vol 14292. Springer, Cham. https://doi.org/10.1007/978-3-031-43587-4_20
Download citation
DOI: https://doi.org/10.1007/978-3-031-43587-4_20
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-43586-7
Online ISBN: 978-3-031-43587-4
eBook Packages: Computer ScienceComputer Science (R0)