## Non-Parametric Bayesian Areal Linguistics

### Cached

### Download Links

### BibTeX

@MISC{Iii_non-parametricbayesian,

author = {Hal Daumé Iii},

title = {Non-Parametric Bayesian Areal Linguistics},

year = {}

}

### OpenURL

### Abstract

We describe a statistical model over linguistic areas and phylogeny. Our model recovers known areas and identifies a plausible hierarchy of areal features. The use of areas improves genetic reconstruction of languages both qualitatively and quantitatively according to a variety of metrics. We model linguistic areas by a Pitman-Yor process and linguistic phylogeny by Kingman’s coalescent. 1

### Citations

4178 | Pattern Classification and Scene Analysis - Duda, Hart - 1973 |

234 | The two-parameter Poisson-Dirichlet distribution derived from a stable subordinator
- Pitman, Yor
- 1997
(Show Context)
Citation Context ...tic areas) and Kingman’s coalescent (for modeling linguistic phylogeny), both described below. 2.2.1 The Pitman-Yor Process One particular example of a non-parametric prior is the Pitman-Yor process (=-=Pitman and Yor, 1997-=-), which can be seen as an extension to the betterknown Dirichlet process (Ferguson, 1974). The Pitman-Yor process can be understood as a particular example of a Chinese Restaurant process (CRP) (Pitm... |

231 |
The coalescent
- Kingman
- 1982
(Show Context)
Citation Context ...eton tables growing as O(α log N). 2.2.2 Kingman’s Coalescent Kingman’s coalescent is a standard model in population genetics describing the common genealogy (ancestral tree) of a set of individuals (=-=Kingman, 1982-=-b; Kingman, 1982a). In its full form it is a distribution over the genealogy of a countable set. Consider the genealogy of n individuals alive at the present time t = 0. We can trace their ancestry ba... |

174 |
Prior distributions on spaces of probability measures
- Ferguson
- 1974
(Show Context)
Citation Context ...2.2.1 The Pitman-Yor Process One particular example of a non-parametric prior is the Pitman-Yor process (Pitman and Yor, 1997), which can be seen as an extension to the betterknown Dirichlet process (=-=Ferguson, 1974-=-). The Pitman-Yor process can be understood as a particular example of a Chinese Restaurant process (CRP) (Pitman, 2002). The idea in all CRPs is that there exists a restaurant with an infinite number... |

171 | On the genealogy of large populations
- Kingman
- 1982
(Show Context)
Citation Context ...eton tables growing as O(α log N). 2.2.2 Kingman’s Coalescent Kingman’s coalescent is a standard model in population genetics describing the common genealogy (ancestral tree) of a set of individuals (=-=Kingman, 1982-=-b; Kingman, 1982a). In its full form it is a distribution over the genealogy of a countable set. Consider the genealogy of n individuals alive at the present time t = 0. We can trace their ancestry ba... |

52 |
An Indoeuropean classification: a lexicostatistical experiment. Transactions of the American Philosophical Society 82 (1-132). Ancestral trees for binary trait data 23
- Dyen, Kruskal, et al.
- 1992
(Show Context)
Citation Context ...trees is significantly aided by knowledge of areal features. We note that Warnow et al. (2005) have independently proposed a model for phonological change in Indo-European (based on the Dyen dataset (=-=Dyen et al., 1992-=-)) that includes notions of borrowing. Our model is different in that we (a) base our model on typological features rather than just lexical patterns and (b) we explicitly represent language areas, no... |

50 | Bayesian Hierarchical Clustering
- Heller, Ghahramani
- 2005
(Show Context)
Citation Context ...ady shown to outperform competing hierarchical clustering algorithms such as average-link agglomerative clustering (see, eg., Duda and Hart (1973)) and the Bayesian Hierarchical Clustering algorithm (=-=Heller and Ghahramani, 2005-=-). We run the same experiment both on the IE subset of data and on the whole-world subset. We evaluate the results qualitatively, by observing the trees found (on the IE subset) and quantitatively (be... |

45 | 2008): “Kernel stick-breaking processes
- Dunson, Park
(Show Context)
Citation Context ... language centers and locations part of the Pitman-Yor distribution ensures this is not an issue. An alternative would be to use a location-sensitive process such as the kernel stickbreaking process (=-=Dunson and Park, 2007-=-), though we do not explore that here. corresponding to the value preferences in the langauge area to which language n belongs. If it is derived genetically, then Xn,f is drawn from a variable corresp... |

44 |
Grey 2001. Language Contact: an Introduction
- Thomason
(Show Context)
Citation Context ...upon in the field of areal linguistics. Different linguists favor different defintions of what it means to be a linguistic area (are two languages sufficient to describe an area or do you need three (=-=Thomason, 2001-=-; Katz, 1975)?), what areal features are (is there a linear ordering of “borrowability” (Katz, 1975; Curnow, 2001) or is that too prescriptive?), and what causes sharing to take place (does social sta... |

27 | Clustering by Committee
- Pantel
- 2003
(Show Context)
Citation Context ...se the value of K for each metric that did best (giving an unfair advantage). Clustering performance is measured on the Indo-European task according to the Rand Index, F-score, Normalized Edit Score (=-=Pantel, 2003-=-) and Normalized Variation of Information (Meila, 2003). In these results, we see that the Pitman-Yor process model dominates the K-means model and the Areal model dominates the Pitman-Yor model. 5.2 ... |

24 | Bayesian agglomerative clustering with coalescents - Teh, Daumé, et al. - 2008 |

15 | A Bayesian model for discovering typological implications - Hal, Campbell - 2007 |

14 | Proto Oceanic and the Austronesian Languages of Western Melanesia - Ross - 1988 |

12 | Dirichlet processes, chinese restaurant processes and all that. Tutorial presentation at the NIPS Conference - Jordan - 2005 |

10 |
Historical Linguistics
- Campbell
- 1998
(Show Context)
Citation Context ...mbrenis in 1194 (Campbell, In press). A recent article provides a short introduction to both the issues that surround areal linguistics, as well as an enumeration of many of the known language areas (=-=Campbell, 2005-=-). A fairly wide, modern treatment of the issues surrounding areal diffusion is also given by essays in a recent book edited by Aikhenvald and Dixon (2001). The essays in this book provide a good intr... |

7 | Language contact - Moravcsik - 1978 |

4 |
Jowan 2001. What language features can be 'borrowed
- Curnow
(Show Context)
Citation Context ...guistic area (are two languages sufficient to describe an area or do you need three (Thomason, 2001; Katz, 1975)?), what areal features are (is there a linear ordering of “borrowability” (Katz, 1975; =-=Curnow, 2001-=-) or is that too prescriptive?), and what causes sharing to take place (does social status or number of speakers play a role (Thomason, 2001)?). In this paper, we attempt to provide a statistical answ... |

4 |
Non-parametric Bayesian Methods’, Tutorial presentation at
- Ghahramani
- 2005
(Show Context)
Citation Context ...o outperform competing hierarchical clustering algorithms such as average-link agglomerative clustering (see, eg., Duda and Hart (1973)) and the Bayesian Hierarchical Clustering algorithm (Heller and =-=Ghahramani, 2005-=-). We run the same experiment both on the IE subset of data and on the whole-world subset. We evaluate the results qualitatively, by observing the trees found (on the IE subset) and quantitatively (be... |

4 | The Analysis of Linguistic Borrowing. Language - Haugen - 1964 |

4 |
External versus internal factors in the development of language' (Norsk tidsskrift for sprogvidenskap 19
- Sommerfelt
- 1960
(Show Context)
Citation Context ...s well as the somewhat surprising Irish. However, in addition to being intuitively plausible, it is not hard to find evidence in the literature for the contact relationship between English and Irish (=-=Sommerfelt, 1960-=-). In the whole-world experiment, the model identified too many linguistic areas to fit (39 in total that contained at least two languages, and contained at least two language families). In Figure 5.1... |

3 | diffusion and genetic inheritance: Problems in comparative linguistics - Aikhenvald, M |

1 |
Areal linguistics: the problem to the answer
- Campbell
- 2006
(Show Context)
Citation Context ...duction Why are some languages more alike than others? This question is one of the most central issues in historical linguistics. Typically, one of three answers is given (Aikhenvald and Dixon, 2001; =-=Campbell, 2006-=-). First, the languages may be related “genetically.” That is, they may have all derived from a common ancestor language. Second, the similarities may be due to chance. Some language properties are si... |

1 |
Generative Phonologie und phonologische Sprachbünde des Ostjakischen un Samojedischen. Wilhelm Fink
- Katz
- 1975
(Show Context)
Citation Context ...d of areal linguistics. Different linguists favor different defintions of what it means to be a linguistic area (are two languages sufficient to describe an area or do you need three (Thomason, 2001; =-=Katz, 1975-=-)?), what areal features are (is there a linear ordering of “borrowability” (Katz, 1975; Curnow, 2001) or is that too prescriptive?), and what causes sharing to take place (does social status or numbe... |

1 | Genetic versus contact relationship: prosodic diffusibility in South-East Asian languages - Matisoff - 2001 |