Phon

 

For people interested in child phonology, there is a separate directory within the CHILDES database for corpora that have been full transcribed phonetically and phonologically. A group of child phonologists are working to construct PhonBank, and our.  PhonBank is supported by grant RO1-HD051698 from NIH-NICHHD to Brian MacWhinney and Yvan Rose. The title is "A Shared Database for the Study of Phonological Development." This databases  rely on the PHON program. We are working with a group of child phonologists to construct PhonBank.

Phon is a software program that greatly facilitates a number of tasks related to the analysis of phonological data. Built to support research in phonological development (including babbling), second language acquisition, and phonological disorders, Phon can also be used to investigate virtually all types of phonological investigations (e.g. loanword phonology, fieldwork in phonology). Phon supports multimedia data linkage, unit segmentation (e.g. utterance, word), multiple-blind transcription, automatic labeling of data (features, syllabification), and systematic comparisons between target (model) and actual (produced) phonological forms. All of these functions are accessible through a user-friendly graphical interface. Databases managed within Phon can also be queried using a powerful search system adapted for the needs of the phonologist.

This software program works on both Mac OS X and Windows platforms and is compliant with the CHILDES (TalkBank) XML data format. Phon is being made freely available to the community as open-source software. Phon facilitates data exchange among researchers and is currently used for the elaboration of the shared PhonBank database, a new initiative within CHILDES to support empirical needs of research in all areas of phonological development.

 

1. Introduction

Phon, one of the latest additions to the CHILDES software toolkit, is designed to facilitate a number of tasks related to the transcription and analysis of phonological data. Built to support research in early phonological development (including babbling), second language acquisition, and phonological disorders, Phon can also be used in virtually all types of investigations based on phonological data (e.g. loanword phonology, fieldwork). Phon is available to the community as free, open-source software. It facilitates data exchange among researchers and is currently used for the elaboration of the PhonBank database, which already provides a large corpus of babbling and phonological development data. We will describe these data in greater detail in the next issue of the Bulletin. The development of Phon is currently supported through a grant from the National Institute of Health to Brian MacWhinney and Yvan Rose, in which about 40 phonologists have joined together to both provide their input for the development of Phon and commit their data to PhonBank. Through our consortium, which has been growing at a steady pace, a large number of people have already been involved in the development and testing of Phon and the development of PhonBank. If you have interests in either of these projects, you are more than welcome to join by contacting Brian MacWhinney or Yvan Rose. We also encourage you to subscribe to the discussion group (no Gmail account is required to subscribe):

  • To join the group, click on the “Apply for group membership” and follow the instructions.
  • Group’s email address (for message posting): phon@googlegroups.com

Programmed in Java, Phon works on both Mac OS X and Windows platforms. Because it is Unicode and XML compliant, files can be transferred across computers without changing the fonts or file format. The design and programming work has been achieved primarily at Memorial University of Newfoundland by Greg Hedlund (chief programmer), in close collaboration with Franklin Chen, Leonid Spektor, and Brian MacWhinney from the CHILDES team at Carnegie Mellon University. I am mostly involved with defining the requirements of Phon’s main functions, in close liaison with members of the consortium.

The 1.3 version of Phon was released on time for the IASCL conference in Edinburgh last summer, during which Brian MacWhinney and Yvan Rose provided a demonstration of the software and offered an outlook on ongoing and future projects. Since then, we have continued to develop Phon’s functionality. In the paragraphs below, I provide more detail on Phon’s key features and current work on their refinement. This discussion is followed by a quick outlook on more substantial, future additions to the project.

 

2. Phon’s key features

In its current version, Phon offers full support for the development of corpora of phonological data. The functions are organized around a logical workflow which can be tailored for the special needs of the researcher. The next subsections are organized according to this general workflow.

2.1 Media linkage and segmentation

The first two functions, through which the user can identify portions of the recorded media that are relevant for research (typically, words or utterances) and directly access them from the associated transcript, are similar to the ones found in CLAN, with the exception of cosmetic differences. For example, the Phon user has access to a graphical user interface (GUI) to playback the segmented utterances and modify their start or end times. In addition, given a recording with multiple speakers, the user can now perform full segmentation of a single speaker and later perform segmentation of the other speaker(s). Phon has functions to resume segmentation for each speaker as well as to reorder the segmented records so that they appear in chronological order in the transcript.

2.2 Phonetic transcription and validation

To our knowledge, Phon offers the first fully-integrated system for multiple-blind IPA transcriptions. Using this system, an unlimited number of transcribers can perform their transcriptions without access to other transcribers’ work. This system also supports password protection for blind transcriptions. However, the decision to use password protection, which may be overkill in many situations, is left to the user.

The use of multi-blind transcriptions implies the need for validation. The user (or, ideally, team of users) responsible for transcript validation can visualize and compare all blind transcriptions for each utterance. The transcription deemed the most accurate is selected with a simple mouse click and can be further modified as needed.

The use of multi-blind transcription and associated validation systems is optional. Depending on research needs, the user can decide whether to use these functions. Only validated transcriptions are included in the transcript for research; other, non-validated transcriptions are saved as part of the project file but cannot be used for research. If the user decides not to use multi-blind transcriptions, the transcriptions are entered directly into the transcript. Except from the mode of entry into the session editor (multi-blind or direct), the interface for multi-blind transcriptions or direct data entry is identical.

Whichever mode of transcription the user prefers, Phon provides useful functions to streamline the inherently time-consuming process of transcription. For example, a built-in IPA map is provided, which interacts with the session editor in a way that reduces the amount of mouse clicks to a minimum. In order to further streamline the work involved with transcription, we recommend that users incorporate an IPA keyboard layout such as IPAKeys (Mac OS X) or Keyman (Windows), available from the Unicode sections listed on the CHILDES main portal.

Phon also supports built-in dictionaries of pronounced forms, which provide quick access to generic target (adult) forms. Already present in the 1.3 version are dictionaries for Catalan, English, French and Spanish. We are now in the process of incorporating the Fonilex dictionary of Dutch (Flemish) pronounced forms, thanks to an initial contact by Steven Gillis and a very positive response from Piets Mertens. Also, in collaboration with Maarten Janssen from the Instituto de Linguística Teórica e Computacional and the Portuguese team led by Maria João Freitas at Universidade de Lisboa, we are also working on adding a European Portuguese dictionary.

2.3 Division of the transcribed utterances into Word Groups

Research in phonology often requires a segmentation of transcribed utterances into smaller domains such as phrases or clitic groups, all of which are potential domains for phonological processes. The Phon user can break the transcribed utterance into domains which we call ‘word groups’. This further division of the utterance, controlled at the level of the orthographic transcription, is automatically reflected in the IPA Target and Actual tiers, or any group-aligned user-defined tier such as the “Morphology” tier in the example below (note also the absence of bracketing in the notes tier, which is not group-aligned).

Example Record:
Orthography: [I love] [to study] [child language]
IPA Target: [ˈaɪ ˈlʌv] [tə ˈstʌdiː] [ˈʧaɪld ˈlæŋɡwəʤ]
IPA Actual: [a ˈlaf] [ə ˈtadi] [ˈʧaɪ ˈjæwəʒ]
Morphology: [pro V] [P V] [N N]
Notes: This is just an example

Word Group alignment effectively provides a system for ‘vertical’ analysis across target and produced forms within each utterance, which can also be used to narrow down searches to particular domains. Because of non-trivial constraints, however, nested or overlapping word groups are currently unsupported. We have no plans to tackle this issue at this stage.

2.4 Automatic segmental feature labelling

Each symbol of the IPA, including diacritics, entered in IPA Target or Actual transcriptions, is associated with a set of descriptive features. This association is effective in the background (invisible to the user). Features and feature combinations can be used in many types of queries. Phon’s feature set is as general and redundant as possible, in order not to impose any bias on the analysis.

We are planning to incorporate in a future version a feature set editor, in order to give the user the ability to add or remove features for a given symbol, and also add additional symbols to the set. This latter function will be useful for the transcription of sounds which have no clear correspondence with the IPA. For example, once the editor is in place, the user will be able to add a cover symbol such as “F” in the transcript, and associate the feature set {Labial, Continuant} to it, irrespective of other potential specifications such as {Voiced} or {Labiodental}.

2.5 Automatic syllabification

Within the IPA Target and Actual tiers, Phon automatically breaks down each transcribed word into syllables, and provides a syllable constituent label (e.g. onset, nucleus, coda) to each segment. The syllabification is derived through an algorithm using a deterministic cascade of rules, in a way similar to traditional theories of syllabification. In cases when the algorithm produces spurious results, for example when dealing with an unusual sequence of segments, the user can modify the syllabification directly from the GUI, using a contextual menu. We currently provide syllabification algorithms for languages such as Catalan, Dutch, English, French, European Portuguese and Spanish, including a number of different algorithms for many of these languages. We are working directly with users to add support for additional research needs, for example an algorithm for Gurindji, to be released with version 1.4 of the application. While we are not considering this a pressing priority at this time, we are considering the possibility of giving the user access to the editing of syllabification algorithms in future versions.

2.6 Automatic phone alignment

In addition to syllabification, Phon also performs automatic alignment of IPA Target and Actual phones. This function relies on a ‘best-guess’ dynamic programming algorithm. The alignment is confined within word groups, considered to be independent domains of analysis but however transcends syllable boundaries within the groups. This enables, for example, the alignment of [s], syllabified in the coda of the first syllable in ‘pasta’, with the [s] syllabified in the onset of the reduced form ‘pasa’.

2.7 Search functions

Phon offers a powerful system to query transcribed data as well as other information contained in the transcript (e.g. participant name or age). In a nutshell, the user can make inventories of target and/or actual phones, syllable types and stress patterns as well as search individual data tiers, aligned phones and aligned word groups. Phon also incorporates support for the detection of consonant and vowel harmony as well as consonant metathesis. In each relevant case, the user can take advantage of the Phonex language to search for detailed phonological information (e.g. features, syllable positions), or build text-based queries using regular expressions. Up until version 1.3, we have favoured heavy GUI support for these searches. However, although GUI support makes many aspects of queries more user-friendly, it also imposes extremely limiting constraints on the types of queries that can be made, especially for queries that combine different types of information from different types of data. To alleviate this problem, Greg has designed a brand new system of queries based on Javascript. This script-based system is both extremely powerful and much more efficient than the search methods available in Phon 1.3. We are currently testing this function in collaboration with a number of Phon users, to whom we are deeply grateful. In order to make our script-based system as friendly as possible, we will incorporate built-in scripts of different kinds in our next release of Phon and maintain a discussion forum, through which we will support our community of users and gather feedback for additional refinement.

The data returned by Phon’s query functions can be visualized directly in the application (and further edited if needed), saved as independent transcript files, or exported as CSV (comma-separated value) text files or Excel spreadsheets. In CSV or Excel exported results, a summary of the search is produced alongside the detailed results. Once our script-based search is finalized, the user will be able to specify which information should be returned in the exported report. This will facilitate the management of search results as well as their transfer to other applications for further processing.

 

3. Future outlook

Looking into the future, we have also begun testing of an updated system to read and write files that is bringing us closer to full compatibility with the CHAT file format utilized by CLAN. Once we complete this testing phase, users will be able to migrate their data from CLAN to Phon using the CHAT2XML application programmed by Franklin Chen. This, alongside the other improvements described above, will be the last significant development included in the upcoming version 1.4 of Phon. After the release of that version, we will tackle the next major step, the design of a new module to support acoustic measurement data imported from Praat. This module will enable the researcher to store acoustic data alongside phonological information (e.g. features, syllabification, alignment) and thus perform research on the phonetics and phonology of various phenomena using an integrated data coding and query system.

Implicit to many aspects of the discussion above is the fact that much of our development efforts would have been much less fruitful without feedback and support from the community. Your response has been tremendous thus far. We look forward to working with you on the further advancement of this exciting project.

 

Yvan Rose
http://iascl.talkbank.org/bulletins/bulletinV28N1.html#phon

 

More information