I can answer you about service discovery, I didn't work with automatic (dynamic) composition of Web Services but I can refer you a good review of these techniques. 

With regard to the service discovery, it depends of the kind of Web Services you are working with, if they are RESTful or SOAP. In my case Web Services were SOAP. For SOAP web services you have to take into account the WSDL descriptor (Web Service Description Language) which specifies the inputs/outpus, and the operations of the Web Service. Therefore you have to analyze this file searching for compatible operations that can be linked in order to be "composed" in a Business Process or other structure of complex services. A common cited work about searching compatible services was published by Paoulucci (Automated discovery, interaction and composition of Semantic Web services) who expressed how to find services that best match a service request in terms of interfaces (inputs/outputs). However, this is an old paper, so I would recommend you to review a new state of the art in this field as for example:  the work of Rostami et al "Web Services Composition Methods and Techniques: A review";  Mello et al "A Review of Dynamic Web Service Composition Techniques" and the paper of Sheng et al "Web services composition: A decade’s overview". Below you can find links to download them. 

Hope it can be useful. 

http://www.sciencedirect.com/science/article/pii/S1570826803000039

http://link.springer.com/chapter/10.1007/978-3-642-17881-8_9

http://www.sciencedirect.com/science/article/pii/S0020025514005428

http://airccse.org/journal/ijcseit/papers/3613ijcseit03.pdf

Thank you very much for your sincere effort to answer the question. 

Dear Gopal,

You can study the documents below:

http://www.cs.cmu.edu/~anupriya/papers/jws03.pdf

http://www.hindawi.com/journals/am/2011/238683/

http://di.ufpe.br/~redis/intranet/bibliography/middleware/chen-composition06.pdf

https://msdn.microsoft.com/en-us/library/aa480728.aspx

Hope it helps you.

To add to the discussion, the primary issue behind discovering and composing services is how the services are described (if we're talking about input/output descriptions) or specified (if we're also dealing with other details, such as preconditions and postconditions of service execution). Let's say, for the sake of simplicity, that you're only working with syntactic interfaces, i.e. inputs and outputs of services, without any semantic information. This means that you're more than likely dealing with WSDL descriptions (version 2 can describe both SOAP and RESTful services). On the one hand, you have the request description, which details the required inputs and expected outputs of the service you want. On the other hand, you have your service repository, which contains candidate services, each one with its own WSDL interface description. The discovery process usually consists of two phases: matchmaking and selection. In the first phase, you find suitable candidates that match your request description and in the second phase, you rank your results and select the most suitable one (usually other factors come to play in this phase, like QoS characteristics). Since you're dealing with WSDL descriptions, matchmaking essentially involves comparing request input/output names and candidate input/output names. If no service description in your repository is a match, then you need to create a composition that does that. This is a wholly different problem and a vast research area. Since you're mentioning BPEL, which is one candidate solution, this would involve manually or automatically combining candidate services using one or more BPEL control constructs (sequence, flow, etc.), so that the resulting composition matches your request, meaning that it requires the same inputs and produces the same outputs.

Service research contains literally thousands of papers that offer discovery and composition techniques, those mentioned in the previous answers should be a good start, you can also check out the link below for a state of the art in service discovery; also, my PhD thesis, available in my profile, contains an extensive literature review on automated service composition.

http://www.igi-global.com/chapter/taxonomy-state-art-service-discovery/55243

US Patent  6,963,863 col 12:1-9 (filed Sept 1999) describes service composition and uses a web search engine to discover services.